# forck.live — 21–27 September 2026 > Live feed of AI lab newsrooms. First-party announcements, traced to the source. forck.live follows what companies announce about their own products. Press may corroborate an official story; it never leads and never appears as an item of its own. A missing date stays missing. A title is the publisher's words — where a company does not publish in English, an English rendering is listed separately as `Title in English` and named as this site's translation, never as theirs. This file is one issue of the weekly archive: every confirmed first-party announcement of ISO week 2026-W39, 21–27 September 2026, newest first. The window never moves, so this URL always names the same seven days — but a post published inside them and detected later is added when it arrives, so a closed week can still gain entries. Every entry is one first-party announcement, and carries the same eight fields in this order before any others: Company · Announced · Category · Coverage · Announcement · Group · Source · Record `Source` is the company's own page the announcement was traced to and `Record` is this site's page for it. Both are safe to cite. `Coverage` is a number of outlets, or `not counted` — which means this record reached no count of its own, never that nobody wrote about it. `Category` is this tracker's own analysis, and is authoritative wherever it is stated. `not stated` means either that nothing has analysed the post yet or that the analysis placed it outside this taxonomy; neither is a verdict about the announcement. `Announcement` is whether the post is news of something the company did, rather than a how-to, a customer story, a changelog listing or a staffing note. `Group` is which of this record's groups the entry falls in, carried on the entry itself so it holds where the groups are headings and where they are not. One of: `models`, `covered`, `announcements`, `routine`. Anything after `Record` is held detail that only some announcements carry. - This week's page: https://forck.live/week/2026-W39 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Sakana AI — SAIL: Scaling In-Context Imitation Learning - Company: Sakana AI (sakana.ai) - Announced: 2026-09-27T15:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/sail - Record: https://forck.live/items/14457-sail-scaling-in-context-imitation-learning - Subject: Sakana models Sakana AI and the University of Tokyo propose SAIL, a method for VLM-based robot trajectory generation that uses test-time scaling with Monte Carlo tree search to refine trajectories in simulation. In six simulated manipulation tasks, increasing the search budget from one to 45 candidates raised the success rate from 25% to 73%. The method was also evaluated on a physical robot. ### GitHub — Enterprise managed settings in-product validator - Company: GitHub (github.com) - Announced: 2026-09-25T23:24:57+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-25-enterprise-managed-settings-in-product-validator - Record: https://forck.live/items/13924-enterprise-managed-settings-in-product-validator - Subject: Copilot GitHub added an in-product validator for enterprise managed settings in GitHub Copilot that detects malformed JSON, unsupported configurations, invalid team mappings, and other errors preventing policy enforcement. The validator identifies affected files and JSON paths to help administrators correct configuration issues. ### GitHub — Usage metrics API adds pull request review stages - Company: GitHub (github.com) - Announced: 2026-09-25T21:09:40+00:00 - Category: api-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-25-usage-metrics-api-adds-pull-request-review-stages - Record: https://forck.live/items/13923-usage-metrics-api-adds-pull-request-review-stages - Subject: Copilot GitHub's Copilot usage metrics API now includes pull request review stage breakdowns, showing median and 90th percentile times for three review phases: ready for review to first review, first review to final review, and final review to merge. The new pull_request_review_times array is available in enterprise and organization repository-level reports, with data building forward from September 21, 2026. ### OpenAI — Proaction boosts sales 60% and saves 75+ hours with Codex - Company: OpenAI (openai.com) - Announced: 2026-09-25 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/proaction - Record: https://forck.live/items/13889-proaction-boosts-sales-60-and-saves-75-hours-with-codex - Subject: GPT / ChatGPT / API Proaction, a fleet management software company, uses Codex, GPT-Live-1, and GPT-6 Astra to accelerate sales and reduce engineering workload. The company's co-founder Colin Knudsen uses Codex to build customized interactive demos in 30–45 minutes each, saving 40–60 engineering hours monthly and increasing deals moving to solution development by 50–60%. Proaction also deploys voice agents built on GPT-Live-1 and GPT-6 Astra to handle fleet operations tasks such as maintenance coordination and toll management. ### Perplexity — Low, medium, and high presets use GPT-6 - Company: Perplexity (perplexity.ai) - Announced: 2026-09-25T17:56:50+00:00 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://docs.perplexity.ai/docs/resources/changelog#september-2026 - Record: https://forck.live/items/14332-low-medium-and-high-presets-use-gpt-6 - Subject: Sonar Perplexity's Agent API low, medium, and high presets now use GPT-6 model variants instead of GPT-5.6. The low and medium presets switched to openai/gpt-6-luna, while the high preset switched to openai/gpt-6-sol. Prompts, reasoning effort, tools, token budgets, and step limits remain unchanged. ### GitHub — Agentic autofix now uses Copilot Memory - Company: GitHub (github.com) - Announced: 2026-09-25T17:25:50+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-25-agentic-autofix-now-uses-copilot-memory - Record: https://forck.live/items/13906-agentic-autofix-now-uses-copilot-memory - Subject: Copilot GitHub's agentic autofix feature now integrates with Copilot Memory for customers who have enabled it. When resolving security alerts, agentic autofix reviews existing memories for context and stores fix patterns as memories for future use, which can help resolve additional alerts and inform other Copilot features like code review and cloud agent. ### GitHub — GitHub Copilot weekly releases — September 21 - Company: GitHub (github.com) - Announced: 2026-09-25T16:42:48+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-25-github-copilot-weekly-releases-september-21 - Record: https://forck.live/items/13897-github-copilot-weekly-releases-september-21 - Subject: Copilot GitHub Copilot added Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and Grok 4.7 to its model lineup, with availability varying by plan. Local sandboxing for agents entered public preview, and assisted approvals for agent sessions also reached public preview. Updates extended to Copilot in Slack, Microsoft Teams, JetBrains, and VS Code, including model switching mid-conversation, duplicate issue detection, and shared organization skills. ### GitHub — Updates to GitHub Copilot for Slack and Microsoft Teams - Company: GitHub (github.com) - Announced: 2026-09-25T16:33:42+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-25-updates-to-github-copilot-for-slack-and-microsoft-teams - Record: https://forck.live/items/13898-updates-to-github-copilot-for-slack-and-microsoft-teams - Subject: Copilot GitHub Copilot in Slack and Microsoft Teams now supports expanded context from conversations, including files, attachments, images, and message links. The update improves how Copilot creates and links GitHub work items back to source discussions, allows users to switch models per message, and enhances handling of longer-running tasks with better status reporting and connection recovery. The public preview is available to organizations on GitHub Copilot Business and GitHub Copilot Enterprise plans. ### Amazon — Scaling MoE reinforcement learning on Amazon EKS with EFA and DeepEP with 40% more throughput - Company: Amazon (amazon.com) - Announced: 2026-09-25T16:29:50+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/scaling-moe-reinforcement-learning-on-amazon-eks-with-efa-and-deepep-with-40-more-throughput/ - Record: https://forck.live/items/13891-scaling-moe-reinforcement-learning-on-amazon-eks-with-efa-and-deepep-with-40 - Subject: Bedrock / Nova When you post-train a Mixture-of-Experts (MoE) model with Reinforcement Learning from Human Feedback (RLHF) or Group Relative Policy Optimization (GRPO) at scale, three simultaneous challenges emerge. The first requires coordinating heterogeneous compute for rollout generation and policy training. Second, sustaining high-throughput communication across hundreds of accelerators. And third, dynamically orchestrating every subsystem to keep them in balance. On AWS, you can address these challenges using Amazon Elastic Kubernetes Service (Amazon EKS), Elastic Fabric Adapter (EFA), and DeepEP. Mixture-of-Experts (MoE) has become a standard architecture for scaling large language models (LLMs) to hundreds of billions or even trillions of parameters, while maintaining efficient inference through sparsity. However, sparsity doesn’t remove infrastructure complexity in training. As part of the standard training pipeline, these models must undergo pre-training, mid-training, supervised fine-tuning (SFT), and reinforcement learning (RL). … ### Amazon — Accelerate multimodal RL training with SkyRL on Amazon SageMaker HyperPod - Company: Amazon (amazon.com) - Announced: 2026-09-25T16:18:07+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/accelerate-multimodal-rl-training-with-skyrl-on-amazon-sagemaker-hyperpod/ - Record: https://forck.live/items/13892-accelerate-multimodal-rl-training-with-skyrl-on-amazon-sagemaker-hyperpod - Subject: Bedrock / Nova Amazon published a guide showing how to use SageMaker HyperPod with the open-source SkyRL framework to run GRPO post-training on a Qwen3-VL-8B vision-language model for visual maze navigation. The training improved the maze solve rate from 43.75% to over 95% on a fixed 64-maze evaluation set. The guide covers cluster setup, Ray integration, and monitoring via Amazon Managed Grafana. ### Amazon — NarrateAI: production-ready LLM quality assurance on Amazon Bedrock - Company: Amazon (amazon.com) - Announced: 2026-09-25T16:15:22+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/narrateai-production-ready-llm-quality-assurance-on-amazon-bedrock/ - Record: https://forck.live/items/13893-narrateai-production-ready-llm-quality-assurance-on-amazon-bedrock - Subject: Bedrock / Nova Amazon published a technical guide describing five quality assurance techniques for LLM applications on Amazon Bedrock: adaptive pipeline orchestration, cross-account multi-model failover, real-time streaming evaluation, composite evaluation framework, and data accuracy verification. The post reports that these techniques achieve approximately 99 percent numerical accuracy while streaming responses in real time. It is the second post in the NarrateAI series, which serves over 4,000 AWS executive leaders. ### Amazon — Deploying real-time personalized speech with Qwen3-TTS on Amazon SageMaker AI - Company: Amazon (amazon.com) - Announced: 2026-09-25T16:09:46+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/deploying-real-time-personalized-speech-with-qwen3-tts-on-amazon-sagemaker-ai/ - Record: https://forck.live/items/13894-deploying-real-time-personalized-speech-with-qwen3-tts-on-amazon-sagemaker-ai - Subject: Bedrock / Nova Amazon published a guide showing how to deploy the publicly available Qwen3-TTS-12Hz-1.7B-Base model from SageMaker JumpStart to a real-time inference endpoint for voice cloning. The model supports 10 languages and cross-lingual cloning, and can generate speech in a target speaker's voice from a short reference recording without retraining. The post covers deployment using the SageMaker Python SDK, endpoint invocation, configuration, and CloudWatch monitoring. ### Amazon — How Datacor built self-service rental analytics with Amazon Quick Sight - Company: Amazon (amazon.com) - Announced: 2026-09-25T15:54:42+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/how-datacor-built-self-service-rental-analytics-with-amazon-quick-sight/ - Record: https://forck.live/items/13878-how-datacor-built-self-service-rental-analytics-with-amazon-quick-sight - Subject: Bedrock / Nova Datacor built a self-service rental analytics solution for its TrackAbout customers by integrating Amazon Quick Sight, providing interactive dashboards and natural language querying. The solution uses an automated cross-cloud data pipeline to refresh datasets in Quick Sight SPICE, enabling business users to explore rental performance data without IT involvement. ### Amazon — Multi-Region training with Amazon SageMaker HyperPod and Qumulo - Company: Amazon (amazon.com) - Announced: 2026-09-25T15:49:44+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/multi-region-training-with-amazon-sagemaker-hyperpod-and-qumulo/ - Record: https://forck.live/items/13879-multi-region-training-with-amazon-sagemaker-hyperpod-and-qumulo - Subject: Bedrock / Nova Amazon published a solution architecture pairing SageMaker HyperPod with Qumulo's Cloud Data Fabric to enable cross-Region training without copying data. Validation showed a spoke cluster in us-west-2 reading data from a hub in us-east-2 matched the hub's throughput of 115–117 samples/sec after a warmup period, achieving 98–100% GPU utilization. The post describes the architecture, NeuralCache predictive caching, and validation results from a 1.02 billion-parameter LLaMA v3 training run. ### Perplexity — Upcoming retirement of older OpenAI models - Company: Perplexity (perplexity.ai) - Announced: 2026-09-25T14:34:21+00:00 - Category: deprecation - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://docs.perplexity.ai/docs/resources/changelog#september-2026 - Record: https://forck.live/items/13876-upcoming-retirement-of-older-openai-models - Subject: Sonar Perplexity will retire several older OpenAI model IDs (openai/gpt-5.4, openai/gpt-5.4-mini, openai/gpt-5.4-nano, openai/gpt-5.2, openai/gpt-5.1, openai/gpt-5, openai/gpt-5-mini) from its Agent API and Router API on October 24, 2026. Users must update direct model selections and fallback chains before the cutoff; after that date the retired IDs will no longer be accepted or returned by model-list endpoints. ### Perplexity — Fast preset uses Fast Search - Company: Perplexity (perplexity.ai) - Announced: 2026-09-25T14:18:34+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://docs.perplexity.ai/docs/resources/changelog#september-2026 - Record: https://forck.live/items/13877-fast-preset-uses-fast-search - Subject: Sonar Perplexity's Agent API fast preset now uses Fast Search, reducing web_search price from $2.50 to $1.00 per 1,000 invocations and making search about 800 ms faster. ### GitHub — Default Enablement of Copilot Features for Copilot Business and Enterprise - Company: GitHub (github.com) - Announced: 2026-09-25T03:13:23+00:00 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-24-default-enablement-of-copilot-features-for-copilot-business-and-enterprise - Record: https://forck.live/items/13826-default-enablement-of-copilot-features-for-copilot-business-and-enterprise - Subject: Copilot GitHub is introducing a new global default policy for generally available Copilot features and supported client capabilities in enterprise and organization settings. The policy, configurable now and effective October 22, lets administrators choose to enable, disable, or let organizations decide on eligible features, while preserving explicit decisions and keeping preview features opt-in. ### Perplexity — AI Alignment: Current Research and Debate - Company: Perplexity (perplexity.ai) - Announced: 2026-09-25 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.perplexity.ai/hub/blog/ai-alignment - Record: https://forck.live/items/13920-ai-alignment-current-research-and-debate - Subject: Perplexity AI alignment used to simply mean “does the AI model follow user instructions?” However, as AI systems get more capable and autonomous, that question has broadened to include judgment and values. This article will explore how AI platforms, government bodies, and non-profits are responding to the challenge of AI alignment. What AI alignment means At its simplest, AI alignment means making sure an AI system does what it is meant to do. An aligned AI model reads a prompt and produces the output the prompt asked for, without gaming the objective it was trained on. In modern alignment research, this is usually framed as intent alignment: building systems that try to do what their operators intend, not just literally follow the text of a prompt. Alignment is both a technical challenge and a topic of public debate. As a technical term, alignment is an engineering problem. Does the specific model, working on a specific task, do what it was built to do? It’s a bounded, testable issue. However, AI alignment is also used in broader public and policy discussion on the general safety of AI systems. … ### Anthropic — Build plugins for Claude - Company: Anthropic (anthropic.com) - Announced: 2026-09-25 - Category: developer-tool-release - Coverage: not counted - Announcement: no - Group: routine - Source: https://claude.com/blog/build-plugins-for-claude - Record: https://forck.live/items/13905-build-plugins-for-claude - Subject: Claude Anthropic opened a new developer portal allowing developers on paid Claude plans to submit plugins to the Claude directory. Plugins package MCP connectors, Agent Skills, or both, and can be submitted as single MCP connectors or plugin bundles combining MCP servers and skills hosted on GitHub. Once approved, plugins are listed in the Claude directory, and developers gain access to usage analytics tracking installs by product surface and version, as well as discovery metrics. ### Amp — Less Noise - Company: Amp (ampcode.com) - Announced: 2026-09-25 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/news/less-noise - Record: https://forck.live/items/13880-less-noise - Subject: Amp Amp has updated its interface to collapse the step-by-step work of agents, allowing users to see only the final outcome while still having the option to expand steps when needed. The change reflects the growing use of multiple agents that run longer and more autonomously. ### Cognition — Cognition Crosses $1B in Annualized Revenue Run Rate - Company: Cognition (cognition.com) - Announced: 2026-09-25 - Category: funding-company-news - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cognition.com/blog/1b-run-rate - Record: https://forck.live/items/13855-cognition-crosses-1b-in-annualized-revenue-run-rate - Subject: Devin / SWE Cognition, the company behind the AI coding assistant Devin, announced it has reached a $1 billion annualized revenue run rate less than two years after Devin became generally available. Devin is used by engineering teams at companies including GE Aerospace, Rivian, Rohlik, and Exa. ### Meta — The Biggest News From Connect 2026 - Company: Meta (meta.com) - Announced: 2026-09-24T21:15:55+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://about.fb.com/news/2026/09/the-biggest-news-from-connect-2026/ - Record: https://forck.live/items/13782-the-biggest-news-from-connect-2026 - Subject: Meta AI We’re building personal AI agents for everyone and a whole family of devices that let you connect with them from anywhere. Bringing Muse to AI Glasses We launched Muse , a first-of-a-kind personal AI agent that proactively helps you meet your goals, earlier this month. At Connect, we announced we’re bringing Muse to our AI glasses in the coming months, so you’ll soon have easy access to your personal agent wherever you go. Mark Zuckerberg showed how Muse keeps getting better with new features, connectors, and a state-of-the-art model that brings your Muse to life. Muse on AI glasses : You’ll soon be able to say your agent’s name, and Muse can act on what you’re looking at, so you won’t have to describe what’s in front of you. Ask about a product on a shelf, a flier on a wall, or a long list of school supplies, and Muse can take action. Voice mode : Talking to an agent as capable as Muse is different from voice chat with an assistant. You can have a long, in-depth conversation, and Muse gets work done in the background while you’re still talking. https://about.fb.com/wp-content/uploads/2026/09/02_Bringing-Muse-to-AI-Glasses.mp4 … ### Google Research — Automating coherent long-form video generation - Company: Google Research (research.google) - Announced: 2026-09-24T19:40:29+00:00 - Category: research-paper - Coverage: not counted - Announcement: no - Group: routine - Source: https://research.google/blog/coherent-long-form-video-generation/ - Record: https://forck.live/items/13758-automating-coherent-long-form-video-generation - Subject: Research Google Research introduced a unified multi-agent framework that autonomously generates temporally consistent, long-form video narratives, addressing identity drift and cascading failures in current linear AI pipelines. The framework, built as an orchestration layer on top of Gemini and Veo, includes components such as Co-Director, CANVAS, A²RD, and VQQA, and uses hierarchical parameterization with a multi-armed bandit algorithm to optimize creative directions. Evaluations showed substantial gains in multi-shot narrative consistency and character persistence, enabling minutes-long videos while mitigating visual drift and pipeline error propagation. ### Microsoft — Ship agents faster with expanded model choice, voice agents, and continuous optimization - Company: Microsoft (microsoft.com) - Announced: 2026-09-24T18:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://azure.microsoft.com/en-us/blog/ship-agents-faster-with-expanded-model-choice-voice-agents-and-continuous-optimization/ - Record: https://forck.live/items/14954-ship-agents-faster-with-expanded-model-choice-voice-agents-and-continuous - Subject: Azure AI The best model for your business will keep changing. Adopting it should move your business forward, not send your team back to rebuild the architecture around it. As models advance, organizations need the freedom to choose the right model for each workload, and an agent foundation that can evolve with those choices. Microsoft Foundry provides that model—and harness—agnostic foundation for building and running agents. Teams can adopt better models as they emerge while preserving their investments in the enterprise systems, knowledge, tools, and controls that make agents useful to their business. Adopting a newer model is only the starting point. Foundry helps teams continuously improve their agents, using production traces to evaluate and refine instructions, skills, tools, and model choice against quality, latency, and cost. That’s the hill-climbing approach: observe, evaluate, optimize, validate, and repeat—with people in control Learn more about Microsoft Foundry Today, we’re advancing that foundation to help teams build and continuously improve agents that deliver business value: Choose from a broader range of models as frontier AI advances. … ### Perplexity — Fast Search - Company: Perplexity (perplexity.ai) - Announced: 2026-09-24T17:09:31+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://docs.perplexity.ai/docs/resources/changelog#september-2026 - Record: https://forck.live/items/13737-fast-search - Subject: Sonar Perplexity introduced a lower-latency search mode called Fast Search for its Search API and Agent API web_search tool, priced at $1.00 per 1,000 requests or invocations, with model tokens billed separately for Agent API. ### Google DeepMind — Introducing Gemini 3.8 Live with Live Avatar - Company: Google DeepMind (deepmind.google) - Announced: 2026-09-24T16:20:39+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://deepmind.google/blog/introducing-gemini-38-live-with-live-avatar/ - Record: https://forck.live/items/13723-introducing-gemini-3-8-live-with-live-avatar - Subject: Gemini Google DeepMind introduced Gemini 3.8 Live with Live Avatar, a feature that adds near real-time video generation and a visual persona to the existing Gemini 3.8 Live dialogue model. The feature supports lip-syncing, expressions, turn-taking, asynchronous tool execution, and native multilingual speech-to-speech synchronization across 97 languages. It is available starting today in Gemini Enterprise, with custom avatar creation available through enterprise allowlisting. ### Amazon — Speaker-labeled transcription with WhisperX on SageMaker AI - Company: Amazon (amazon.com) - Announced: 2026-09-24T16:20:12+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/speaker-labeled-transcription-with-whisperx-on-sagemaker-ai/ - Record: https://forck.live/items/13732-speaker-labeled-transcription-with-whisperx-on-sagemaker-ai - Subject: Bedrock / Nova Amazon released a WhisperX Deep Learning Container (DLC) for SageMaker AI that packages Whisper, alignment models, and diarization weights into a GPU-ready image. The container can be deployed to real-time or asynchronous endpoints without building a custom image, supporting per-word timestamps and speaker labels for use cases like contact centers, meetings, and media captioning. ### Amazon — Build a multi-account AI agent with AgentCore Gateway and MCP - Company: Amazon (amazon.com) - Announced: 2026-09-24T16:12:47+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/build-a-multi-account-ai-agent-with-agentcore-gateway-and-mcp/ - Record: https://forck.live/items/13719-build-a-multi-account-ai-agent-with-agentcore-gateway-and-mcp - Subject: Bedrock / Nova Amazon published a guide for building a multi-account AI agent architecture using Amazon Bedrock AgentCore Gateway and Model Context Protocol (MCP). The design keeps each team's data in its own AWS account while providing a unified query endpoint through a central platform account. The post walks through setting up cross-account MCP integration, authentication with AgentCore Identity and Okta, and fine-grained authorization with Policy in Amazon Bedrock AgentCore. ### Amazon — Aderant builds intelligent ticket triage with Amazon Nova - Company: Amazon (amazon.com) - Announced: 2026-09-24T16:06:46+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/aderant-builds-intelligent-ticket-triage-with-amazon-nova/ - Record: https://forck.live/items/13720-aderant-builds-intelligent-ticket-triage-with-amazon-nova - Subject: Bedrock / Nova This guest post is co-written by Angela Mapes and Adam Walker of Aderant. In this post, we share how Aderant , a global provider of business management software for the legal industry, built an intelligent ticket triage system using Amazon Nova Lite through Amazon Bedrock. Aderant’s solution automates much of the context gathering, classification, routing, and knowledge enrichment required for support-ticket triage. The Intelligent Ticket Analyzer supports Aderant’s 38-person SierraOps team, which operates Expert Sierra across 268 client environments globally. It reviews newly submitted, unassigned tickets during scheduled hourly processing cycles on business days. It gathers operational context from Jira, Confluence, Amazon Athena , and Microsoft SharePoint to recommend a team assignment and resolution starting point, and it automates approved routing and communication actions. During the first 2.5 weeks of production, from June 30 through July 17, 2026, the analyzer reviewed 109 tickets and achieved approximately 96% routing accuracy. … ### Hugging Face — Accelerating vision-language models with LFM2.5-VL-DSpark - Company: Hugging Face (huggingface.co) - Announced: 2026-09-24T14:08:57+00:00 - Category: model-update - Coverage: 1 outlet - Announcement: yes - Group: models - Source: https://huggingface.co/blog/LiquidAI/lfm2-5-vl-dspark - Record: https://forck.live/items/13710-accelerating-vision-language-models-with-lfm2-5-vl-dspark - Subject: Platform Liquid AI released an experimental DSpark draft model for its vision-language model LFM2.5-VL-3B, adding a speculative decoding path that increases memory footprint by 8.9% (280M parameters) while achieving decode speedups up to 3.13x on device and 2.66x on an H100, with end-to-end gains up to 2.62x and 2.27x. The drafter uses a simplified attention-only architecture with 4 layers and a block size of 9, and ships with day-one support for llama.cpp, MLX-VLM, and SGLang. The model is open-weight and available in Safetensors and GGUF formats. ### NVIDIA — How Open Science Can Help Researchers Prepare for the Next Pandemic - Company: NVIDIA (nvidia.com) - Announced: 2026-09-24T14:00:50+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://blogs.nvidia.com/blog/open-protein-dataset/ - Record: https://forck.live/items/13704-how-open-science-can-help-researchers-prepare-for-the-next-pandemic - Subject: AI platform When COVID-19 emerged, scientists had a crucial advantage: Decades of prior research on coronaviruses meant they understood the virus’ key proteins well enough to design vaccines in record time. The next pandemic may not offer the same head start. To help improve the odds, NVIDIA has joined a coalition of global research organizations, including Google DeepMind and the European Molecular Biology Laboratory’s European Bioinformatics Institute (EMBL-EBI), to release predicted 3D structures for the protein complexes of more than 2,800 viruses — openly available to any scientist, anywhere, through the AlphaFold Database. The structures in the newly released dataset were inferred using AlphaFold2 — Google DeepMind’s AI model for predicting how proteins fold into 3D shapes — with optimization from NVIDIA BioNeMo Inference Runtime . This allowed the team to scale inference to thousands of viral proteomes, predicting the complexes, or groups of interacting proteins, encoded within each virus. “Our ambition with the AlphaFold Database has always been to democratize access to foundational biology at scale,” said Risha Patel, life sciences partnerships manager at Google DeepMind. … ### OpenAI — Ringg’s AI agents resolve up to 65% of customer calls with OpenAI - Company: OpenAI (openai.com) - Announced: 2026-09-24 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/ringg - Record: https://forck.live/items/13442-ringg-s-ai-agents-resolve-up-to-65-of-customer-calls-with-openai - Subject: GPT / ChatGPT / API Ringg, a voice and chat agent platform, uses OpenAI's GPT-5.6 models to power multilingual agents handling over 7 million connected calls per month. Migrating real-time workloads from GPT-4.1 to GPT-5.6 reduced model costs by approximately 90%. The platform routes tasks to different GPT-5.6 variants based on performance needs, achieving up to 65% call resolution and an average customer satisfaction score of 4.8. ### Tencent — Tencent’s Quirky, Accidental Mascot Turns 10 - Company: Tencent (tencent.com) - Announced: 2026-09-24T10:45:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.tencent.com/tencents-quirky-accidental-mascot-turns-10/ - Record: https://forck.live/items/14412-tencent-s-quirky-accidental-mascot-turns-10 - Subject: Hunyuan Most mascots begin with a brief: a document setting out who the character is, what they stand for, which adjectives apply. But when marketing teams start running low on ideas, some odd characters can result — with Long Penguin the most notorious. Long Penguin never got a brief. What he got instead was ten years of moments nobody planned, and a team who found him funny enough to keep going. Here is the whole thing, head to feet. 2016 – Starting with a doodle A team was stuck when trying to come up with an idea to mark the Mid-Autumn Festival, an annual holiday in China that brings families together. In middle of a meeting, someone drew Tencent’s penguin mascot with a ridiculously long bird — the kind of thing you draw when you’ve run out of good ideas and start entertaining bad ones. In Chinese, the team jokingly called it 长鹅 (cháng é, literally “long penguin”). When said aloud in Chinese, it sounds like the moon goddess at the heart of the Mid-Autumn legend. No one knew that would become the beginning of Chang’e’s long history. 2017-2019 – Developing a cult following For four years he lived in group chats. … ### Lovable — Inside Chats: How Lovable's Agents Work Together - Company: Lovable (lovable.dev) - Announced: 2026-09-24T08:40:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://lovable.dev/blog/how-lovable-agents-work-together - Record: https://forck.live/items/13646-inside-chats-how-lovable-s-agents-work-together - Subject: Lovable / AI app builder Lovable published a technical guide explaining the engineering architecture behind Chats, its feature for connecting conversations to app-building work. The post describes how Lovable's Trajectory System—an append-only, forkable event log modeled after Git—separates what happened from what agents need to know, enabling agents to delegate tasks, maintain context across conversations, and coordinate work through durable inboxes and activations. ### Amp — The Mac App Is Your Runner - Company: Amp (ampcode.com) - Announced: 2026-09-24 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/news/the-mac-app-is-your-runner - Record: https://forck.live/items/13783-the-mac-app-is-your-runner - Subject: Amp Amp's macOS app now automatically starts a runner, eliminating the need to run amp --no-tui in a terminal. Users can add folders or projects in app settings, and the Mac appears as 'This Mac' in the thread picker. A 'Keep This Mac Awake' option prevents sleep while plugged in, with the screen still turning off and locking. ### Anthropic — Claude Tag now supports personal connectors in channels - Company: Anthropic (anthropic.com) - Announced: 2026-09-24 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://claude.com/blog/claude-tag-now-supports-personal-connectors-in-channels - Record: https://forck.live/items/13780-claude-tag-now-supports-personal-connectors-in-channels - Subject: Claude Anthropic announced that Claude Tag now supports personal connectors in Slack channels, allowing users to access their own connected tools and data when making requests. Users can control how responses are surfaced, with options to review before posting, use auto mode with content screening, or have Enterprise admins require review. The feature is rolling out to Team plans with Enterprise to follow. ### Apple — A Practical Recipe for Semi-Supervised Federated ASR: Online Pseudo-Labels with Server Update Stabilization - Company: Apple (apple.com) - Announced: 2026-09-24 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://machinelearning.apple.com/research/practical-recipe-federated-asr - Record: https://forck.live/items/13751-a-practical-recipe-for-semi-supervised-federated-asr-online-pseudo-labels-with - Subject: Machine Learning Research Apple researchers present a method for semi-supervised federated learning in automatic speech recognition that uses per-client online pseudo-label generation stabilized by server-side updates on labeled data. The approach improves over prior methods on 9 of 11 evaluation pairs, narrowing the gap to fully-supervised federated learning. ### Perplexity — Photon: Building a Retrieval and Ranking Engine From Scratch - Company: Perplexity (perplexity.ai) - Announced: 2026-09-24 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.perplexity.ai/hub/blog/photon - Record: https://forck.live/items/13742-photon-building-a-retrieval-and-ranking-engine-from-scratch - Subject: Perplexity Perplexity built Photon, an in-house retrieval and ranking engine designed to replace an open-source platform that faced scaling costs, tail latency, and recovery time constraints. Photon powers Perplexity's search pipeline and a new fast preset for its Search API, achieving single-search-call latency of 160 ms at p50 and 230 ms at p95, and reducing estimated model-plus-search cost per task by 68% across six public benchmarks while maintaining comparable task quality. ### Perplexity — Portable Computer comes to AMD-Powered Agentic PCs - Company: Perplexity (perplexity.ai) - Announced: 2026-09-24 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.perplexity.ai/hub/blog/portable-computer-comes-to-amd-powered-agentic-pcs - Record: https://forck.live/items/13738-portable-computer-comes-to-amd-powered-agentic-pcs - Subject: Perplexity People want to put their own compute to work on more of their everyday tasks, from analyzing local files to running recurring workflows. Completing that work on their own hardware gives them more control over sensitive information and credit spend. To let more people work this way, local AI needs to run on more devices and connect to the tools they already use. Today we’re expanding Portable Computer to systems powered by AMD Ryzen AI Max Series processors including the Ryzen AI Halo developer platform, giving more people a way to run local AI on their own hardware. Local inference doesn’t consume Computer credits. Users can run routine analysis and file processing on their Ryzen AI Max-powered system. The local model can still escalate to the cloud for more advanced research and reasoning when needed. The local agent stack comes to AMD Ryzen AI Max as one system AMD support gives users another way to run Perplexity Computer locally. They can put their own hardware to work on demanding tasks while limiting credit spend. … ### Anthropic — Coding sessions are longer and use more context. Claude Opus 5.5 is built with that in mind. - Company: Anthropic (anthropic.com) - Announced: 2026-09-24 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://claude.com/blog/claude-opus-5-5-built-for-coding-sessions-that-use-more-context - Record: https://forck.live/items/13726-coding-sessions-are-longer-and-use-more-context-claude-opus-5-5-is-built-with - Subject: Claude Anthropic released Claude Opus 5.5, a model update optimized for long-running coding sessions with higher context usage. The model costs approximately 40% less to run than Opus 5 for typical token-billed workloads, with significant savings on cached token reads. Opus 5.5 generates output more than 30% faster than Opus 5 and can complete open-ended tasks in fewer turns. ### Apple — Compressing Streaming Neural Audio Encoders via Latent-Space Distillation - Company: Apple (apple.com) - Announced: 2026-09-24 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://machinelearning.apple.com/research/latent-space-distillation - Record: https://forck.live/items/13718-compressing-streaming-neural-audio-encoders-via-latent-space-distillation - Subject: Machine Learning Research Apple researchers present a method for compressing streaming neural audio encoders used in on-device dictation through latent-space distillation. The technique trains a student encoder to regress the teacher's pre-quantizer latent representation, achieving 2.8× compression while maintaining within 1.9% relative word error rate of the teacher on most model pairs without fine-tuning. ### Amp — Shared Runners - Company: Amp (ampcode.com) - Announced: 2026-09-24 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/news/shared-runners - Record: https://forck.live/items/13700-shared-runners - Subject: Amp Amp added the ability to share runners with workspace members, allowing teams to spawn agents on shared machines such as GPUs, Macs, or dev boxes in specific networks. Shared runners receive workspace and project secrets and environment variables but not personal ones, and workspace admins can disable runner sharing in Member Settings. ### Meta — Bringing Private Processing to Meta AI Glasses - Company: Meta (meta.com) - Announced: 2026-09-24 - Category: capability-change - Coverage: 5 outlets - Announcement: yes - Group: covered - Source: https://engineering.fb.com/2026/09/23/security/private-processing-meta-ai-glasses/ - Record: https://forck.live/items/13495-bringing-private-processing-to-meta-ai-glasses - Subject: Llama / infrastructure Meta is extending its Private Processing confidential computing infrastructure to Meta AI Glasses, enabling the glasses to offload intensive AI workloads to cloud data centers while keeping user data encrypted and inaccessible to Meta. The system uses trusted execution environments (TEEs) in CPUs and GPUs to process personal context at scale while maintaining data confidentiality, integrity, and code integrity. ### Runway — Runway is Now in DaVinci Resolve - Company: Runway (runway.com) - Announced: 2026-09-23 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/news/company-news/runway-davinci-resolve - Record: https://forck.live/items/13450-runway-is-now-in-davinci-resolve - Subject: Gen / Aleph Runway released a plugin for DaVinci Resolve Studio that allows users to generate and restyle video and images directly within the editing application. The plugin integrates Runway's models into the editor's workflow, enabling users to prompt generation, preview results, and import them directly to the timeline without leaving the application. ### GitHub — More ways to request and configure Copilot code reviews - Company: GitHub (github.com) - Announced: 2026-09-23T21:25:58+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-23-copilot-code-review-more-ways-to-request-and-configure-reviews - Record: https://forck.live/items/13476-more-ways-to-request-and-configure-copilot-code-reviews - Subject: Copilot GitHub Copilot code review now offers a dedicated personal settings page for automatic review configuration available across all Copilot plans, including Copilot Business and Copilot Enterprise. Enterprise administrators can set a default review effort level for the entire enterprise, which applies to organization-owned repositories. ### Hugging Face — How to Use NVIDIA Warp and MjWarp to Accelerate Robotics Simulation and Learning Workflows - Company: Hugging Face (huggingface.co) - Announced: 2026-09-23T18:41:40+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/nvidia/how-to-use-nvidia-warp-and-mjwarp - Record: https://forck.live/items/13455-how-to-use-nvidia-warp-and-mjwarp-to-accelerate-robotics-simulation-and - Subject: Platform Hugging Face and NVIDIA published a guide on using NVIDIA Warp and MuJoCo Warp (MJWarp) to accelerate robotics simulation on GPUs. The article demonstrates how to scale robot simulation from single CPU-based environments to up to 2,048 parallel GPU-accelerated environments, enabling faster batch processing for reinforcement learning and large-scale sampling workflows. ### Amazon — From portal-hopping to instant answers: HEMA’s journey with MCP and Amazon Bedrock - Company: Amazon (amazon.com) - Announced: 2026-09-23T18:41:09+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/from-portal-hopping-to-instant-answers-hemas-journey-with-mcp-and-amazon-bedrock/ - Record: https://forck.live/items/13459-from-portal-hopping-to-instant-answers-hema-s-journey-with-mcp-and-amazon - Subject: Bedrock / Nova HEMA, a Dutch retailer, built HAL, an internal AI assistant using Model Context Protocol (MCP) and Amazon Bedrock AgentCore, to consolidate fragmented knowledge across disconnected portals and wikis. The solution allows engineers and other roles to access procedural and infrastructure knowledge from tools they already use, such as IDEs and chat applications, reducing onboarding friction and context-switching. ### Amazon — Agentic conversational video intelligence built on AWS - Company: Amazon (amazon.com) - Announced: 2026-09-23T18:21:54+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/agentic-conversational-video-intelligence-built-on-aws/ - Record: https://forck.live/items/13453-agentic-conversational-video-intelligence-built-on-aws - Subject: Bedrock / Nova With video intelligence powered by agentic AI, you can ask natural language questions about uploaded videos and get answers within seconds. Organizations across media, security, insurance, and professional services are generating more video than their teams can review. Meeting recordings accumulate in shared drives, and security cameras capture weeks of unreviewed footage. Field inspection videos sit in object storage long after the initial review. The information inside these videos is often valuable: a design decision discussed three weeks ago, the exact moment a person arrived at a door, or the sequence of events leading to a vehicle collision. But accessing it has traditionally required watching hours of content manually. The alternative, building custom machine learning (ML) pipelines for each specific question type, demands significant development effort. Each new use case meant new development work A transcription pipeline for meeting queries. A computer vision pipeline for visual search. A face-matching integration. … ### Amazon — Use open weight models as your AI coding agent with Amazon Bedrock - Company: Amazon (amazon.com) - Announced: 2026-09-23T18:17:44+00:00 - Category: capability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/use-open-weight-models-as-your-ai-coding-agent-with-amazon-bedrock/ - Record: https://forck.live/items/13454-use-open-weight-models-as-your-ai-coding-agent-with-amazon-bedrock - Subject: Bedrock / Nova Amazon Bedrock now enables developers to run AI coding agents using open weight models privately within their AWS account. The post demonstrates how to configure OpenCode, an open source terminal-native coding agent, with open weight models on Bedrock to support multi-model workflows for coding tasks, and describes how Ethara.AI deploys this architecture in production. ### Black Forest Labs — Introducing FLUX 3 Action - Company: Black Forest Labs (bfl.ai) - Announced: 2026-09-23T18:11:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://bfl.ai/blog/flux-3-action - Record: https://forck.live/items/13747-introducing-flux-3-action - Subject: FLUX Images, video and audio represent different aspects of the underlying reality. Training across these modalities lets us build on a much broader source of data than action demonstrations alone, resulting in more generalization. We then adapt this foundation through joint video-action training and finetuning for a target embodiment and its corresponding action space. In robotics, the fine-tuning recipe is just as important as the weights. We are publishing this report along with the weights to make this process transparent; from the pretraining and midtraining phases to finetuning and inference optimizations for action prediction - where efficiency in particular is a critical and necessary capability for local deployments. In addition, we analyze hybrid systems that combine fast action prediction with the planning capabilities of frontier reasoning models - and find that fast control makes embodied reasoning more cost- and time efficient - up to 53.64% more success per dollar. The focus of this report is action prediction applied to robotics - but action prediction extends to digital environments as well. … ### Microsoft Research — Offloaded inference for real-world physical AI robotics - Company: Microsoft Research (microsoft.com) - Announced: 2026-09-23T16:01:36+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.microsoft.com/en-us/research/blog/offloaded-inference-for-real-world-physical-ai-robotics/ - Record: https://forck.live/items/13440-offloaded-inference-for-real-world-physical-ai-robotics - Subject: Research / Phi Microsoft Research presents a systematic study of robotics inference workloads showing that offloading AI inference from onboard GPUs to edge or cloud infrastructure improves task success rates, enables larger models, and extends battery life by up to 160% compared to onboard GPU compute. The research introduces a Kubernetes-based toolset for containerizing and orchestrating distributed robotics AI workloads across robots, edge infrastructure, and the cloud. ### Google DeepMind — Advancing Private AI Compute with secure, server-side memory - Company: Google DeepMind (deepmind.google) - Announced: 2026-09-23T16:00:57+00:00 - Category: capability-change - Coverage: 1 outlet - Announcement: no - Group: covered - Source: https://deepmind.google/blog/advancing-private-ai-compute-with-secure-server-side-memory/ - Record: https://forck.live/items/13439-advancing-private-ai-compute-with-secure-server-side-memory - Subject: Gemini Google DeepMind announced an update to its Private AI Compute platform that adds persistent, server-side memory while maintaining on-device privacy standards. The architecture uses hardware-enforced secure enclaves, double-encrypted channels, and device-derived encryption keys to allow AI models to retain context across devices without exposing user data to Google or others. ### Microsoft — Your architecture diagram is not your resilience - Company: Microsoft (microsoft.com) - Announced: 2026-09-23T16:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://azure.microsoft.com/en-us/blog/your-architecture-diagram-is-not-your-resilience/ - Record: https://forck.live/items/13752-your-architecture-diagram-is-not-your-resilience - Subject: Azure AI Microsoft publishes the first article in a resilience series discussing how resilience validation is changing in the AI era. The post, drawing on a conversation with Mark Russinovich, explains that traditional architecture diagrams no longer capture operational reality, particularly as AI model dependencies become critical to workload availability. Microsoft argues that resilience must shift from a one-time project to a continuously validated property, and that roughly 70 percent of cloud outages are related to change rather than dramatic failures. ### OpenAI — Two years of OpenAI Academy - Company: OpenAI (openai.com) - Announced: 2026-09-23 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/two-years-of-openai-academy - Record: https://forck.live/items/13426-two-years-of-openai-academy - Subject: GPT / ChatGPT / API OpenAI marks two years of its Academy program, which provides AI skills training to communities. ### Google DeepMind — Gemini 3.8 text-to-speech says hello - Company: Google DeepMind (deepmind.google) - Announced: 2026-09-23T15:25:14+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://deepmind.google/blog/say-hello-to-gemini-38-text-to-speech/ - Record: https://forck.live/items/13433-gemini-3-8-text-to-speech-says-hello - Subject: Gemini Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS are our most expressive audio generation models yet. Generate custom character voices and direct scene dialogue across Google AI Studio, Gemini API, Gemini Enterprise, Gemini Notebook, and Google Vids. Leland Rechis Group Product Manager Alan Cowen Director, Research Science, on Behalf of the Gemini Audio Team Today, we’re introducing two new text-to-speech models to the Gemini family, transforming voice generation from static presets into a dynamic creative studio. These models enable creators, developers, and enterprises to create richer, more expressive audio experiences, while enabling improved user experiences in products like Gemini Notebook and Google Vids . Gemini 3.8 Flash TTS: Built for deep creative direction and character design. Create entirely new voices from scratch using natural language prompts to bring characters to life across gaming, immersive audiobooks, podcasts, and interactive media. Direct every performance line by line with granular control over acting cues, pacing, dialect shifts, and backchanneling. Gemini 3.8 Flash-Lite TTS: Built for high-volume, cost-efficient scale. … ### GitHub — Local sandboxing in the GitHub Copilot app - Company: GitHub (github.com) - Announced: 2026-09-23T15:00:57+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-23-local-sandboxing-in-the-github-copilot-app - Record: https://forck.live/items/13432-local-sandboxing-in-the-github-copilot-app - Subject: Copilot GitHub Copilot app now offers local sandboxing in public preview, allowing users to configure per-project restrictions on filesystem access, network resources, and credentials to limit the impact of unintended commands. ### NVIDIA — Sakeena Fiza Helps NVIDIA Hardware Succeed at Scale - Company: NVIDIA (nvidia.com) - Announced: 2026-09-23T15:00:32+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://blogs.nvidia.com/blog/nvidia-life-sakeena-fiza/ - Record: https://forck.live/items/13431-sakeena-fiza-helps-nvidia-hardware-succeed-at-scale - Subject: AI platform NVIDIA published a profile of validation engineer Sakeena Fiza, describing her role in testing and debugging hardware systems before mass production and customer deployment. The article documents her work investigating failures across firmware, hardware, software, mechanical design, and thermal behavior in data center systems, including the NVIDIA Rubin GPU. ### Microsoft — Designing agent-first platforms: What changes when agents do the work - Company: Microsoft (microsoft.com) - Announced: 2026-09-23T15:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://azure.microsoft.com/en-us/blog/designing-agent-first-platforms-what-changes-when-agents-do-the-work/ - Record: https://forck.live/items/13753-designing-agent-first-platforms-what-changes-when-agents-do-the-work - Subject: Azure AI For decades, applications have been designed to wait. A user clicks, a request arrives, code runs, a response goes back. And we got very good at this. We learned to forecast traffic, scale on demand, wrap everything in enterprise controls, and run it all with the operational discipline that keeps critical business systems available around the clock. That model is being turned on its head, and being asked to serve apps that continuously act, and wait for no one. What changed is not the infrastructure underneath, but the software being written on top of it. The work we used to capture as deterministic code, where every branch was defined in advance and every step was known before the first line ran, is now being rewritten as multi-agent applications that work out the steps at runtime. A developer used to encode the path, and now a developer describes the outcome and lets a set of agents reason their way toward it. Agents work differently. Given an outcome, an agent reasons through the problem, breaks it into steps, writes code to solve it, runs that code, looks at the result, and goes again. The work happens in a loop that no human is standing inside. … ### Sakana AI — The Next Frontier: Welcoming AI Pioneer Jürgen Schmidhuber to Sakana AI - Company: Sakana AI (sakana.ai) - Announced: 2026-09-23T15:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/schmidhuber - Record: https://forck.live/items/13717-the-next-frontier-welcoming-ai-pioneer-j-rgen-schmidhuber-to-sakana-ai - Subject: Sakana models The next era of artificial intelligence will move beyond digital interfaces and directly into the physical world. To lead this paradigm shift, Sakana AI is incredibly proud to announce that Jürgen Schmidhuber , universally recognized as the father of modern AI, is officially joining Sakana AI as Chief Scientific Advisor, alongside his current positions, and will be involved in our Recursive Self-Improvement (RSI) Lab . “Japan is a birthplace of foundational neural network architectures and advanced robotics. It is a privilege to join Sakana AI to help bridge these two worlds. The future of intelligence is not just language; it is physical AI powered by World Models. Sakana AI is uniquely positioned to reclaim Japan’s legacy of innovation by building autonomous systems that can truly understand, simulate, and interact with the physical universe.” — Jürgen Schmidhuber A core strength of Sakana AI is our ability to reverse the global AI brain drain and attract top-tier international talent to Tokyo. We are actively transforming Japan into a premier magnet for world-class technical talent. Having Jürgen join Sakana AI is a testament to this commitment. … ### Hugging Face — **Know Who Spoke When: Build Real-Time, Multi-Speaker AI with NVIDIA Nemotron 3 Diarization** - Company: Hugging Face (huggingface.co) - Announced: 2026-09-23T13:17:01+00:00 - Category: open-weight-release - Coverage: 4 outlets - Announcement: yes - Group: models - Source: https://huggingface.co/blog/nvidia/nemotron-diarization - Record: https://forck.live/items/13419-know-who-spoke-when-build-real-time-multi-speaker-ai-with-nvidia-nemotron-3 - Subject: Platform NVIDIA released Nemotron 3 Diarization, an open-weight 100M-parameter model that identifies which speaker is active at each moment in conversations with up to eight speakers. The model ranks #1 on Voice Arena's Diarization-Bench leaderboard with a 14.72% Diarization Error Rate and supports both live streaming and recorded audio with overlapping speech handling. ### OpenAI — OpenAI extends cyber access to Ukraine for civilian defense - Company: OpenAI (openai.com) - Announced: 2026-09-23 - Category: availability-change - Coverage: 4 outlets - Announcement: yes - Group: covered - Source: https://openai.com/index/openai-extends-cyber-access-to-ukraine-for-civilian-defense - Record: https://forck.live/items/13400-openai-extends-cyber-access-to-ukraine-for-civilian-defense - Subject: GPT / ChatGPT / API OpenAI will offer the Government of Ukraine access to its Daybreak program to support the cyber defense of civilian infrastructure, providing tools to identify software vulnerabilities and develop and test fixes more quickly. ### OpenAI — Sam Altman’s remarks at the United Nations Security Council - Company: OpenAI (openai.com) - Announced: 2026-09-23T12:00:00+00:00 - Category: safety-policy-update - Coverage: 30 outlets - Announcement: no - Group: covered - Source: https://openai.com/index/sam-altman-un-security-council-remarks - Record: https://forck.live/items/13470-sam-altman-s-remarks-at-the-united-nations-security-council - Subject: GPT / ChatGPT / API OpenAI CEO Sam Altman addressed the United Nations Security Council on AI safety and governance, emphasizing the need to keep powerful AI systems under human control, avoid concentrating power in too few hands, and pursue international cooperation. He outlined principles for responsible AI development, including maintaining human agency in decision-making and ensuring alignment and monitorability as systems become more capable. ### OpenAI — Harvey turns legal context into stronger drafts with GPT-6 Astra - Company: OpenAI (openai.com) - Announced: 2026-09-23T12:00:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/harvey-from-context-to-confidence-with-astra - Record: https://forck.live/items/13457-harvey-turns-legal-context-into-stronger-drafts-with-gpt-6-astra - Subject: GPT / ChatGPT / API Harvey, a legal AI platform, now uses GPT-6 Astra to generate more structured, context-aware legal documents with improved formatting and the ability to incorporate lawyer preferences directly into the drafting workflow. ### OpenAI — How invideo improves color grading 3x with GPT‑6 Astra - Company: OpenAI (openai.com) - Announced: 2026-09-23T12:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/invideo-builds-with-gpt-6-astra - Record: https://forck.live/items/13456-how-invideo-improves-color-grading-3x-with-gpt-6-astra - Subject: GPT / ChatGPT / API With GPT‑6 Astra, invideo plans edits with greater precision, improves color correction and grading threefold, and produces 50 custom effects in one day. Results Custom effects created in one day Results Increase in success rate for color-grading and correction Editing videos is one creative and technical decision after another. Editors have to shape the story, place sound effects, build transitions, adjust color, and more. Invideo is an agentic video editor designed to handle that work while keeping the human editor in control. With GPT‑6 Astra, invideo sees greater capability to plan and execute complex edits and place changes at the correct points in the timeline. “How Astra can plan a particular edit on a frame-level accuracy is quite stunning.” Planning complex edits and staying on brief The AI agent must translate the editor’s direction into a sequence of steps, choose the right tools, execute the work, and verify the result. Sanket says GPT‑6 Astra can plan these operations in fewer reasoning steps than the models it previously tested: “Astra is using far fewer reasoning sets and thus output tokens to complete complex work.” … ### Meta — Introducing Meta VR Glasses: A Cinema, Courtside Seat, and Workspace in Just 100 Grams - Company: Meta (meta.com) - Announced: 2026-09-23T11:42:46+00:00 - Category: product-launch - Coverage: 4 outlets - Announcement: no - Group: covered - Source: https://about.fb.com/news/2026/09/introducing-meta-vr-glasses-3d-movies-immersive-live-sports-100-grams/ - Record: https://forck.live/items/13493-introducing-meta-vr-glasses-a-cinema-courtside-seat-and-workspace-in-just-100 - Subject: Meta AI Meta announced Meta VR Glasses, a lightweight VR device weighing approximately 100 grams featuring a 5K Infinite Display, Qualcomm Snapdragon Reality Elite processor, and a separate battery puck. The device will launch in Spring 2027 and includes partnerships with Disney+, ESPN, and other media platforms for 3D content and immersive sports viewing, as well as hologram calling technology. ### Meta — Introducing Ray-Ban Meta Audio and More AI Glasses Styles - Company: Meta (meta.com) - Announced: 2026-09-23T11:36:25+00:00 - Category: not stated - Coverage: 10 outlets - Announcement: yes - Group: covered - Source: https://about.fb.com/news/2026/09/introducing-ray-ban-meta-audio-glasses-new-styles-plus-muse/ - Record: https://forck.live/items/13492-introducing-ray-ban-meta-audio-and-more-ai-glasses-styles - Subject: Meta AI Today at Connect, we announced Ray-Ban Meta Audio, our first-ever audio glasses, and our biggest expansion of AI glasses yet through our partnership with EssilorLuxottica. We’re building AI glasses for everyone. By the end of the year, we’ll offer more than 100 different glasses options across Ray-Ban, Oakley, and Meta Glasses, including lightweight frames made for all-day wear and new styles with slimmer designs and longer battery life. And with regular software updates, new AI features, and our growing ecosystem of apps and experiences, these glasses will only keep getting better. AI glasses help you on the go, from giving you sports scores and local restaurant picks to translating live conversations. With a connection to your personal AI agent from Muse , they’ll offer more ways than ever to help you manage your daily life — whether you’re building healthy habits or navigating a busy calendar — all without reaching for your phone. Ray-Ban Meta Audio Audio has been our most popular feature since day one because our glasses give you high-quality sound without closing you off from the world around you. … ### OpenAI — Introducing MentalHealthBench - Company: OpenAI (openai.com) - Announced: 2026-09-23T10:00:00+00:00 - Category: not stated - Coverage: 2 outlets - Announcement: yes - Group: covered - Source: https://openai.com/index/introducing-mentalhealthbench - Record: https://forck.live/items/13461-introducing-mentalhealthbench - Subject: GPT / ChatGPT / API An open benchmark developed with more than 80 licensed mental health experts to evaluate AI responses in realistic mental health conversations. People turn to AI for many kinds of conversations: navigating a difficult relationship, working through everyday stress, supporting someone they care about, or deciding how to approach a challenging situation. These conversations require accuracy, practical judgment, and respect for people’s agency. With more than one billion people using ChatGPT each week, our research focuses on helping models respond with care across a wide range of needs and put people’s safety and well-being first. Most evaluations of AI in this domain have focused primarily on emergency scenarios, given their importance to safety, and measure success using broad, predefined criteria. This has left a gap in understanding how models perform across the full range of mental health conversations, and how well their responses align with expert guidance for each situation, beyond whether they avoid disallowed responses. … ### NVIDIA — At AI Day Singapore, NVIDIA and Partners Showcase AI Advancements Across Southeast Asia - Company: NVIDIA (nvidia.com) - Announced: 2026-09-23T02:30:22+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://blogs.nvidia.com/blog/ai-day-singapore/ - Record: https://forck.live/items/13203-at-ai-day-singapore-nvidia-and-partners-showcase-ai-advancements-across - Subject: AI platform At NVIDIA AI Day Singapore, NVIDIA and regional partners showcased AI advancements across Southeast Asia. Singapore's HTX is using Nemotron 3 Super and Nemotron 3 Nano Omni models for public safety research. Across the region, organizations are fine-tuning Nemotron models for local languages and applications: Malaysia's YTL AI Labs and Vietnam's Viettel AI for enterprise services, Thailand's iApp Technology for legal applications using OpenThai 2.0 Legal, and AI Singapore expanding its SEA-LION model family with Nemotron. NVIDIA Cosmos and VSS Blueprint are being applied to smart city solutions in Malaysia and Thailand. ### GitHub — OpenTelemetry in the GitHub Copilot app - Company: GitHub (github.com) - Announced: 2026-09-23T02:14:54+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-22-opentelemetry-in-the-github-copilot-app - Record: https://forck.live/items/13224-opentelemetry-in-the-github-copilot-app - Subject: Copilot GitHub Copilot app now supports OpenTelemetry configuration through enterprise-managed settings, allowing administrators to send agent activity data to compatible monitoring tools for analyzing sessions, investigating behavior, and managing telemetry centrally. ### OpenAI — ChatGPT Ads expands to Southeast Asia and Taiwan - Company: OpenAI (openai.com) - Announced: 2026-09-23T02:00:00+00:00 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/chatgpt-ads-expands-southeast-asia-taiwan - Record: https://forck.live/items/13550-chatgpt-ads-expands-to-southeast-asia-and-taiwan - Subject: GPT / ChatGPT / API OpenAI expanded ChatGPT Ads to seven additional markets in Southeast Asia and Taiwan—Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam, and Taiwan—bringing the total availability to more than 60 countries. Ads are shown only to users on Free and Go plans, while Pro, Plus, and Enterprise subscriptions remain ad-free. ### OpenAI — Airbnb widens access to GPT-6 Astra and OpenAI frontier models - Company: OpenAI (openai.com) - Announced: 2026-09-23T01:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/airbnb-gpt-6-astra - Record: https://forck.live/items/13467-airbnb-widens-access-to-gpt-6-astra-and-openai-frontier-models - Subject: GPT / ChatGPT / API Airbnb widens access to GPT‑6 Astra and OpenAI frontier models Under a new agreement, Airbnb is giving its engineering and product development teams broader access to OpenAI frontier models, including GPT‑6 Astra. The expansion builds on Airbnb’s use of Codex and longstanding collaboration with OpenAI. Airbnb engineers use an internal AI assistant for writing software and creating remote AI agents powered by Codex and models like GPT‑5.6 Sol, Terra, and Luna. The new agreement gives Airbnb broader access to OpenAI models through OpenAI APIs and Amazon Bedrock. It arrives amid accelerating adoption of GPT‑6 Astra among developers, builders, and enterprises, many of which were already benefiting from the improved price-performance and efficiency of GPT‑5.6 Terra and Luna. Airbnb’s early use of GPT‑6 Astra has shown strong results beyond coding. Astra helps Airbnb engineers accelerate their work by tracking down hard bugs, shaping system designs, and brainstorming engineering approaches. In one test involving strategic documents and other non-coding work, one Airbnb user reported reaching impressive output in 3–4 passes, compared with 20+ rounds using other models. … ### GitHub — New features and improvements in Copilot for JetBrains - Company: GitHub (github.com) - Announced: 2026-09-23T00:34:18+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-22-new-features-and-improvements-in-copilot-for-jetbrains - Record: https://forck.live/items/13202-new-features-and-improvements-in-copilot-for-jetbrains - Subject: Copilot GitHub Copilot for JetBrains 1.18.0 adds AI-assisted tool approvals in public preview, allowing low-risk tool calls to receive automatic approval while higher-risk actions require user confirmation. The update also introduces message re-editing in agent sessions, support for organization and enterprise skills and custom instructions, plan mode for the Codex agent, persistent per-tool controls for MCP servers, and a side-by-side chat panel switcher. ### Apple — How to Guide Your Language Flow - Company: Apple (apple.com) - Announced: 2026-09-23 - Category: research-paper - Coverage: not counted - Announcement: no - Group: routine - Source: https://machinelearning.apple.com/research/guide-language-flow - Record: https://forck.live/items/13489-how-to-guide-your-language-flow - Subject: Machine Learning Research Apple researchers introduce probe guidance, a method to guide flow matching models that uses frozen internal states from existing diffusion models to construct guidance signals. Applied to a 1.7B diffusion language model, probe guidance achieves state-of-the-art performance on unconditional generation and improves multiple choice question answering benchmarks while eliminating the need for additional forward passes at inference time. ### Together AI — How to train your own Jev for $17 - Company: Together AI (together.ai) - Announced: 2026-09-23 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.together.ai/blog/how-to-train-your-own-jev - Record: https://forck.live/items/13488-how-to-train-your-own-jev-for-17 - Subject: Inference platform 40+ Models Chosen for Production...40+ Models Chosen for Production...40+ Models Chosen for Production... Summary We just launched our own Jev-like classifier, together/Tev1-4B-experimental , on top of Qwen3.5 4B on Together’s serverless platform. In this blog post we’ll show you how to fine-tune your own version! Jev has quickly become one of the most talked about model releases in the AI space. It’s a powerful classification model that’s both fast and incredibly cheap to run. Give Jev a piece of state plus predefined questions and it will quickly give back a result in the form of a score, boolean value, or multiple choice answer. This sort of classification model has many real-world applications, such as an e-commerce site evaluating automated customer returns, categorizing ML papers, or even providing a sentiment rating for a piece of text. Today we’re going to fine-tune our own Jev-like classification model that takes state and returns an answer. Our goal is to create a model that can quickly and efficiently answer questions like: Customer message: Hi, I checked my statement and your company charged my card twice for the October subscription. … ### Lovable — You can now chat with Lovable for free - Company: Lovable (lovable.dev) - Announced: 2026-09-23 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lovable.dev/blog/chat-for-free - Record: https://forck.live/items/13475-you-can-now-chat-with-lovable-for-free - Subject: Lovable / AI app builder Lovable introduced free daily chat functionality across Free, Pro, and Business workspaces, allowing users to explore ideas, evaluate changes, and work with existing projects before committing to paid build or plan work. The feature enables users to connect tools like Notion, Granola, and Linear for context, inspect app code and Lovable Cloud databases, and reuse components across projects. ### Meta — Bringing Your Muse to Life - Company: Meta (meta.com) - Announced: 2026-09-23 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://research.meta.ai/blog/bringing-your-muse-to-life - Record: https://forck.live/items/13473-bringing-your-muse-to-life - Subject: Muse Today, we’re introducing Muse Realtime Avatar, our state-of-the-art embodiment technology that turns Muse Realtime Voice into expressive, interactive avatars. Conditioned on reference media, Muse Realtime Avatar brings any character into a live conversation. A photographic portrait responds through subtle expressions, while a full-body illustration gestures and shifts posture as it speaks. Animals and everyday objects become expressive without losing what makes them distinctive. Frame by frame, the avatar’s appearance and mannerisms remain coherent from one conversational turn to the next. Beyond talking heads, Muse Realtime Avatar brings any image to life in real time, with expressive facial, hand, and full-body movement. From Intelligence to Real-Time Presence Muse Realtime Voice and Muse Realtime Avatar form a single streaming system connecting intelligence, voice, and embodiment. Muse Realtime Voice provides the conversational intelligence and produces a stream of speech tokens (VQs) carrying both what is said and how it’s delivered. … ### Perplexity — AI for market research: a step-by-step guide - Company: Perplexity (perplexity.ai) - Announced: 2026-09-23 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.perplexity.ai/hub/blog/ai-for-market-research - Record: https://forck.live/items/13471-ai-for-market-research-a-step-by-step-guide - Subject: Perplexity Perplexity published a guide on using AI tools to conduct market research, covering the full process from defining research objectives through market sizing, customer segmentation, competitive analysis, and positioning decisions. The guide includes example prompts and verification steps for each stage. ### Cursor — Rollouts and Security Review - Company: Cursor (cursor.com) - Announced: 2026-09-23 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cursor.com/changelog/rollouts-and-security-reviewer - Record: https://forck.live/items/13464-rollouts-and-security-review - Subject: Cursor Today we're launching two Cursor bots for the last mile of shipping code. Rollouts watches every change as it deploys and reports its health per environment. Security Review reports exploitable bugs on every pull request. Both are available today on Teams and Enterprise plans. Rollouts Rollouts attaches a monitor to every pull request and watches the change as it deploys, reporting change health per environment: verified healthy, regression detected, or inconclusive. It's the Cursor version of Firetiger Change Monitors, rebuilt with the Bot Development Kit. Enable it from the dashboard and connect source control, your deploy system, and your telemetry provider. Rollouts starts watching on the next pull request. Monitoring plans When a pull request opens, Rollouts reads the diff and the systems it touches, then writes a monitoring plan as a PR comment. The plan lists the risks it identified, the effect the change is meant to have, the signals it will check, and any gaps in instrumentation that would make the change hard to verify. Edit the plan in the PR and Rollouts uses your version. Deploy tracking … ### Cartesia — Introducing Multilingual Voices - Company: Cartesia (cartesia.ai) - Announced: 2026-09-23 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.cartesia.ai/blog/multilingual-voices - Record: https://forck.live/items/13458-introducing-multilingual-voices - Subject: Sonic / Ink Cartesia introduced Multilingual Voices, a feature enabling 50+ library voices to speak up to 25 languages natively while maintaining consistent brand identity across languages. The company also added the ability to make custom voice clones multilingual by adding up to 9 new accents and languages, with native speaker evaluation of accent quality and localization of dates, numbers, and currency. ### Anthropic — Claude discovers a novel enzyme system with CRISPR-like repeats - Company: Anthropic (anthropic.com) - Announced: 2026-09-23 - Category: research-paper - Coverage: 11 outlets - Announcement: yes - Group: covered - Source: https://www.anthropic.com/news/claude-discovers-novel-enzyme-system - Record: https://forck.live/items/13449-claude-discovers-a-novel-enzyme-system-with-crispr-like-repeats - Subject: Claude Anthropic introduced a new life sciences research group and laboratory focused on fundamental biology research using Claude, and shared early results in which Claude autonomously discovered a novel enzyme system with CRISPR-like properties. The system, called array-associated reverse transcriptases (ARTs), was identified after Claude agents searched a DNA database for 21 hours using 210 million tokens, and the findings are detailed in a pre-print. The lab operates at BSL-1 and BSL-2 biosafety levels and does not handle human-infecting pathogens. ### Anthropic — Claude Marketplace: one place to discover plugins, agents, and services from our partners - Company: Anthropic (anthropic.com) - Announced: 2026-09-23 - Category: product-launch - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://claude.com/blog/claude-marketplace - Record: https://forck.live/items/13446-claude-marketplace-one-place-to-discover-plugins-agents-and-services-from-our - Subject: Claude Anthropic launched Claude Marketplace, a unified platform where customers can discover and integrate plugins, connectors, agents, and services from partners including Atlassian, Google, Microsoft, Notion, Salesforce, CrowdStrike, Cursor, Harvey, Legora, Lovable, Snowflake, and consulting firms. The marketplace enables builders to list tools built with Model Context Protocol and Agent Skills, and allows customers to apply committed Anthropic spend toward Claude-powered software and partner services. ### Anthropic — How CodeRabbit, Power Digital, and ThoughtSpot scale with Snowflake and Vercel on Claude Marketplace - Company: Anthropic (anthropic.com) - Announced: 2026-09-23 - Category: not stated - Coverage: 4 outlets - Announcement: yes - Group: covered - Source: https://claude.com/blog/how-coderabbit-power-digital-and-thoughtspot-scale-with-snowflake-and-vercel-on-claude-marketplace - Record: https://forck.live/items/13445-how-coderabbit-power-digital-and-thoughtspot-scale-with-snowflake-and-vercel - Subject: Claude CodeRabbit expanded its Vercel plan through Claude Marketplace, and Power Digital and ThoughtSpot expanded their Snowflake capacity using their existing Anthropic commitment. Category Enterprise AI Product Claude Platform Date September 23, 2026 Reading time 5 min Share Copy link https://claude.com/blog/how-coderabbit-power-digital-and-thoughtspot-scale-with-snowflake-and-vercel-on-claude-marketplace Companies building with Claude also rely on other software to get the work done, like Snowflake to store and analyze their data or Vercel to run their apps. Claude Marketplace lets companies with an Anthropic commitment put part of it toward tools their teams already rely on, so one investment covers more of what their teams use every day, without a new budget request. Here's how CodeRabbit, Power Digital and ThoughtSpot used Claude Marketplace to expand their work with Vercel and Snowflake. Power Digital runs its clients' data on Snowflake Power Digital is a marketing agency that builds Nova, a platform for its teams and clients. Clients connect their data through Nova, which runs on Snowflake and powers reporting, in-house tools and models that improve marketing performance. … ### Anthropic — How to prepare for AI-driven code modernization projects - Company: Anthropic (anthropic.com) - Announced: 2026-09-23 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://claude.com/blog/how-to-prepare-for-ai-driven-code-modernization-projects - Record: https://forck.live/items/13420-how-to-prepare-for-ai-driven-code-modernization-projects - Subject: Claude Anthropic published a guide for enterprises on organizing AI-driven code modernization projects. The guide covers six steps: defining the target state, creating acceptance criteria, setting a promotion policy, preparing prerequisites, building an agentic workflow using Claude Code, and running the modernization. It addresses how organizations must adapt their change management and review processes when AI agents accelerate code production. ### GitHub — Faster C++ code intelligence with whole codebase indexing - Company: GitHub (github.com) - Announced: 2026-09-22T22:24:48+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-22-faster-c-code-intelligence-with-whole-codebase-indexing - Record: https://forck.live/items/13131-faster-c-code-intelligence-with-whole-codebase-indexing - Subject: Copilot GitHub Copilot CLI now supports whole codebase indexing for C++ repositories, creating a persistent index of symbols across projects to speed up code-intelligence requests for definitions, references, and symbol searches. The feature is enabled by default and uses the Microsoft C++ Language Server to resolve types and relationships between files, though initial indexing may increase memory usage on large repositories. ### OpenAI — Better prompt caching for GPT-6 - Company: OpenAI (openai.com) - Announced: 2026-09-22 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/better-prompt-caching-for-gpt-6 - Record: https://forck.live/items/13119-better-prompt-caching-for-gpt-6 - Subject: GPT / ChatGPT / API OpenAI improved prompt caching for GPT-6 with higher cache hit rates by default within a 30-minute window. The company introduced a Prompt Caching Dashboard to monitor cache performance, a diagnostics tool to identify cache misses, and new controls including explicit cache breakpoints, reasoning effort adjustment without breaking cache, and cache prewarming to help developers reduce latency and costs. ### Perplexity — GPT-6 Sol - Company: Perplexity (perplexity.ai) - Announced: 2026-09-22T20:04:06+00:00 - Category: api-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://docs.perplexity.ai/docs/resources/changelog#september-2026 - Record: https://forck.live/items/13130-gpt-6-sol - Subject: Sonar Perplexity's Agent API now supports four additional models: OpenAI's GPT-6 Sol and GPT-6 Luna, Anthropic's Claude Opus 5.5, and xAI's Grok 4.7. ### Perplexity — GPT-6 Luna - Company: Perplexity (perplexity.ai) - Announced: 2026-09-22T20:04:06+00:00 - Category: api-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://docs.perplexity.ai/docs/resources/changelog#september-2026 - Record: https://forck.live/items/13127-gpt-6-luna - Subject: Sonar Perplexity's Agent API now supports three additional models: OpenAI's GPT-6 Luna, Anthropic's Claude Opus 5.5, and xAI's Grok 4.7. ### Microsoft — GPT-6 Astra, Sol, and Luna: For production agents in Microsoft Foundry - Company: Microsoft (microsoft.com) - Announced: 2026-09-22T18:30:00+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://azure.microsoft.com/en-us/blog/gpt-6-astra-sol-and-luna-for-production-agents-in-microsoft-foundry/ - Record: https://forck.live/items/13754-gpt-6-astra-sol-and-luna-for-production-agents-in-microsoft-foundry - Subject: Azure AI Today, we are expanding our GPT-6 series by welcoming GPT-6 Sol and GPT-6 Luna to our generally available lineup in Microsoft Foundry . Building on the exceptional customer momentum of GPT-5.6 Sol and GPT-6 Astra , this launch continues our work to deliver transformative capabilities in Microsoft Foundry that produce less noise and are more capable at completing full tasks with agents . Explore GPT models in Foundry today Astra brings advanced reasoning , software engineering and computer use to demanding work that requires both judgment and action. Azure customers report a step-change in capabilities, and strong cost-to-performance with the model using fewer, higher-value tokens to drive agents. Completing the lineup, GPT-6 Sol is excellent for general-purpose use, while Luna brings efficient intelligence to high-volume data and preparatory work. Put the right intelligence behind every agent The right model for a job should be determined through evaluations: an agent handling a complex business decision and one routing routine requests have different needs. Microsoft recommends customers start with GPT-6 Astra for demanding work. … ### Amazon — Bring more intelligence to everyday work with GPT-6 Sol and GPT-6 Luna on Amazon Bedrock - Company: Amazon (amazon.com) - Announced: 2026-09-22T18:10:22+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/bring-more-intelligence-to-everyday-work-with-gpt-6-sol-and-gpt-6-luna-on-amazon-bedrock/ - Record: https://forck.live/items/13108-bring-more-intelligence-to-everyday-work-with-gpt-6-sol-and-gpt-6-luna-on - Subject: Bedrock / Nova GPT-6 Sol and GPT-6 Luna are now generally available on Amazon Bedrock, giving you more options to match intelligence and efficiency to each workload. The value of AI at scale depends on two dimensions: what a model can do and how often you can put it to use. Greater intelligence expands the complexity a model can handle, from subtle coding problems to multistep processes across tools. Efficiency determines how broadly that intelligence can support everyday activity and repeatable tasks, where every additional token, retry, and second of latency multiplies across requests. GPT-6 Astra established the upper end of the GPT-6 family for the most ambitious projects, where achieving the highest-quality result matters more than cost. Organizations also need advanced intelligence for the recurring tasks that keep products and operations moving. GPT-6 Sol brings strong reasoning and coding capabilities to complex tasks performed throughout the week, with economics suited to regular use. GPT-6 Luna makes focused, repeatable tasks practical at high volume, where small differences in latency and cost multiply across requests. … ### OpenAI — Introducing GPT-6 Sol and Luna - Company: OpenAI (openai.com) - Announced: 2026-09-22T18:00:00+00:00 - Category: new-model - Coverage: 18 outlets - Announcement: yes - Group: models - Source: https://openai.com/index/introducing-gpt-6-sol-and-luna - Record: https://forck.live/items/13114-introducing-gpt-6-sol-and-luna - Subject: GPT / ChatGPT / API OpenAI announced GPT-6 Sol and GPT-6 Luna, two new models in the GPT-6 family that offer improved cost efficiency while bringing advances from GPT-6 Astra to professional work, factuality, coding, and alignment. API prices for Sol and Luna are reduced by 50% compared to their GPT-5.6 promotional pricing. The models outperform similarly priced competitors on benchmarks such as AutomationBench, Agents' Last Exam, FrontierCode, and DeepSWE at lower cost per task. ### Runway — Introducing DIFFUSE: A New Hiring Platform for the AI-Native Creative Workforce - Company: Runway (runway.com) - Announced: 2026-09-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://runway.com/news/company-news/introducing-diffuse - Record: https://forck.live/items/13073-introducing-diffuse-a-new-hiring-platform-for-the-ai-native-creative-workforce - Subject: Gen / Aleph Runway launched DIFFUSE, a hiring platform connecting AI-native creative talent with brands, agencies and studios. The platform allows creatives to build profiles and showcase portfolios, while enabling companies to search for and hire talent skilled in generative AI tools. ### Amazon — Claude Opus 5.5 is now available on AWS - Company: Amazon (amazon.com) - Announced: 2026-09-22T17:28:01+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/claude-opus-5-5-is-now-available-on-aws/ - Record: https://forck.live/items/13103-claude-opus-5-5-is-now-available-on-aws - Subject: Bedrock / Nova Today, we’re excited to announce the availability of Claude Opus 5.5 on Amazon Bedrock and Claude Platform on AWS , the first of the Claude 5.5 model family. Claude Opus 5.5 is Anthropic’s most capable Opus model suitable for agentic coding, knowledge work, and long-running tasks. This post covers Claude Opus 5.5’s improvements, practical guidance, and how to start building with the model on Amazon Bedrock. What makes Claude Opus 5.5 different According to Anthropic, Claude Opus 5.5 does more with fewer tokens than Claude Opus 5, and new pricing passes those gains straight to customers. Lower per-token prices and much cheaper cache reads stack on top of the efficiency gains. The result is an average lower cost per task than Claude Opus 5, so teams can run more ambitious agentic work at scale. Claude Opus 5.5 is trained to communicate more clearly. As it works, it surfaces what it did, what it found, and what it needs, making long-running tasks easier to follow. Adaptive thinking is always on, and Opus 5.5 decides how much reasoning each task needs. You can use effort as your control instead of manual thinking budgets. … ### Amazon — Evaluate skill-equipped agents with Strands Evals and Amazon Bedrock AgentCore - Company: Amazon (amazon.com) - Announced: 2026-09-22T17:18:13+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/evaluate-skill-equipped-agents-with-strands-evals-and-amazon-bedrock-agentcore/ - Record: https://forck.live/items/13102-evaluate-skill-equipped-agents-with-strands-evals-and-amazon-bedrock-agentcore - Subject: Bedrock / Nova Amazon Bedrock AgentCore Evaluations and Strands Evals SDK add skill-focused evaluators to measure agent behavior with modular skills. Skill Selection Accuracy determines whether invoked skills are appropriate for tasks, Skill Instruction Following rates how fully agents follow skill instructions on a five-level scale, and Skill Invoked provides deterministic checks for skill loading. ### GitHub — Claude Opus 5.5 is now available in GitHub Copilot - Company: GitHub (github.com) - Announced: 2026-09-22T17:10:23+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://github.blog/changelog/2026-09-22-claude-opus-5-5-is-now-available-in-github-copilot - Record: https://forck.live/items/13106-claude-opus-5-5-is-now-available-in-github-copilot - Subject: Copilot Claude Opus 5.5, Anthropic’s newest Opus model, is now available in GitHub Copilot. You can use it for agentic coding, long-running agentic tasks, and knowledge work. In early testing, Opus 5.5 resolved tasks comparably to Claude Opus 5 while using significantly fewer steps and tokens. It also quickly recovered from errors in multistep tasks. Claude Opus 5.5 watermarks its text outputs. The watermark doesn’t change the meaning, quality, or readability of outputs, nor does it add any tokens or cost. To learn more visit Anthropics’s How Claude’s text watermark works . This model is billed at provider list pricing under usage-based billing. See Models and pricing for GitHub Copilot for details. Claude Opus 5.5 is available to Copilot Pro+, Max, Business, and Enterprise users. You can select the model in the model picker in: Visual Studio Code Visual Studio Copilot CLI GitHub Copilot coding agent GitHub Copilot app github.com GitHub Mobile on iOS and Android JetBrains IDEs Xcode Eclipse Rollout will be gradual. Check back soon if you don’t see it yet. … ### Microsoft — Claude Opus 5.5 comes to Microsoft Foundry for long-running coding and knowledge work - Company: Microsoft (microsoft.com) - Announced: 2026-09-22T17:00:25+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/claude-opus-5-5-comes-to-microsoft-foundry-for-long-running-coding-and-knowledge/4558051 - Record: https://forck.live/items/13755-claude-opus-5-5-comes-to-microsoft-foundry-for-long-running-coding-and - Subject: Azure AI AI models are increasingly taking on work that extends far beyond a single prompt: building a feature across a codebase, investigating a complex issue, synthesizing hundreds of pages of information, or working through a multi-step business process. As that work gets longer, raw intelligence is only part of what matters. The model also needs to stay focused, make good decisions along the way, communicate what it is doing, and produce work that people can quickly review and use. Today, Claude Opus 5.5 is available in Microsoft Foundry, bringing Anthropic’s most capable Opus model to developers and enterprises building AI applications and agents. Claude Opus 5.5 is designed for everyday complex work. It advances Opus 5 across agentic coding, knowledge work, and long-running tasks while making it easier for people to understand what the model did, what it found, and what it needs next. Claude Opus 5.5 also does more with fewer tokens. Lower per-token prices and much cheaper cache reads stack on top of the efficiency gains. Built for work that takes time Writing a function is one thing. … ### GitHub — OpenAI’s GPT-6 Sol and GPT-6 Luna now available - Company: GitHub (github.com) - Announced: 2026-09-22T17:00:14+00:00 - Category: capability-change - Coverage: 4 outlets - Announcement: yes - Group: covered - Source: https://github.blog/changelog/2026-09-22-openais-gpt-6-sol-and-gpt-6-luna-now-available - Record: https://forck.live/items/13113-openai-s-gpt-6-sol-and-gpt-6-luna-now-available - Subject: Copilot OpenAI's GPT-6 Sol and GPT-6 Luna models are now available in GitHub Copilot alongside the previously released GPT-6 Astra. GPT-6 Sol is a balanced model for interactive and agentic coding available to Copilot Pro+, Max, Business, and Enterprise plans. GPT-6 Luna is a lightweight, cost-efficient model for smaller tasks available to Copilot Pro, Pro+, Max, Business, and Enterprise plans. Both models can be selected through the model picker across multiple IDEs and platforms. ### Lovable — Faster builds, same quality: Opus 5.5 now in Lovable - Company: Lovable (lovable.dev) - Announced: 2026-09-22T16:30:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lovable.dev/blog/opus-5-5-now-in-lovable - Record: https://forck.live/items/13100-faster-builds-same-quality-opus-5-5-now-in-lovable - Subject: Lovable / AI app builder Lovable now uses Claude Opus 5.5, which completes app-building tasks in a third to half fewer steps than Opus 5 while maintaining equivalent quality. The model gathers context in a single pass, makes fewer and more complete edits, and reduces unnecessary verification loops. ### Amazon — How Reactiv automates mobile commerce 80% faster with Amazon Bedrock AgentCore - Company: Amazon (amazon.com) - Announced: 2026-09-22T15:46:07+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/how-reactiv-automates-mobile-commerce-80-faster-with-amazon-bedrock-agentcore/ - Record: https://forck.live/items/13084-how-reactiv-automates-mobile-commerce-80-faster-with-amazon-bedrock-agentcore - Subject: Bedrock / Nova Reactiv, a mobile commerce platform for Shopify merchants, built an AI Scheduler using Amazon Bedrock AgentCore to automate app updates, reducing merchant configuration time by 80 percent and achieving 33 percent faster production deployment. The system uses a three-agent architecture with persistent memory and native Model Context Protocol support to enable merchants to schedule autonomous app updates via natural language commands. ### Amazon — Right-size generative AI endpoints with concurrency sweeps on Amazon SageMaker AI - Company: Amazon (amazon.com) - Announced: 2026-09-22T15:35:53+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/right-size-generative-ai-endpoints-with-concurrency-sweeps-on-amazon-sagemaker-ai/ - Record: https://forck.live/items/13081-right-size-generative-ai-endpoints-with-concurrency-sweeps-on-amazon-sagemaker - Subject: Bedrock / Nova Amazon published a guide demonstrating how to use concurrency sweeps in Amazon SageMaker AI Inference Recommendations to right-size generative AI endpoints. The walkthrough covers deploying the NVIDIA Nemotron-3 Nano 30B model, configuring workload profiles, running automated benchmarks via the CreateAIBenchmarkJob API, and using results to optimize instance selection and capacity planning. ### Amazon — How Trane gets building insights 60x faster with Amazon Bedrock AgentCore - Company: Amazon (amazon.com) - Announced: 2026-09-22T15:30:34+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/how-trane-gets-building-insights-60x-faster-with-amazon-bedrock-agentcore/ - Record: https://forck.live/items/13082-how-trane-gets-building-insights-60x-faster-with-amazon-bedrock-agentcore - Subject: Bedrock / Nova Trane Technologies manages millions of connected heating, ventilation, and air conditioning (HVAC) assets worldwide, but getting a single operational answer could mean cross-referencing multiple dashboards and drilling through menus for 20 minutes or more. For organizations operating at this scale, that kind of friction slows operations, defers corrective action, and creates material business impact across the enterprise. In 3–4 weeks, Trane’s engineering team built an AI-powered agentic solution on Amazon Bedrock AgentCore that reduced a 20-minute multi-screen diagnostic workflow to a 20-second natural language interaction. This is based on Trane’s internal benchmarking with technicians over several weeks. This represents a 60x improvement in time-to-insight, helping shift operations from reactive response to more proactive, data-driven optimization. In this post, we describe the architectural approach and key design decisions behind the solution: Separating agent logic from tool execution. Integrating real-time telemetry through a centralized tool gateway. Tailoring responses to different personas. Trane Technologies and the building intelligence challenge … ### Amazon — How Tata Elxsi detects industrial safety risks in seconds on AWS - Company: Amazon (amazon.com) - Announced: 2026-09-22T15:19:54+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/how-tata-elxsi-detects-industrial-safety-risks-in-seconds-on-aws/ - Record: https://forck.live/items/13078-how-tata-elxsi-detects-industrial-safety-risks-in-seconds-on-aws - Subject: Bedrock / Nova Tata Elxsi's case study documents how the company built IRIS, a real-time industrial safety platform on AWS that uses computer vision to detect unsafe conditions in manufacturing and warehouse environments. The platform processes video at the edge using AWS IoT Greengrass and GPU-equipped servers, streams safety-relevant metadata through Amazon Kinesis Data Streams, and reduces cloud-bound frame volume by 70–80 percent through edge filtering. ### Amazon — Extending public sector intelligence with Agentforce and AWS - Company: Amazon (amazon.com) - Announced: 2026-09-22T15:17:45+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/extending-public-sector-intelligence-with-agentforce-and-aws/ - Record: https://forck.live/items/13077-extending-public-sector-intelligence-with-agentforce-and-aws - Subject: Bedrock / Nova Public sector agencies process large volumes of unstructured evidence, such as body camera footage, surveillance video, and scanned documents, that require extracting insights before anyone can act on them. This post shows how to combine Amazon Bedrock Data Automation with the Model Context Protocol (MCP) to turn unstructured data into structured insights. You can then expose those insights through natural language queries in an AI agent, such as Salesforce Agentforce . In our previous post, Modernizing evidence management in Salesforce Public Sector Solutions with Amazon S3 , we used the External Storage of Files with Amazon Simple Storage Service (Amazon S3) integration from Agentforce Public Sector (formerly Public Sector Solutions) as an example implementation. With that foundation in place, you now have durable, cost-efficient storage for body camera footage, surveillance video, photographs, audio recordings, and scanned documents. However, storage is only half the challenge. Without automation, you spend significant time manually reviewing, classifying, and extracting relevant details from these files before you can act on them. … ### NVIDIA — NVIDIA Isaac ROS 5.0 Advances Agentic, Open Source Robotics Development - Company: NVIDIA (nvidia.com) - Announced: 2026-09-22T12:00:41+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://blogs.nvidia.com/blog/isaac-ros-5-0-agentic-open-source-robotics/ - Record: https://forck.live/items/13022-nvidia-isaac-ros-5-0-advances-agentic-open-source-robotics-development - Subject: AI platform NVIDIA released Isaac ROS 5.0, a collection of GPU-accelerated packages for robotics development built on the open source ROS framework. The release introduces agentic workflows, support for ROS Lyrical and Ubuntu 24.04, new reusable skills for setup and manipulation tasks, and agent-ready documentation to help AI agents assist with robotics development. FoundationPose now provides faster object pose estimation and tracking, while a new FoundationStereo fine-tuning skill enables AI agents to adapt stereo perception models to specific camera and environment configurations. ### OpenAI — Parallel cut research time and cost in half with GPT‑6 Astra - Company: OpenAI (openai.com) - Announced: 2026-09-22T12:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/parallel-cuts-time-and-cost-with-astra - Record: https://forck.live/items/13128-parallel-cut-research-time-and-cost-in-half-with-gpt-6-astra - Subject: GPT / ChatGPT / API OpenAI reports that GPT-6 Astra enabled Parallel's agents to research and synthesize labor-market data in half the time and at half the cost compared to prior models. ### Lovable — Lovable is a founding member of the Blueprint Alliance - Company: Lovable (lovable.dev) - Announced: 2026-09-22T12:00:00+00:00 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lovable.dev/blog/lovable-founding-member-blueprint-alliance - Record: https://forck.live/items/13045-lovable-is-a-founding-member-of-the-blueprint-alliance - Subject: Lovable / AI app builder Lovable joined AWS, CrowdStrike, Databricks, Docker, Google Cloud, Okta, Proofpoint, Salesforce, ServiceNow, Wiz, and Zscaler as a founding member of the Blueprint Alliance, a cross-industry coalition developing an open reference architecture for securing and governing AI agents at enterprise scale. ### Tencent — Tencent Esports Competition Solution to Support Esports at the 20th Asian Games - Company: Tencent (tencent.com) - Announced: 2026-09-22T04:34:12+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.tencent.com/tencent-esports-competition-solution-to-support-esports-at-the-20th-asian-games/ - Record: https://forck.live/items/14413-tencent-esports-competition-solution-to-support-esports-at-the-20th-asian-games - Subject: Hunyuan Tencent's Esports Competition Solution (ECS) will provide end-to-end technical support across all 11 esports events at the 20th Asian Games, covering pre-match equipment checks, real-time monitoring, and post-match review. The solution builds on Tencent's experience at the 19th Asian Games, where it supported 217 matches with zero technical incidents, and underwent three test events in Japan, Singapore, and Vietnam in 2026. The deployment is part of the Olympic Council of Asia's 10-year strategic partnership with Tencent Esports from 2025 to 2035. ### Together AI — Canary rollouts: upgrade models in production without downtime - Company: Together AI (together.ai) - Announced: 2026-09-22 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.together.ai/blog/canary-rollouts-upgrade-models-in-production-without-downtime - Record: https://forck.live/items/13150-canary-rollouts-upgrade-models-in-production-without-downtime - Subject: Inference platform Together AI released canary rollouts, a feature for upgrading models in production without downtime. The system uses staged traffic ramps, metric gates, and automatic rollback on dedicated inference, allowing operators to migrate traffic between two deployments with health checks and optional performance gates at each step. ### Perplexity — The AI literacy gap is growing. Here’s how to close it. - Company: Perplexity (perplexity.ai) - Announced: 2026-09-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.perplexity.ai/hub/blog/ai-literacy - Record: https://forck.live/items/13149-the-ai-literacy-gap-is-growing-here-s-how-to-close-it - Subject: Perplexity Perplexity publishes a guide on AI literacy, defining it as the ability to understand how AI works, recognize its limits and risks, and make informed decisions about its output. The guide emphasizes that evaluation—the ability to verify accuracy, completeness, currency, relevance, and safety of AI-generated responses—is the most critical skill separating novices from experts. It outlines a three-step verification process (Trace, Verify, Decide) and provides examples of how to evaluate different types of AI claims, from factual assertions to calculations and recommendations. ### xAI — How SpaceXAI is using Grok Bot to scale customer support - Company: xAI (x.ai) - Announced: 2026-09-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://x.ai/news/grok-bot-customer-support - Record: https://forck.live/items/13112-how-spacexai-is-using-grok-bot-to-scale-customer-support - Subject: Grok We rebuilt the combined SpaceXAI and Cursor support operation around Grok Bot, expanding to a much broader product portfolio without adding headcount. When Cursor became part of SpaceXAI on August 14, our two customer support teams began coming together around a much broader product portfolio. At the same time, we were preparing to launch Grok Bot , an AI teammate you can give real work to. We expected the product to grow quickly, bringing another wave of users and support demand. We decided to use Grok Bot itself to help meet that demand, putting it to work throughout the support operation. It signed into the same tools our team used and its role stretched from resolving individual tickets to helping us understand and improve the operation as a whole. Our new combined team has seen a 175% increase in support tickets, but we have not had to hire any new people thanks to Grok Bot. We might have hired 200 additional people otherwise. We are also doing it at a fraction of the usual cost. Traditional AI support tools charge a flat $1 to $4 per resolution. With Grok Bot, you only pay for your actual usage, which is already included in your plan. … ### Cohere — AI change management: A human-centric approach - Company: Cohere (cohere.com) - Announced: 2026-09-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://cohere.com/blog/ai-change-management - Record: https://forck.live/items/13111-ai-change-management-a-human-centric-approach - Subject: Command Cohere publishes a guide on AI change management for enterprises, arguing that AI should be treated as an active participant in workflows rather than simply as software to be deployed. The post outlines how organizations must redesign roles, responsibilities, and workflows to accommodate human-AI collaboration, and identifies three categories of work—verifiable, judgment-based, and hybrid—to help determine which tasks can be delegated to AI and which require human judgment. ### OpenAI — Priorities and principles for effective third party assessments - Company: OpenAI (openai.com) - Announced: 2026-09-22 - Category: safety-policy-update - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/priorities-principles-third-party-assessments - Record: https://forck.live/items/13105-priorities-and-principles-for-effective-third-party-assessments - Subject: GPT / ChatGPT / API OpenAI published a framework outlining priorities and principles for independent third-party assessments of its frontier AI models. The company commits to providing deep access to assessors across training, evaluation, and deployment stages, and proposes four priority areas: independent assessment of safety cases spanning training through external deployment, assessment of critical safeguards, and evaluation of safety claims and evidence. OpenAI emphasizes that effective assessments require strong independence mechanisms, scientific rigor, robust security practices, and shared international standards for safety and security. ### Hugging Face — How UK AISI and EvalEval Are Making Benchmark Results Reproducible - Company: Hugging Face (huggingface.co) - Announced: 2026-09-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/evaleval-aisi - Record: https://forck.live/items/13097-how-uk-aisi-and-evaleval-are-making-benchmark-results-reproducible - Subject: Platform The EvalEval Coalition is thrilled to share that the UK AI Security Institute (AISI) is using EvalEval's infrastructure to openly share evaluation results, supporting more reproducible and verifiable evaluation science. AISI and EvalEval have previously collaborated on research that began at a joint workshop alongside NeurIPS 2025 , and feedback from the Institute has helped shape the Every Eval Ever (EEE) schema . This next phase of the collaboration puts that shared infrastructure into practice. Why reproducible evaluation reporting matters As AI deployment accelerates, evaluations are becoming increasingly important sources of evidence about model and system performance. Yet results are reported across many formats, platforms, and outlets, often without enough information to reproduce them. Running the evaluations again may itself be prohibitively expensive. EvalEval's mission is to improve this ecosystem through a shared reporting schema, Every Eval Ever , and an open platform, Evaluation Cards , that brings evaluation results and the information needed to interpret them into a common structure. … ### Anthropic — What a task costs on Opus 5.5 - Company: Anthropic (anthropic.com) - Announced: 2026-09-22 - Category: pricing-change - Coverage: 6 outlets - Announcement: yes - Group: covered - Source: https://claude.com/blog/what-a-task-costs-on-opus-5-5 - Record: https://forck.live/items/13095-what-a-task-costs-on-opus-5-5 - Subject: Claude Anthropic announced pricing for Claude Opus 5.5, which costs less per token than Opus 5. Input and output tokens are 20% cheaper, and cache reads are 60% cheaper. The post explains how task costs vary based on the number of turns, cache hit rates, output token usage, and model selection. ### Anthropic — Introducing Claude Opus 5.5 - Company: Anthropic (anthropic.com) - Announced: 2026-09-22 - Category: model-update - Coverage: 22 outlets - Announcement: yes - Group: models - Source: https://www.anthropic.com/claude-opus-5-5 - Record: https://forck.live/items/13091-introducing-claude-opus-5-5 - Subject: Claude Anthropic released Claude Opus 5.5, a new model that performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5. The model achieves the best scores on Anthropic's automated behavioral audit and generates output more than 30% faster than Opus 5. Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks. ### Cognition — Building the Future of Software Engineering in Latin America - Company: Cognition (cognition.com) - Announced: 2026-09-22 - Category: availability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://cognition.com/blog/devin-comes-to-sao-paulo - Record: https://forck.live/items/13076-building-the-future-of-software-engineering-in-latin-america - Subject: Devin / SWE Cognition announced its expansion into Latin America, with São Paulo as the starting point. The company highlighted that major regional institutions including Itaú, Nubank, Santander, Natura, and EBANX are already using Devin, with Itaú reporting a 20-30% increase in engineering throughput and Nubank completing a multi-million-line monolith migration in weeks at over 20x lower cost. ### Suno — What Does a Compressor Actually Do? - Company: Suno (suno.com) - Announced: 2026-09-22 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://suno.com/blog/about-compression - Record: https://forck.live/items/13070-what-does-a-compressor-actually-do - Subject: Music / Studio Suno Studio, the company's browser-based DAW, now includes a suite of audio effects including a compressor plugin that allows users to manage dynamic range in their mixes. The compressor is available to all Suno Premier subscribers at no additional cost. ### Amp — One Runner, Many Worktrees - Company: Amp (ampcode.com) - Announced: 2026-09-22 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/news/one-runner-many-worktrees - Record: https://forck.live/items/13044-one-runner-many-worktrees - Subject: Amp Amp's runner now supports creating Git worktrees, allowing users to start threads in fresh checkouts without affecting the main repository. The runner can also access configured Secrets & Env Vars from ampcode.com, with variables applied per thread based on personal, project, and workspace scopes. ### Hugging Face — Transformers now runs llama.cpp quants - Company: Hugging Face (huggingface.co) - Announced: 2026-09-22 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/transformers-llama-cpp-quants - Record: https://forck.live/items/13000-transformers-now-runs-llama-cpp-quants - Subject: Platform Hugging Face added support for running GGUF-quantized models efficiently in transformers through the familiar transformers APIs. Users can load GGUF checkpoints from the Hub using from_pretrained and generate locally on their machines, with initial focus on Apple Silicon and the Qwen3.5 architecture. ### LG AI Research — [ICML2026] MCFlow, a New Integrated Model for AI-Generated Inorganic Materials - Company: LG AI Research (lgresearch.ai) - Announced: 2026-09-22 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.lgresearch.ai/blog/view?seq=695 - Record: https://forck.live/items/12954-icml2026-mcflow-a-new-integrated-model-for-ai-generated-inorganic-materials - Subject: EXAONE LG AI Research developed MCFlow (Multimodal Crystal Flow), a single integrated generative model for inorganic materials that can perform crystal structure prediction from composition, de novo generation of both composition and structure, and generation of atomic species for a given structure. The research was accepted to ICML 2026. ### Hugging Face — Jun Kim, oMLX creator and maintainer, joins Hugging Face to support the MLX community - Company: Hugging Face (huggingface.co) - Announced: 2026-09-22 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/omlx - Record: https://forck.live/items/12919-jun-kim-omlx-creator-and-maintainer-joins-hugging-face-to-support-the-mlx - Subject: Platform Jun Kim, creator and maintainer of oMLX, has joined Hugging Face to support the MLX community. oMLX will transition from a side project to a fully maintained and funded initiative while remaining Apache 2.0 licensed, with Kim continuing to lead the project. ### Amazon — xAI’s Grok 4.6 is now available in Amazon Bedrock - Company: Amazon (amazon.com) - Announced: 2026-09-21T18:30:34+00:00 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://aws.amazon.com/blogs/machine-learning/xais-grok-4-6-is-now-available-in-amazon-bedrock/ - Record: https://forck.live/items/12741-xai-s-grok-4-6-is-now-available-in-amazon-bedrock - Subject: Bedrock / Nova xAI's Grok 4.6 is now available in Amazon Bedrock as of August 18, 2026, accessible via both bedrock-mantle and bedrock-runtime endpoints with support for the Converse API. The model offers a 500K token context window and configurable reasoning effort at four levels, with cross-region inference options for US and global routing. ### NVIDIA — NVIDIA Launches DSX Ready to Qualify Power and Cooling Products for AI Factories - Company: NVIDIA (nvidia.com) - Announced: 2026-09-21T18:00:07+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://blogs.nvidia.com/blog/dsx-ready-ai-factories-power-cooling/ - Record: https://forck.live/items/12738-nvidia-launches-dsx-ready-to-qualify-power-and-cooling-products-for-ai-factories - Subject: AI platform Every AI factory needs power and cooling that fit its computing architecture. As AI infrastructure expands, power, cooling, water, site and grid constraints are shaping what builders can deploy. Choosing products that fit the complete factory design helps builders turn computing capacity into useful AI output. To help builders make those decisions, NVIDIA is introducing NVIDIA DSX Ready, a qualification program for partner products and solutions that meet applicable NVIDIA DSX AI factory reference design requirements. The program launches with two initial categories: battery energy storage systems (BESS) and cooling distribution units (CDUs). Category-specific requirements and review through the program help builders evaluate offerings with greater confidence, reduce integration risk and move toward deployment. Qualified Building Blocks for Building AI Factory The NVIDIA DSX AI factory platform unifies AI factory design and operations across compute, networking, power, cooling, facilities and software. It helps partners design and operate the factory as one system to produce more useful AI output within available power, cooling, water and grid constraints. … ### Amazon — How BMW Group detects cost anomalies across 14,000 cloud accounts - Company: Amazon (amazon.com) - Announced: 2026-09-21T16:36:10+00:00 - Category: not stated - Coverage: 3 outlets - Announcement: no - Group: covered - Source: https://aws.amazon.com/blogs/machine-learning/how-bmw-group-detects-cost-anomalies-across-14000-cloud-accounts/ - Record: https://forck.live/items/12729-how-bmw-group-detects-cost-anomalies-across-14-000-cloud-accounts - Subject: Bedrock / Nova BMW Group's case study documents how its Cloud Efficiency Analytics (CLEA) system, built on AWS with Reply, detects cost anomalies across 14,000 cloud accounts using Prophet forecasting and daily anomaly detection that sends alerts to account owners when spending departs from expected patterns. ### Amazon — Run Positron on Amazon SageMaker AI for data science workflows - Company: Amazon (amazon.com) - Announced: 2026-09-21T16:34:21+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/run-positron-on-amazon-sagemaker-ai-for-data-science-workflows/ - Record: https://forck.live/items/12730-run-positron-on-amazon-sagemaker-ai-for-data-science-workflows - Subject: Bedrock / Nova Data science teams often move among separate tools for governed data access, R analysis, Python model development, deployment, application development, and reporting. Positron, Posit’s integrated development environment (IDE) for data science, now runs on Amazon SageMaker AI. For a data scientist, running Positron on SageMaker AI means: Data access without managing credentials. Positron runs under the Space execution role, so you query Amazon Athena, the AWS Glue Data Catalog, and Amazon Simple Storage Service (Amazon S3) directly from the IDE. Access follows the role’s permissions, with no keys to store or rotate. Compute that is ready when you are. You launch a Space on the instance size you need, and teams can reserve capacity with SageMaker AI training plans so compute is available for scheduled training. AI assistance that stays in your account. Posit Assistant, Posit’s AI coding assistant, can use Amazon Bedrock as its model provider, so AI help runs on models in your own AWS account and AWS Region. Room to work in parallel and together. … ### Amazon — How Benchling secured multi-tenant AI agents with Amazon Bedrock AgentCore - Company: Amazon (amazon.com) - Announced: 2026-09-21T16:27:34+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/how-benchling-secured-multi-tenant-ai-agents-with-amazon-bedrock-agentcore/ - Record: https://forck.live/items/12731-how-benchling-secured-multi-tenant-ai-agents-with-amazon-bedrock-agentcore - Subject: Bedrock / Nova Benchling's case study documents how the company secured multi-tenant AI agent code execution using Amazon Bedrock AgentCore in VPC mode, combining account-level isolation, DNS firewall controls, and VPC endpoint policies to prevent data exfiltration while processing over 600 code execution sessions per day across more than 250 tenants with zero security incidents. ### Amazon — Reducing medical claims review time with AI on AWS: The EXL Medical IDP solution - Company: Amazon (amazon.com) - Announced: 2026-09-21T16:24:40+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://aws.amazon.com/blogs/machine-learning/reducing-medical-claims-review-time-with-ai-on-aws-the-exl-medical-idp-solution/ - Record: https://forck.live/items/12727-reducing-medical-claims-review-time-with-ai-on-aws-the-exl-medical-idp-solution - Subject: Bedrock / Nova EXL's Medical IDP solution, built on AWS, automates the review of medical records for insurance claims by combining intelligent document processing with domain-specific large language models to extract, summarize, and query medical information, reducing manual review time from over 100 minutes per case. ### NVIDIA — Why Deploying Physical AI at Scale Demands Safety at Every Layer - Company: NVIDIA (nvidia.com) - Announced: 2026-09-21T16:00:46+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://blogs.nvidia.com/blog/physical-ai-halos-safety/ - Record: https://forck.live/items/12724-why-deploying-physical-ai-at-scale-demands-safety-at-every-layer - Subject: AI platform Physical AI is moving rapidly from research to large-scale deployment. By 2035, ABI Research projects an installed base of 49 million level 3-5 autonomous vehicles (AVs) , while Omdia estimates that roughly 60 million industrial robots will be deployed between 2026 and 2035 . As these machines enter roads, factories, warehouses and other environments shared with people, safety must scale with them. Physical AI safety means proving that AI-driven machines — AVs , humanoid robots , industrial robots and more — behave safely when their decisions turn into physical action. That requires safety across the hardware, software, AI, operating environment and deployment lifecycle — not a one-time check before deployment. Why Is Safety the Key to Scaling Physical AI? After years of testing and benchmarking, AVs continue to expand commercially. That progress has required developers to demonstrate how automated systems address potential hardware and software failures, limitations in intended functionality and AI-specific risks. Robotics is approaching a similar inflection point as autonomous machines move into factories, warehouses and other environments shared with people. … ### Meta — Open-Sourcing Rebalancer: A Generic, High-Performance Library for Solving Assignment Problems - Company: Meta (meta.com) - Announced: 2026-09-21T16:00:37+00:00 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://engineering.fb.com/2026/09/21/open-source/rebalancer-generic-high-performance-library-assignment-problems/ - Record: https://forck.live/items/12723-open-sourcing-rebalancer-a-generic-high-performance-library-for-solving - Subject: Llama / infrastructure We’re open-sourcing Rebalancer , the assignment-problem solver that has been used to solve resource allocation problems throughout Meta for over nine years. Rebalancer separates several related concerns: how to specify an assignment problem, how to store it efficiently in memory, how to solve it, and how to debug it. This separation of concerns is crucial to Rebalancer’s usability, scalability, and extensibility. For a more detailed technical exposition, see the accompanying paper , “ Optimizing Resource Allocation in Hyperscale Datacenters: Scalability, Usability, and Experiences ,” published at OSDI’24. Given a set of objects and a set of bins, how do we assign objects to bins in a way that optimizes specific objectives while meeting certain constraints ? This question arises at all layers of Meta’s infrastructure stack including in Hardware placement: racks (objects) need to be positioned in datacenters (bins) to optimize the spread of racks across electrical fault domains while honoring power and cooling limitations. … ### NVIDIA — From Enablement to Execution, Egypt’s AI Ecosystem Reaches Production Scale - Company: NVIDIA (nvidia.com) - Announced: 2026-09-21T16:00:10+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://blogs.nvidia.com/blog/egypt-africa-ai-ecosystem/ - Record: https://forck.live/items/12725-from-enablement-to-execution-egypt-s-ai-ecosystem-reaches-production-scale - Subject: AI platform NVIDIA hosted an event in Egypt highlighting the nation's growing AI ecosystem, featuring keynotes and panels on AI development and deployment. The event showcased NVIDIA Inception program members across industries including healthcare, robotics, and financial services, while announcing expansions of NVIDIA's Deep Learning Institute learner base in Egypt and new data center investments including a $400 million Hassan Allam data center project and GPU-as-a-service initiatives with Cassava Technologies and Vodafone Egypt. ### Tencent — When Do Larger Batches Help Scale LLM Reinforcement Learning? - Company: Tencent (tencent.com) - Announced: 2026-09-21T16:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://hunyuan.tencent.com/research/100116 - Record: https://forck.live/items/13339-when-do-larger-batches-help-scale-llm-reinforcement-learning - Subject: Hunyuan Tencent researchers investigate how batch size affects the efficiency of reinforcement learning training for large language models, analyzing the trade-off between throughput gains and sample efficiency. The study develops a framework to identify the batch size that minimizes wall-clock time to reach target performance, showing that while moderate batch increases with learning-rate retuning can improve efficiency, excessively large batches incur sample costs that outweigh throughput benefits. ### Microsoft Research — Improving synthesis prediction of small molecules at scale with RetroChimera - Company: Microsoft Research (microsoft.com) - Announced: 2026-09-21T15:30:19+00:00 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.microsoft.com/en-us/research/blog/improving-synthesis-prediction-of-small-molecules-at-scale-with-retrochimera/ - Record: https://forck.live/items/12722-improving-synthesis-prediction-of-small-molecules-at-scale-with-retrochimera - Subject: Research / Phi Microsoft Research published RetroChimera, an open-weight retrosynthesis model that predicts synthesis routes for small molecules by combining two complementary neural models—R-SMILES 2 and NeuralLoc—using a learned ensembling strategy. In blind tests, expert chemists preferred RetroChimera's reaction predictions over preceding models and literature reactions. The implementation and weights are available on GitHub and via Microsoft Foundry. ### GitHub — Grok 4.7 is now available in GitHub Copilot - Company: GitHub (github.com) - Announced: 2026-09-21T14:54:12+00:00 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://github.blog/changelog/2026-09-21-grok-4-7-is-now-available-in-github-copilot - Record: https://forck.live/items/12728-grok-4-7-is-now-available-in-github-copilot - Subject: Copilot xAI's Grok 4.7 reasoning model is rolling out in GitHub Copilot for Copilot Pro, Pro+, Max, Business, and Enterprise SKUs. The model is designed for agentic coding and complex multistep workflows and will be available in Visual Studio Code, Visual Studio, Copilot CLI, GitHub Copilot cloud agent, GitHub Copilot app, JetBrains, Xcode, and Eclipse. ### NVIDIA — AI Security Is an Engineering Problem — How to Solve It at Every Layer of the Agent Stack - Company: NVIDIA (nvidia.com) - Announced: 2026-09-21T14:51:34+00:00 - Category: safety-policy-update - Coverage: not counted - Announcement: no - Group: routine - Source: https://blogs.nvidia.com/blog/ai-security-agent-stack/ - Record: https://forck.live/items/12716-ai-security-is-an-engineering-problem-how-to-solve-it-at-every-layer-of-the - Subject: AI platform NVIDIA outlines an engineering approach to AI agent security, emphasizing that protection requires enforceable controls across the full agent stack—models, harnesses, runtime environments and infrastructure. The company describes security responsibilities at each layer, including identity management, access control, sandboxed execution and audit logging, and introduces OpenShell, an open-source secure runtime that enforces policies outside the agent's reach while governing access to data, network and system resources. ### Hugging Face — Pruning LLMs Like a Physicist: Block Removal as an Ising Optimization Problem - Company: Hugging Face (huggingface.co) - Announced: 2026-09-21T13:44:34+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/MultiverseComputingCAI/pruning-llms-like-a-physicist-block-removal-as-an - Record: https://forck.live/items/12709-pruning-llms-like-a-physicist-block-removal-as-an-ising-optimization-problem - Subject: Platform Hugging Face and Multiverse Computing published a research paper proposing a physics-inspired approach to pruning large language models by reformulating block removal as a constrained binary optimization problem mapped to an Ising glass. The method achieves significant compression gains, reaching 23 percentage points improvement on MMLU over competing block-removal methods at 50% compression of Llama-3.3-70B-Instruct. ### OpenAI — Higgsfield AI ships new video features in a day with GPT-6 Astra - Company: OpenAI (openai.com) - Announced: 2026-09-21T12:00:00+00:00 - Category: not stated - Coverage: 2 outlets - Announcement: no - Group: covered - Source: https://openai.com/index/higgsfield-from-prompt-to-production-with-astra - Record: https://forck.live/items/12744-higgsfield-ai-ships-new-video-features-in-a-day-with-gpt-6-astra - Subject: GPT / ChatGPT / API Higgsfield AI, a video creation platform, used GPT-6 Astra to accelerate feature development and improve ad creation capabilities. The model enabled a single engineer to ship new video exploration features in a day and helped the company generate ad variations and customizations for small businesses. ### Meta — Inside Petal: Building the World’s First Petabit-Class Transoceanic Subsea Cable - Company: Meta (meta.com) - Announced: 2026-09-21T12:00:00+00:00 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://engineering.fb.com/2026/09/21/connectivity/petal-petabit-transoceanic-subsea-cable/ - Record: https://forck.live/items/12685-inside-petal-building-the-world-s-first-petabit-class-transoceanic-subsea-cable - Subject: Llama / infrastructure Meta announced Petal, a subsea cable system expected to enter service in 2029 that will deliver 1 petabit per second (Pbps) of capacity between France and the United States over approximately 7,000 km. Petal will be the first transoceanic subsea cable to deploy multi-core fiber technology at scale, doubling capacity per fiber without proportional increases in power or physical infrastructure. The cable is being built in partnership with NEC and Sumitomo Electric Industries, with support from Orange on the French landing. ### NVIDIA — 5 Companies Using NVIDIA AI for Clean Energy - Company: NVIDIA (nvidia.com) - Announced: 2026-09-21T10:00:35+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://blogs.nvidia.com/blog/clean-energy-nvidia-ai/ - Record: https://forck.live/items/12644-5-companies-using-nvidia-ai-for-clean-energy - Subject: AI platform NVIDIA highlighted five companies using AI to accelerate clean energy projects, including ThinkLabs AI using digital twins to optimize grid operations, Atomic Canyon deploying AI-powered platforms for nuclear plant operations, Redwood Materials integrating recycled EV batteries with AI intelligence for data center power, TerraPower using NVIDIA Omniverse for digital twin software to speed reactor deployment, and Commonwealth Fusion Systems compressing fusion research timelines with NVIDIA tools. ### OpenAI — Building standards for the next phase of AI - Company: OpenAI (openai.com) - Announced: 2026-09-21T10:00:00+00:00 - Category: safety-policy-update - Coverage: 8 outlets - Announcement: no - Group: covered - Source: https://openai.com/index/building-standards-next-phase-ai - Record: https://forck.live/items/12735-building-standards-for-the-next-phase-of-ai - Subject: GPT / ChatGPT / API OpenAI outlines its approach to safely advancing AI through three main goals: building an automated AI researcher while maintaining human oversight, delivering scientific and economic benefits, and empowering individuals with personal AGI. The company emphasizes that international standards for safety and security in frontier AI development are essential to manage risks from recursive self-improvement, prevent fragmentation across nations, enable collective action, and address uneven capacity in AI expertise globally. ### OpenAI — Expanding OpenAI Academy with new learning paths - Company: OpenAI (openai.com) - Announced: 2026-09-21T07:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/expanding-openai-academy-with-new-learning-paths - Record: https://forck.live/items/12736-expanding-openai-academy-with-new-learning-paths - Subject: GPT / ChatGPT / API OpenAI expanded OpenAI Academy with new role-based learning courses for developers, leaders, educators, and college students, joining the existing Apply AI at Work pathway. The courses use real tasks and assessments to help learners build practical AI skills and earn OpenAI Academy badges. ### Perplexity — Learning from Real-World Experience - Company: Perplexity (perplexity.ai) - Announced: 2026-09-21 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.perplexity.ai/hub/blog/learning-from-real-world-experience - Record: https://forck.live/items/13117-learning-from-real-world-experience - Subject: Perplexity How users interact with our products in the real world is a valuable source for model training. The data is abundant, reflects the actual distribution of user tasks, and captures user corrections and tool failures that synthetic environments may miss. A common way to learn from real-world data is rejection sampling fine-tuning: judge each session’s outcome, keep the successful ones, and train the model to imitate them. But a successful outcome does not mean every step was correct, so imitating the whole trajectory risks reinforcing bad intermediate behaviors in addition to good ones. Discarding unsuccessful sessions also loses critical evidence of where the model falls short. We combine rejection sampling fine-tuning with hint-guided self-distillation to learn from both successful and unsuccessful sessions. A hint is a short corrective instruction grounded in information the model already had when it made the mistake. Useful steps from successful sessions remain imitation targets, while grounded hints turn avoidable mistakes into correction targets. … ### xAI — Introducing Grok 4.7 - Company: xAI (x.ai) - Announced: 2026-09-21 - Category: model-update - Coverage: 12 outlets - Announcement: yes - Group: models - Source: https://x.ai/news/grok-4-7 - Record: https://forck.live/items/12720-introducing-grok-4-7 - Subject: Grok xAI released Grok 4.7, an updated model for coding and knowledge work trained with longer reinforcement learning on harder tasks. The model features improved verification capabilities, better context management, and a new safeguard stack, and is available through Cursor, Grok Build, the Grok API, and third-party platforms. ### Perplexity — Learning from Real-World Mistakes - Company: Perplexity (perplexity.ai) - Announced: 2026-09-21 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.perplexity.ai/hub/blog/learning-from-real-world-mistakes - Record: https://forck.live/items/12717-learning-from-real-world-mistakes - Subject: Perplexity Perplexity describes a training method combining rejection sampling fine-tuning with hint-guided self-distillation to learn from real-world user sessions. The approach distinguishes between successful and unsuccessful sessions, using user corrections and tool errors as training signals to reduce tool-call failures by 21.2% relative to an earlier checkpoint. ### Hugging Face — tokenizers v1: encode, decode and scaling, measured - Company: Hugging Face (huggingface.co) - Announced: 2026-09-21 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/tokenizers-v1 - Record: https://forck.live/items/12710-tokenizers-v1-encode-decode-and-scaling-measured - Subject: Platform Hugging Face released tokenizers v1, a major performance update to its tokenization library that aims to be tens of times faster than v0.23 while preserving identical token IDs, API, vocabulary and merge ranks. The update introduces optimizations including SIMD-based bitstream splitting instead of regex, thread-local word caching for repeated pre-tokens, and improvements across normalization, pre-tokenization, model, and post-processing stages to prevent tokenization from becoming a bottleneck in machine learning workflows. ### OpenAI — V7 cuts costs 78% while boosting accuracy with GPT-5.6 Luna - Company: OpenAI (openai.com) - Announced: 2026-09-21 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/v7 - Record: https://forck.live/items/12699-v7-cuts-costs-78-while-boosting-accuracy-with-gpt-5-6-luna - Subject: GPT / ChatGPT / API How V7 gives AI agents institutional memory V7 turns company files into agent context, with GPT‑6 Astra reaching 89% accuracy on its hardest graph-query tests. Results 89% Accuracy for GPT-6 Astra on the hardest queries Results 78% Lower cost per document with GPT-5.6 Luna Results +11.6 pts Higher accuracy with GPT-5.6 Luna Today’s models can reason through complex tasks, but they don’t automatically understand the underlying business context of those tasks. Which fund report is current? How is the same entity named across three systems? That context lives in documents, data rooms, spreadsheets, emails, and internal tools: scattered, unresolved, and invisible to agents. For teams in finance, insurance, and real estate, retrieval accuracy within workflows is non-negotiable. After building a widely used computer vision accessibility app together, Rizzoli and Edwardsson started V7 in 2018 to help companies teach AI systems how their businesses work. V7 Go is an agentic platform to build mission critical workflows, and organize buried context into memory that agents can query and act on. … ### LG AI Research — LG Aimers 9th: Forecasting Pitch Control with AI - Company: LG AI Research (lgresearch.ai) - Announced: 2026-09-21 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://www.lgresearch.ai/news/view?seq=707 - Record: https://forck.live/items/12592-lg-aimers-9th-forecasting-pitch-control-with-ai - Subject: EXAONE LG AI Research hosted the LG Aimers 9th Hackathon, a two-day program in which 96 finalists from a record 3,393 applicants developed AI models to predict pitch control success probability in baseball, with 100 million KRW in funding support. The event included a job fair with four LG affiliates and awarded top teams for predictive precision and field usability.