# forck.live

> AI lab announcements, news reports and podcasts.

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 the current issue, not the archive: every confirmed first-party
announcement dated 9–15 October 2026 — the last 7 days.

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.

The entries are grouped, and the groups are the order to brief in: model
releases and updates, then whatever outlets covered, then the rest of the
announcements, then the how-tos and company notes. Each heading states the
held fields that put an entry under it, and inside a group the entries are
newest first. Nothing is ranked and nothing is scored — a group is a heading,
not a position — and nothing is left out: every announcement in the window is
in exactly one group, the last one included.

Older announcements are in the weekly archive listed at the end of this file,
one file per ISO week. Seven days is not an ISO week, so the newest of those
files overlaps this window rather than sitting entirely before it. This URL
takes no parameters: adding `?since=` or `?days=` changes nothing, and
filtering by date inside one week's entries is not a diff of the archive. For a
scheduled or recurring briefing, read https://forck.live/feed.xml instead and keep
the items newer than your last run.

- Feed: https://forck.live
- Catalogue: https://forck.live/llms.txt
- RSS: https://forck.live/feed.xml

## Model releases and updates

_Every entry this tracker's own analysis filed as `new-model`, `model-update` or `open-weight-release` — a model, a version, or weights. A post that only mentions a model is not one of these._

### TypeSafe AI — Jev: For the builders AND dreamers

- Company: TypeSafe AI (typesafe.ai)
- Announced: not stated by the source (first seen 2026-10-14T23:05:25.851211+00:00)
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://typesafe.ai/blog/jev-1-14
- Record: https://forck.live/items/22458-jev-for-the-builders-and-dreamers
- Subject: Jev / System One

TypeSafe AI released Jev 1.14.0 with small improvements across the board, especially in multi-lingual support. It also introduced experimental features under 'danger zone' flags: composable image support for Jev 1.13.0 and 1.14.0, a new 'box' type returning a single bounding box for Jev 1.14.0, and the ability to use longer context lengths. These features are not recommended for production use.

### Ant Ling — Ling-3.1-flash is now open source: ~560B total parameters, ~25B active per token and up to 1M context.

- Company: Ant Ling (ant-ling.com)
- Announced: 2026-10-14T15:19:49+00:00
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://x.com/AntLingAGI/status/2110390124743970952
- Record: https://forck.live/items/22356-ling-3-1-flash-is-now-open-source-560b-total-parameters-25b-active-per-token

Ant Ling released Ling-3.1-flash as open source, a mixture-of-experts model with approximately 560 billion total parameters, about 25 billion active parameters per token, and support for up to 1 million tokens of context.

### xAI — Introducing Grok Voice 3.0

- Company: xAI (x.ai)
- Announced: 2026-10-14
- Category: new-model
- Coverage: 7 outlets
- Announcement: yes
- Group: models
- Source: https://x.ai/news/grok-voice-3
- Record: https://forck.live/items/22303-introducing-grok-voice-3-0
- Subject: Grok

xAI announced Grok Voice 3.0, a speech-to-speech model that ranks #1 on Artificial Analysis's Speech-to-Speech Index with a score of 86.9 and achieves 97% on Full Duplex Bench. The model has sub-second latency, costs $0.08 per minute in the API (unchanged from its predecessor), and powers millions of daily conversations across Tesla and SpaceX. On ServiceNow's EVA-Bench, it achieved the highest accuracy score at 74% and is the best-performing model when jointly scored across accuracy and customer experience.

### Ai2 — HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost

- Company: Ai2 (allenai.org)
- Announced: 2026-10-14
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://allenai.org/blog/hiro-acev2
- Record: https://forck.live/items/22275-hiro-acev2-simulating-extreme-weather-at-3-km-resolution-for-a-fraction-of-the
- Subject: Olmo / Molmo / Asta

Ai2 released HiRO-ACEv2, an update to its climate emulator system. The new HiROv2 downscaling model predicts multiple high-resolution fields (rain, wind, pressure) simultaneously and improves accuracy for extreme events, matching X-SHiELD's high-resolution statistics to within 3% for events occurring once per million data points. The full system generates global 3 km climate data at roughly 1/300th the cost of X-SHiELD, and a web demo and downloadable models are available.

### Google DeepMind — Gemini 4 Argon: Built to solve your hardest problems

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-10-13T16:01:52+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://deepmind.google/blog/introducing-gemini-4-argon/
- Record: https://forck.live/items/20513-gemini-4-argon-built-to-solve-your-hardest-problems
- Subject: Gemini

Google DeepMind released Gemini 4 Argon, a new frontier model available globally to developers, enterprises, and consumers via the API, AI Studio, Google Antigravity, the Gemini app, and AI Mode in Search. The model excels at complex, multi-step tasks across coding, finance, legal, cybersecurity, and science, and is priced at $2 per million input tokens and $10 per million output tokens with 95% off cached input tokens.

### Meta — Meta’s Generative Ads Model (GEM): Scaling the Central Model and Compute Engine for Frontier Ads AI

- Company: Meta (meta.com)
- Announced: 2026-10-13T16:01:00+00:00
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://engineering.fb.com/2026/10/13/ml-applications/generative-ads-model-gem-scaling-frontier-ads-ai/
- Record: https://forck.live/items/20516-meta-s-generative-ads-model-gem-scaling-the-central-model-and-compute-engine
- Subject: Llama / infrastructure

Meta's 2026 generation of GEM shifts from per-domain optimizations to a single central foundation model that learns across all Meta ads platforms, delivering an 18.9% conversion improvement and 12% ad CTR improvement across surfaces. The model uses a new LLM-style architecture called Kunlun, co-designed with the latest GPU hardware, enabling 2.5x model complexity growth and LLM-like scaling laws. A post-training technique called LoopFM transfers GEM's representations to downstream models at no additional compute cost or latency.

### Google Cloud — Gemini 4 Argon available on Google Cloud, built to help solve your business’ hardest problems

- Company: Google Cloud (cloud.google.com)
- Announced: 2026-10-13T16:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://cloud.google.com/blog/products/ai-machine-learning/gemini-4-argon-available-on-google-cloud/
- Record: https://forck.live/items/20517-gemini-4-argon-available-on-google-cloud-built-to-help-solve-your-business
- Subject: Gemini Enterprise

Google Cloud announced the availability of Gemini 4 Argon, a new frontier model trained for depth and precision across long, multi-step tasks. The model is designed for complex work in software development, finance, legal, and cybersecurity, and has shown improvements in agentic code submissions and code rollback reduction within Google. Customers such as GitLab, Harvey, and ServiceNow have tested the model and reported positive results.

### Liquid AI — LFM2.5-24B-A2B: Built for Fast On-Device Agents

- Company: Liquid AI (liquid.ai)
- Announced: 2026-10-13
- Category: new-model
- Coverage: 1 outlet
- Announcement: yes
- Group: models
- Source: https://www.liquid.ai/blog/lfm2-5-24b-a2b
- Record: https://forck.live/items/20538-lfm2-5-24b-a2b-built-for-fast-on-device-agents
- Subject: LFM / d1 models

Liquid AI released LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters, designed for fast, private on-device agents. The model is competitive with larger MoEs on agentic and tool-use benchmarks, scoring 48.09 on LiveClaw versus 38.19 and 27.69 on AutomationBench versus 8.72. It runs up to 1.7× faster than comparable MoEs on-device, achieving 184 tokens/s on an Apple M5 Max and over 21K output tokens/s on a single NVIDIA B300 GPU.

### Meta — Introducing Muse Wonder and Muse Wonder Full: Frontier Models for Agents

- Company: Meta (meta.com)
- Announced: 2026-10-12T17:08:05+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://about.fb.com/news/2026/10/introducing-muse-wonder-models/
- Record: https://forck.live/items/19957-introducing-muse-wonder-and-muse-wonder-full-frontier-models-for-agents
- Subject: Meta AI

Meta introduced the Muse Wonder series, including Muse Wonder and Muse Wonder Full, its most capable models to date, built on the Muse Spark foundation. The models deliver state-of-the-art performance on personal and professional agentic tasks in expert domains including law, finance, and health, along with frontier coding performance. They feature Active Memory for carrying context across conversations and Active Focus for persisting through difficult tasks, and are available through the Meta Model API and Muse Code.

### Z.ai — Introducing GLM-Legal

- Company: Z.ai (z.ai)
- Announced: 2026-10-12T16:10:47+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://x.com/Zai_org/status/2109678174468534511
- Record: https://forck.live/items/19959-introducing-glm-legal

Z.ai introduced GLM-Legal, a model built on GLM-5.3-Flash that matches frontier models on mainstream legal benchmarks. It is designed for real legal workflows including contract review, evidence analysis, and compliance.

### Meta — Muse Wonder: A Frontier Model For Personal & Professional AI Agents

- Company: Meta (meta.com)
- Announced: 2026-10-12
- Category: new-model
- Coverage: 29 outlets
- Announcement: yes
- Group: models
- Source: https://research.meta.ai/blog/muse-wonder
- Record: https://forck.live/items/19953-muse-wonder-a-frontier-model-for-personal-professional-ai-agents
- Subject: Muse

Meta announced the Muse Wonder series, including Muse Wonder and Muse Wonder Full, as its most capable models for personal and professional agentic tasks. The models feature Active Memory and Active Focus capabilities, and are available through the Meta Model API and Muse Code, with Muse Wonder Full available for Muse Code subscribers. The series achieves state-of-the-art results on benchmarks for agents, memory, and coding.

### Cloudflare — Introducing Clef-omni with full multimodality, plus a faster Clef and a cheaper Clef-flash

- Company: Cloudflare (cloudflare.com)
- Announced: 2026-10-09T18:27:34+00:00
- Category: new-model
- Coverage: 1 outlet
- Announcement: yes
- Group: models
- Source: https://blog.cloudflare.com/clef-faster-cheaper-multimodal/
- Record: https://forck.live/items/18828-introducing-clef-omni-with-full-multimodality-plus-a-faster-clef-and-a-cheaper
- Subject: Cloudflare AI

Following last week’s release of Clef and Clef-flash , Cloudflare’s open-weight decision models, we decided to bring forth more gifts. Today, we’re releasing Clef-omni, which takes in audio and video input alongside text and image. We also cut the price of Clef-flash so it is now cheaper than Jev, and we made Clef faster. Although the model game is still early for Cloudflare, innovation and iteration is in our DNA, and we apply these principles to everything we do. In fact, the story of Clef came together over the course of less than a week. We decided we wanted to do something in the decision model space on a Friday evening, trained the model over the weekend, and launched it on Thursday. Even with such a short timeline, we were able to ship performant, high-quality, open-weight models for the community — imagine what more we can do in the future. For today, we’re excited to keep up the momentum with new additions and improvements to our Clef family of models. This is just the beginning, and we’ll continue to get better, faster, cheaper, and more innovative. …

### Perplexity — Decisions API: pplx-decider-v1.1-27b replaces v1

- Company: Perplexity (perplexity.ai)
- Announced: 2026-10-09T01:20:07+00:00
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://docs.perplexity.ai/docs/resources/changelog#october-2026
- Record: https://forck.live/items/17219-decisions-api-pplx-decider-v1-1-27b-replaces-v1
- Subject: Sonar

Perplexity replaced pplx-decider-v1-27b with pplx-decider-v1.1-27b as the Decisions API model, and cut the input price to $0.02 per million tokens from $0.04. Requests naming the older model continue to work and bill at the same $0.02 rate.

## Widely covered

_Announcements outlets wrote about, whatever their category — each carries the `Coverage` line that counts them. An acquisition or a safety post reaches this band the same way a model does._

### Baseten — Deploy Runway’s video models on Baseten

- Company: Baseten (baseten.co)
- Announced: 2026-10-14T12:01:44+00:00
- Category: availability-change
- Coverage: 1 outlet
- Announcement: no
- Group: covered
- Source: https://www.baseten.co/blog/deploy-runway-video-models-dedicated-inference
- Record: https://forck.live/items/22193-deploy-runway-s-video-models-on-baseten
- Subject: Baseten

Basten partnered with Runway to offer Runway's Gen-4.5 and Aleph 2.0 video models for dedicated deployment on Baseten's single-tenant infrastructure. Customers can license model weights from Runway and deploy them in isolated environments with dedicated GPU resources, fine-tuning support, and a 99.99% uptime SLA.

### NVIDIA — European Cities Deploy NVIDIA-Powered AI Agents to Run Smarter Urban Operations

- Company: NVIDIA (nvidia.com)
- Announced: 2026-10-14T06:00:28+00:00
- Category: not stated
- Coverage: 1 outlet
- Announcement: no
- Group: covered
- Source: https://blogs.nvidia.com/blog/european-cities-agents-urban-operations/
- Record: https://forck.live/items/21923-european-cities-deploy-nvidia-powered-ai-agents-to-run-smarter-urban-operations
- Subject: AI platform

NVIDIA announced that cities and partners across Europe are deploying AI agents and digital twins built on NVIDIA accelerated computing, Cosmos, Nemotron, Omniverse, and the Metropolis VSS blueprint. Projects in France, Italy, Belgium, the UK, Sweden, and elsewhere aim to improve urban operations such as energy management, mobility, safety, and document processing. The Department of Côte-d’Or in France reported a 10% reduction in total operating costs and a reduction in document search time from weeks to minutes across over 500 public buildings.

### Anthropic — Preparing for the 2026 US midterms

- Company: Anthropic (anthropic.com)
- Announced: 2026-10-14
- Category: safety-policy-update
- Coverage: 1 outlet
- Announcement: no
- Group: covered
- Source: https://www.anthropic.com/news/2026-us-midterms
- Record: https://forck.live/items/22374-preparing-for-the-2026-us-midterms
- Subject: Claude

Anthropic updated its election safeguards ahead of the 2026 US midterms, including integrating official voting information from Democracy Works into Claude.ai responses and showing election night results. The company reported that Claude Opus 5.5, Sonnet 5.5, and Haiku 5.5 scored 99% on its political even-handedness evaluation, and that independent reviews identified areas for improvement such as unsolicited caveats and uneven hedging. Anthropic also shared usage data showing that about 5.75% of US Claude.ai conversations touch on politics, with 0.51% specifically about elections or voting.

### Anthropic — Agent observability on the Claude Platform

- Company: Anthropic (anthropic.com)
- Announced: 2026-10-14
- Category: capability-change
- Coverage: 1 outlet
- Announcement: yes
- Group: covered
- Source: https://claude.com/resources/articles/agent-observability-on-the-claude-platform
- Record: https://forck.live/items/22291-agent-observability-on-the-claude-platform
- Subject: Claude

Anthropic announced that Claude Managed Agents, the Messages API, and Claude SDKs can now export OpenTelemetry traces to any OTel-compatible tool, allowing users to trace model calls, tool use, and MCP server calls. The feature is in public beta and works with tools such as Grafana Cloud, Honeycomb, Sentry, Splunk, Arize AI, Braintrust, Datadog, Langfuse, LangSmith, New Relic, and PostHog. Managed Agents and the Messages API require a destination configured in the Claude Console, while Claude SDKs export traces automatically.

### Anthropic — Anthropic opens Singapore office and signs MOU with the Government of Singapore

- Company: Anthropic (anthropic.com)
- Announced: 2026-10-14
- Category: partnership-acquisition
- Coverage: 2 outlets
- Announcement: yes
- Group: covered
- Source: https://www.anthropic.com/news/singapore-office-opening
- Record: https://forck.live/items/22000-anthropic-opens-singapore-office-and-signs-mou-with-the-government-of-singapore
- Subject: Claude

Anthropic opened a new office in Singapore, its fifth in the Asia-Pacific region, and signed a Memorandum of Understanding with the Singapore Ministry of Digital Development and Information. The MOU outlines cooperation in AI safety, security, model evaluation, and responsible adoption across public and private sectors.

### Google Antigravity — Gemini 4 Argon in Google Antigravity

- Company: Google Antigravity (antigravity.google)
- Announced: 2026-10-13T08:00:00+00:00
- Category: availability-change
- Coverage: 29 outlets
- Announcement: yes
- Group: covered
- Source: https://antigravity.google/blog/gemini-4-argon-in-google-antigravity
- Record: https://forck.live/items/20581-gemini-4-argon-in-google-antigravity
- Subject: Antigravity

Google has made Gemini 4 Argon available inside Google Antigravity for Google AI Ultra subscribers, accessible via the desktop app and CLI. For businesses, it is also available via the API in Gemini platform and within Antigravity on eligible Gemini Enterprise subscriptions. The post describes internal usage at Alphabet, including a 35% increase in agentic code and a 30% decrease in post-submit rollbacks over four weeks.

### Amazon — Accessing OpenAI GPT models on Amazon Bedrock with global cross-Region inference in Thailand, Malaysia, Singapore, Indonesia, and Taiwan

- Company: Amazon (amazon.com)
- Announced: 2026-10-12T17:52:22+00:00
- Category: availability-change
- Coverage: 1 outlet
- Announcement: yes
- Group: covered
- Source: https://aws.amazon.com/blogs/machine-learning/accessing-openai-gpt-models-on-amazon-bedrock-with-global-cross-region-inference-in-thailand-malaysia-singapore-indonesia-and-taiwan/
- Record: https://forck.live/items/19991-accessing-openai-gpt-models-on-amazon-bedrock-with-global-cross-region
- Subject: Bedrock / Nova

Amazon Bedrock made OpenAI GPT-6 Sol, GPT-6 Luna, and GPT-6 Astra available through global cross-Region inference in Thailand, Malaysia, Singapore, Indonesia, and Taiwan. GPT-6 Sol and GPT-6 Luna are accessible from all five AWS Regions, while GPT-6 Astra is available from Asia Pacific (Singapore). The post explains how to configure calls, set permissions, and monitor inference from the source Region.

### Microsoft — Deep dive into Azure Maia 300 architecture

- Company: Microsoft (microsoft.com)
- Announced: 2026-10-12T13:00:00+00:00
- Category: infrastructure-release
- Coverage: 14 outlets
- Announcement: yes
- Group: covered
- Source: https://azure.microsoft.com/en-us/blog/deep-dive-into-azure-maia-300-architecture/
- Record: https://forck.live/items/19945-deep-dive-into-azure-maia-300-architecture
- Subject: Azure AI

Microsoft published a deep dive into the architecture of its third-generation AI accelerator, Maia 300, which is designed for cloud-scale agentic inference. The system-on-chip uses a multi-chiplet design on TSMC N2P, introduces enhanced MXFP4 and sub-3-bit storage formats, and provides 2.5 times the memory bandwidth and 33% more capacity than Maia 200. The rack doubles to 96 accelerators and the scale-up domain grows to a superpod of 12,288 accelerators.

### Microsoft — Introducing Microsoft-Decision-1, our model for fast decision-making

- Company: Microsoft (microsoft.com)
- Announced: 2026-10-10T01:35:51+00:00
- Category: not stated
- Coverage: 6 outlets
- Announcement: yes
- Group: covered
- Source: https://commandline.microsoft.com/microsoft-decision-1-model-foundry/
- Record: https://forck.live/items/19120-introducing-microsoft-decision-1-our-model-for-fast-decision-making
- Subject: Foundry / Responsible AI

Decision models are quickly emerging as an important new category in AI. Unlike LLMs, which are designed to generate text or reason through complex problems, decision models are purpose-built to deliver structured outputs that software can immediately act on.

### OpenAI — Sophos cuts threat investigation time by 96% with OpenAI Daybreak

- Company: OpenAI (openai.com)
- Announced: 2026-10-09T07:00:00+00:00
- Category: not stated
- Coverage: 1 outlet
- Announcement: no
- Group: covered
- Source: https://openai.com/index/sophos
- Record: https://forck.live/items/17394-sophos-cuts-threat-investigation-time-by-96-with-openai-daybreak
- Subject: GPT / ChatGPT / API

Sophos integrated OpenAI Daybreak into its Sophos Fusion cyber defense system, enabling AI agents to investigate and respond to security cases. The average response time for cases using these agents dropped from 38 minutes to 89 seconds, and 52% of Managed Detection and Response (MDR) cases are now resolved end-to-end by AI within boundaries set by Sophos analysts. The system processes trillions of daily sensor events into 1,000–2,000 cases across nine security operations centres.

### TypeSafe AI — TypeSafe A raises Series AI

- Company: TypeSafe AI (typesafe.ai)
- Announced: 2026-10-09
- Category: funding-company-news
- Coverage: 3 outlets
- Announcement: yes
- Group: covered
- Source: https://typesafe.ai/blog/series-ai
- Record: https://forck.live/items/18769-typesafe-a-raises-series-ai
- Subject: Jev / System One

TypeSafe AI announced it raised an $870 million Series A led by Andreessen Horowitz, with participation from Sequoia Capital, DCVC, and angel investors. The company said a third of the Fortune 500 use its Jev product and that it has saved customers millions of dollars in production. It plans to use the funding to expand machine-native models, add enterprise features, and build out infrastructure for building smart software.

## Other announcements

_First-party announcements no outlet this tracker counts has covered yet, including posts nothing has analysed — no category is a queue, not a verdict._

### GitHub — HydraFusion updates and new Max option

- Company: GitHub (github.com)
- Announced: 2026-10-14T20:01:35+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://github.blog/changelog/2026-10-14-hydrafusion-updates-and-new-max-option
- Record: https://forck.live/items/22404-hydrafusion-updates-and-new-max-option
- Subject: Copilot

GitHub updated HydraFusion with an expanded model pool including Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, improved orchestration workflows (tailored reasoning effort, cascade with judging step, critique using tools and parallel critics), and introduced HydraFusion Max, a tier tuned for frontier capability on more demanding tasks. Both remain in research preview and are available across VS Code, the GitHub Copilot app, and GitHub Copilot CLI on Copilot Pro, Pro+, Business, and Enterprise plans with experimental or preview features enabled.

### Databricks — Power real-time applications and agents with Document Intelligence

- Company: Databricks (databricks.com)
- Announced: 2026-10-14T20:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.databricks.com/blog/power-real-time-applications-and-agents-document-intelligence
- Record: https://forck.live/items/22383-power-real-time-applications-and-agents-with-document-intelligence
- Subject: Mosaic AI

Databricks announced the general availability of Document Intelligence REST APIs for parsing, classification, and extraction, and a beta launch of the decisions REST API for document routing. These synchronous APIs enable real-time document processing within applications and agent workflows, allowing users to extract, classify, and act on document content as it arrives.

### Google Research — Putting Earth AI to work: Geospatial agents for our planet's biggest challenges

- Company: Google Research (research.google)
- Announced: 2026-10-14T16:14:38+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/putting-earth-ai-to-work-geospatial-agents-for-our-planets-biggest-challenges/
- Record: https://forck.live/items/22243-putting-earth-ai-to-work-geospatial-agents-for-our-planet-s-biggest-challenges
- Subject: Research

Google Research published a technical report on a Geospatial Reasoning agent that combines Earth AI foundation models, spatial tools, and upfront planning to perform complex spatial analysis and planetary prediction. The agent connects Gemini to Earth AI models including WeatherNext, Flood Forecasting, the Population Dynamics Foundation Model, AlphaEarth Foundations, and Remote Sensing Foundations, alongside Google Maps Platform and Earth Engine data. Google is expanding experimental access via Ask Google Earth and an Agent-to-Agent (A2A) API service.

### Google Research — Analyzing technical documents with Deep Review

- Company: Google Research (research.google)
- Announced: 2026-10-14T16:14:37+00:00
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/analyzing-technical-documents-with-deep-review/
- Record: https://forck.live/items/22244-analyzing-technical-documents-with-deep-review
- Subject: Research

Google Research publicly launched Deep Review, an agentic AI system for rigorous feedback and technical verification on complex documents. Initially designed for computer science academic paper analysis, it now supports documents across domains such as legal and finance. Every user receives two complimentary document reviews per day, with higher usage limits available via a Gemini API key.

### Google Research — Research@ 2026: 10 ways we're amplifying human ingenuity

- Company: Google Research (research.google)
- Announced: 2026-10-14T16:14:34+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/research-2026-10-ways-were-amplifying-human-ingenuity/
- Record: https://forck.live/items/22245-research-2026-10-ways-we-re-amplifying-human-ingenuity
- Subject: Research

Google Research announced expanded early access to Google Earth AI with new experimental capabilities including geospatial analysis, risk assessment, and an agent-to-agent API. They also introduced Deep Review for public use, a multi-agent orchestrator for scientific research, and a next-gen Edge AI chip co-designed with Synaptics.

### Google Cloud — Predictable AI spend in BigQuery: Introducing cost management for generative AI functions

- Company: Google Cloud (cloud.google.com)
- Announced: 2026-10-14T16:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cloud.google.com/blog/products/data-analytics/using-bigquerys-generative-ai-cost-management-features/
- Record: https://forck.live/items/22241-predictable-ai-spend-in-bigquery-introducing-cost-management-for-generative-ai
- Subject: Gemini Enterprise

Google Cloud announced general availability of token-based cost management for BigQuery generative AI functions. Administrators can set four daily quotas (input tokens per day, input tokens per day per user, output tokens per day, output tokens per day per user) at the project and per-user level, enforced through the IAM quota system. The quotas apply to functions such as AI.GENERATE, AI.CLASSIFY, and AI.SCORE, but not to embedding functions.

### Lovable — From building your app to growing it: Lovable now helps you find customers

- Company: Lovable (lovable.dev)
- Announced: 2026-10-14T12:50:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/introducing-growth-tools
- Record: https://forck.live/items/22157-from-building-your-app-to-growing-it-lovable-now-helps-you-find-customers
- Subject: Lovable / AI app builder

Lovable introduced growth tools that let users set a goal, get a plan, and launch ad campaigns on Google, YouTube, Facebook, and Instagram directly from the chat interface. The tools also include SEO improvements via Semrush and the ability to create landing pages, ad copy, and social media posts. No ad campaigns go live without user approval, and billing remains within the user's Google Ads and Meta accounts.

### OpenAI — Amp puts AI agents to work beyond code with OpenAI models

- Company: OpenAI (openai.com)
- Announced: 2026-10-14T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/amp-beyond-coding
- Record: https://forck.live/items/22273-amp-puts-ai-agents-to-work-beyond-code-with-openai-models
- Subject: GPT / ChatGPT / API

See how Amp pushes the frontier of what agents can do with GPT-6 Astra, Ultrafast, GPT-Live-1, and Sign in with ChatGPT.

### Reason Machines — We just launched on Product Hunt, would love the support!

- Company: Reason Machines (reasonmachines.com)
- Announced: 2026-10-14T09:17:04+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.com/reasonmachines/status/2110298833968353585
- Record: https://forck.live/items/22061-we-just-launched-on-product-hunt-would-love-the-support

making coding cheaper, faster and better for you! https://t.co/EGB4PJmEzu

### Apple — Code-Switching Detection via Language-Selective Dimensions in Speech Foundation Models

- Company: Apple (apple.com)
- Announced: 2026-10-14
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://machinelearning.apple.com/research/code-switching-detection
- Record: https://forck.live/items/22247-code-switching-detection-via-language-selective-dimensions-in-speech
- Subject: Machine Learning Research

Apple researchers propose a lightweight method to detect code-switching directly from speech using frame-level language information in pretrained speech encoders. The Whisper-small encoder achieves a mean F1 of 0.798 across 14 language pairs in CS-FLEURS, outperforming CNN and HuBERT baselines.

### Apple — The Design Space of Tri-Modal Masked Diffusion Models

- Company: Apple (apple.com)
- Announced: 2026-10-14
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://machinelearning.apple.com/research/tri-modal-masked-diffusion
- Record: https://forck.live/items/22236-the-design-space-of-tri-modal-masked-diffusion-models
- Subject: Machine Learning Research

Apple researchers introduced the first tri-modal Masked Diffusion Model (MDM) pretrained from scratch on text, image-text, and audio-text data. They systematically analyzed multimodal scaling laws, modality mixing ratios, noise schedules, and batch-size effects, proposing a novel SDE-based reparameterization that decouples physical and logical batch sizes. A preliminary 3B-parameter model trained on 6.4T tokens achieved strong results in text generation, text-to-image, and text-to-speech tasks.

### Factory AI — Introducing OpenCompute

- Company: Factory AI (factory.com)
- Announced: 2026-10-14
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/opencompute
- Record: https://forck.live/items/22230-introducing-opencompute
- Subject: Droid

Factory announced OpenCompute, a self-hosted orchestrator for agent sandboxes that runs on customer-controlled infrastructure, now in private preview. It provides on-demand provisioning and lifecycle management for fleets of computers, integrated with Factory's existing Environments, Automations, and Remote Control features. OpenCompute is designed to let customers run autonomous software development agents on their own choice of runtime without handing over infrastructure control.

### GitHub — Gemini 4 Argon in GitHub Copilot

- Company: GitHub (github.com)
- Announced: 2026-10-13T18:39:56+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://github.blog/changelog/2026-10-13-gemini-4-argon-in-github-copilot
- Record: https://forck.live/items/20629-gemini-4-argon-in-github-copilot
- Subject: Copilot

GitHub made Gemini 4 Argon generally available in Copilot for Pro+, Max, Business, and Enterprise users, with gradual rollout across multiple IDEs and tools.

### Runway — Introducing Runway for web design.

- Company: Runway (runway.com)
- Announced: 2026-10-13T17:34:15+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.com/runwayml/status/2110061567383577037
- Record: https://forck.live/items/20575-introducing-runway-for-web-design

Runway announced a new integration with Replit that allows users to generate images, video, and other assets directly within their Replit projects, from initial concept to a completed site.

### Amazon — Manage private pricing offers for Amazon Bedrock with managed entitlements

- Company: Amazon (amazon.com)
- Announced: 2026-10-13T17:17:25+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://aws.amazon.com/blogs/machine-learning/manage-private-pricing-offers-for-amazon-bedrock-with-managed-entitlements/
- Record: https://forck.live/items/20565-manage-private-pricing-offers-for-amazon-bedrock-with-managed-entitlements
- Subject: Bedrock / Nova

Amazon published a sample repository with two automation paths for activating AWS Marketplace Private Pricing Offer (MPPO) entitlements for third-party models on Amazon Bedrock across multiple AWS accounts. The lightweight script provides one-time activation without infrastructure, while the CDK-based path deploys ongoing event-driven automation that verifies sellers and creates distributed grants. The solution aims to reduce the operational friction of manually activating grants in AWS License Manager.

### Microsoft — Infrastructure for the frontier era: Scaling intelligence through open industry collaboration

- Company: Microsoft (microsoft.com)
- Announced: 2026-10-13T16:15:00+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://azure.microsoft.com/en-us/blog/infrastructure-for-the-frontier-era-scaling-intelligence-through-open-industry-collaboration/
- Record: https://forck.live/items/22170-infrastructure-for-the-frontier-era-scaling-intelligence-through-open-industry
- Subject: Azure AI

Microsoft announced its next-generation AI accelerator, Azure Maia 300, designed for inference workloads and expected to deliver 40% more tokens per watt than leading GPU-based systems. The company also introduced Cobalt 200, an agent-native CPU with over 50% performance improvement over Cobalt 100 and 33% lower agent call latency. Additionally, Microsoft detailed networking innovations including Multipath Reliable Connection (MRC) and the Optical Compute Interconnect Multi-Source Agreement (OCI MSA), as well as power infrastructure advances through the Open Compute Project.

### Microsoft — Copilot to autopilot: Why no single safeguard is enough

- Company: Microsoft (microsoft.com)
- Announced: 2026-10-13T16:03:25+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://commandline.microsoft.com/ai-agent-copilot-autopilot-layered-model-production-network-incidents-on-call-engineers/
- Record: https://forck.live/items/20535-copilot-to-autopilot-why-no-single-safeguard-is-enough
- Subject: Foundry / Responsible AI

Microsoft describes a layered safety architecture for AI agents that can take production actions, arguing that no single safeguard—model, prompt, or guardrail—is sufficient. The post defines an agent as deterministic software wrapped around a non-deterministic model, and presents four boundaries (identity, tool, mutation, time) that together constrain agent behavior. It emphasizes that authorization, not intelligence, is the hard part of safe autonomy.

### GitHub — Enterprise managed settings enforcement updates

- Company: GitHub (github.com)
- Announced: 2026-10-13T15:20:10+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://github.blog/changelog/2026-10-13-enterprise-managed-settings-enforcement-updates
- Record: https://forck.live/items/20536-enterprise-managed-settings-enforcement-updates
- Subject: Copilot

GitHub added a forceRemoteSettingsRefresh setting that lets enterprise administrators require Copilot clients to fetch current server-managed settings at startup, failing closed if the policy cannot be confirmed. The company also added minimum client version enforcement to the sandbox key, forcing users on older Copilot clients to update.

### Reason Machines — Introducing Reason:

- Company: Reason Machines (reasonmachines.com)
- Announced: 2026-10-13T13:33:49+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.com/reasonmachines/status/2110001062216376724
- Record: https://forck.live/items/20472-introducing-reason

A state-of-the-art harness on cost efficiency for free Beating Codex+Astra on Terminal-Bench 4.0 while being 22% cheaper.

### Baseten — Delivering real-time text-to-speech with Gradium

- Company: Baseten (baseten.co)
- Announced: 2026-10-13T13:32:17+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.baseten.co/blog/delivering-fast-real-time-text-to-speech-with-gradium
- Record: https://forck.live/items/21693-delivering-real-time-text-to-speech-with-gradium
- Subject: Baseten

Achieve natural-sounding conversational experiences using Gradium TTS on Baseten.

### NVIDIA — All Systems Go: Global PC App Ecosystem Ready for NVIDIA RTX Spark

- Company: NVIDIA (nvidia.com)
- Announced: 2026-10-13T12:00:49+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://blogs.nvidia.com/blog/local-ai-rtx-spark-laptops-software-updates/
- Record: https://forck.live/items/20412-all-systems-go-global-pc-app-ecosystem-ready-for-nvidia-rtx-spark
- Subject: AI platform

NVIDIA RTX Spark is a new beginning for the PC — a reinvention combining the capabilities of a gaming PC, a creative studio and a personal AI supercomputer into a single superchip.

### Manus — Introducing Manus for Business: Become an AI-native company today

- Company: Manus (manus.im)
- Announced: 2026-10-13
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://manus.im/blog/manus-team-plan
- Record: https://forck.live/items/22062-introducing-manus-for-business-become-an-ai-native-company-today
- Subject: Manus

In the last 6 months, we have seen a fundamental change in how our team works at Manus. Our product managers use Manus to consolidate and prioritize product feedback the team posts in Slack.

### Supabase — Ship changes to your Supabase backend with Grok Bot

- Company: Supabase (supabase.com)
- Announced: 2026-10-13
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/ship-changes-to-your-supabase-backend-with-grok-bot
- Record: https://forck.live/items/21264-ship-changes-to-your-supabase-backend-with-grok-bot
- Subject: Supabase

Supabase announced it is joining the SpaceXAI Partner Network. The integration allows users to request changes via Grok Bot in Cursor, which then builds and runs those changes against the user's Supabase backend through the Supabase MCP server, including querying databases, inspecting schemas, reading logs, and deploying Edge Functions.

### Supabase — Declarative Schemas 2.0 with pg-delta

- Company: Supabase (supabase.com)
- Announced: 2026-10-13
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/declarative-schemas-2-0-pg-delta
- Record: https://forck.live/items/21263-declarative-schemas-2-0-with-pg-delta
- Subject: Supabase

Supabase shipped Declarative Schemas 2.0, replacing the migra diff engine with its own open-source pg-delta tool. pg-delta adds support for RLS policy changes, grants, view settings, partitions, domains, and extension-managed objects, and uses a dependency graph to order output. The new engine is opt-in for existing projects and default for new ones.

### Cursor — Origin App Marketplace

- Company: Cursor (cursor.com)
- Announced: 2026-10-13
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/origin-app-marketplace
- Record: https://forck.live/items/20624-origin-app-marketplace
- Subject: Cursor

Cursor's Origin platform now integrates with over 20 apps from partners including Cloudflare, Linear, Sentry, and Socket Security for deployments, CI, code review, security, monitoring, and project management. The marketplace is open to public submissions, allowing any team building developer tools to publish an app for Origin.

### Anthropic — Uber's CISO on building an insider threat program for AI

- Company: Anthropic (anthropic.com)
- Announced: 2026-10-13
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://claude.com/resources/articles/uber-s-ciso-on-building-an-insider-threat-program-for-ai
- Record: https://forck.live/items/20619-uber-s-ciso-on-building-an-insider-threat-program-for-ai
- Subject: Claude

Uber runs one of the world's largest real-time marketplaces, with over 40 million trips and delivery orders a day.

### Amp — Take Puck with You

- Company: Amp (ampcode.com)
- Announced: 2026-10-13
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/take-puck-with-you
- Record: https://forck.live/items/20580-take-puck-with-you
- Subject: Amp

Amp announced that its AI agent Puck can now be accessed from ChatGPT, Claude, and other MCP-compatible clients. The Puck MCP integration provides two tools: one for interacting with Puck in Amp and another for viewing running orbs with portal links. Users connect by adding the MCP server URL in their client and signing in with their Amp account.

### Apple — Mix, Don't Tune: Bilingual Pre-Training Outperforms Hyperparameter Search in Data-Constrained Settings

- Company: Apple (apple.com)
- Announced: 2026-10-13
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://machinelearning.apple.com/research/mix-dont-tune
- Record: https://forck.live/items/20485-mix-don-t-tune-bilingual-pre-training-outperforms-hyperparameter-search-in
- Subject: Machine Learning Research

Apple researchers compared two strategies for improving language model pre-training when target-language data is scarce: aggressive hyperparameter tuning (e.g., high weight decay) versus mixing in data from a high-resource auxiliary language. Across roughly 1000 runs at scales from 150M to 1.43B parameters, using Arabic as the low-resource target and English as the auxiliary, they found that mixing consistently outperformed hyperparameter tuning on both validation loss and downstream accuracy, with the gap widening at larger model sizes. The paper recommends mixing in a high-resource language, prioritizing the mixing ratio over hyperparameter tuning, and transferring hyperparameters from a small proxy model via μP.

### Manus — Manus Partners with OPPO to Bring Agentic AI Experiences to Smartphones

- Company: Manus (manus.im)
- Announced: 2026-10-13
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://manus.im/blog/manus-partners-oppo
- Record: https://forck.live/items/20320-manus-partners-with-oppo-to-bring-agentic-ai-experiences-to-smartphones
- Subject: Manus

Manus announced a partnership with OPPO to integrate its agentic AI capabilities into OPPO’s AI Mind Pilot, enabling smartphones to turn complex multi-step tasks into finished results. The integration debuts on the upcoming OPPO Find X10 Series with ColorOS 17 through a new Agent Mode. AI Mind Pilot uses on-device processing for tasks like organizing files and photos, and requires user confirmation before sharing content externally.

### LG AI Research — [ICML2026] Solving the Problem of Data Scarcity Using Riemannian Geometry

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-10-13
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/blog/view?seq=715
- Record: https://forck.live/items/20274-icml2026-solving-the-problem-of-data-scarcity-using-riemannian-geometry
- Subject: EXAONE

LG AI Research presented GEAR and Sim-GEAR at ICML 2026, two research projects that address data scarcity in materials science by using Riemannian geometry to improve transfer learning between physical property prediction models. The method calculates curvature in the latent spaces of a data-rich source model and a data-poor target model, training them so their geometric structures become similar.

### GitHub — Client telemetry user attribution and configurable capture

- Company: GitHub (github.com)
- Announced: 2026-10-12T22:38:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://github.blog/changelog/2026-10-12-client-telemetry-user-attribution-and-configurable-capture
- Record: https://forck.live/items/20149-client-telemetry-user-attribution-and-configurable-capture
- Subject: Copilot

GitHub added a telemetry setting that lets enterprise administrators attribute Copilot activity to a user without collecting prompts, responses, or tool content. The telemetry.capture.identity setting adds user.name, process.user.name, and host.name attributes to OpenTelemetry data, and is off by default. Administrators can also independently control five additional capture classes — prompts, responses, toolArguments, toolOutput, and policyDetail — each off by default, with explicit managed values taking precedence over local settings.

### Meta — OCP Summit 2026: Powering AI with Open Innovations in Scale-Up, Scale-Out, and Scale-Across Networks

- Company: Meta (meta.com)
- Announced: 2026-10-12T21:19:18+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://engineering.fb.com/2026/10/12/open-source/ocp-summit-2026-powering-ai-with-open-innovations-in-scale-up-scale-out-and-scale-across-networks/
- Record: https://forck.live/items/20032-ocp-summit-2026-powering-ai-with-open-innovations-in-scale-up-scale-out-and
- Subject: Llama / infrastructure

At the 2026 OCP Global Summit, Meta announced several open networking infrastructure innovations: IcePack, its first liquid-cooled network rack; IceCube, a 102.4 Tbps switch blade delivering 128×800G non-blocking capacity; Flattened Planar Fabric (FPF), a multiplane backend scale-out fabric; MetaRoCE, an open-source RDMA transport protocol; and the Open Switch Platform Architecture (OSPA) initiative co-led with Microsoft and Alibaba. Meta also shared details on petabit-scale backend aggregation for multi-site clusters and Ethernet-based scale-up networking for its MTIA 300 system.

### Meta — Shipping Media Over QUIC at Meta Scale for Instagram Live Ingest

- Company: Meta (meta.com)
- Announced: 2026-10-12T16:01:32+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://engineering.fb.com/2026/10/12/video-engineering/shipping-media-over-quic-at-meta-scale-for-instagram-live-ingest/
- Record: https://forck.live/items/19915-shipping-media-over-quic-at-meta-scale-for-instagram-live-ingest
- Subject: Llama / infrastructure

Meta has begun deploying Media over QUIC (MoQ), an emerging open protocol for real-time video streaming, for Instagram Live ingest on iOS and Android. MoQ is designed to simplify Meta's protocol stack for live streaming and improve latency and resilience. The deployment replaces Meta's in-house RUSH protocol and is the first large-scale use of MoQ at Meta.

### Meta — Turning Opportunistic AI Inference Capacity Into Better Ads Recommendations

- Company: Meta (meta.com)
- Announced: 2026-10-12T16:00:50+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://engineering.fb.com/2026/10/12/ml-applications/turning-opportunistic-ai-inference-capacity-into-better-ads-recommendations/
- Record: https://forck.live/items/19916-turning-opportunistic-ai-inference-capacity-into-better-ads-recommendations
- Subject: Llama / infrastructure

At Meta, we’ve built a system that enhances ads recommendations by leveraging idle inference capacity in our fleet. Two elasticity levers in the system allow ads recommendation work to expand across opportunistic capacity, without compromising baseline performance. This post describes how we designed and now operate the system at Meta’s scale and how it drives significant advertiser value and capacity efficiency. Inference fleets face a capacity paradox: They can be globally constrained while meaningful pockets of capacity remain locally idle. At Meta, we built a system that aggregates these pockets into a shared opportunistic capacity pool and turns them into better ads recommendations. By putting otherwise idle capacity to work, the system helps people see more relevant ads and improves advertiser outcomes without requiring additional capacity orders. Using this opportunistic capacity safely is difficult. It’s distributed across team-owned capacity allocations, regions, and accelerator types, and may be reclaimed with little notice. …

### SenseTime — 调度效率提升80%！商汤大装置助力上海量子城市首批智能体落地

- Company: SenseTime (sensetime.com)
- Announced: 2026-10-12T16:00:18+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.sensetime.com/cn/news/sensecore-shanghai-quantum-city-ai-agents-20261013-1702
- Record: https://forck.live/items/21984-80
- Title in English (translated by forck.live from zh): Dispatch efficiency improved by 80%! SenseTime's large facility helps Shanghai quantum city's first AI agents land
- Subject: SenseNova / 商汤

SenseTime's SenseCore team delivered three AI agents for Shanghai's quantum city initiative, based on the upgraded 'Yunyu Xingkong' large model. The 'Yuzhi Heluo' flood control agent achieved a 65% improvement in spatial intelligence computing, 70% in dynamic simulation, and 80% in intelligent dispatch efficiency. The agents integrate spatial reasoning and task execution capabilities into urban governance workflows.

### Microsoft Research — Simulating student learning for better AI tutoring

- Company: Microsoft Research (microsoft.com)
- Announced: 2026-10-12T16:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.microsoft.com/en-us/research/blog/simulating-student-learning-for-better-ai-tutoring/
- Record: https://forck.live/items/19914-simulating-student-learning-for-better-ai-tutoring
- Subject: Research / Phi

Microsoft Research introduced StudentSim, a system that creates personalized AI student simulators from de-identified learner records to evaluate and improve AI tutors before human testing. The system is evaluated across chess, second-language English writing, and mathematics using a new benchmark called StudentSimEval. StudentSim is a research system, not a product, and is not intended to replace teachers.

### NVIDIA — OCP Global Summit: Amid Ecosystem Adoption, NVIDIA Opens Vera Rubin Architecture to Help Scale AI Factories Worldwide

- Company: NVIDIA (nvidia.com)
- Announced: 2026-10-12T15:00:25+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://blogs.nvidia.com/blog/ocp-global-summit-vera-rubin-scale-ai-factories/
- Record: https://forck.live/items/19893-ocp-global-summit-amid-ecosystem-adoption-nvidia-opens-vera-rubin-architecture
- Subject: AI platform

At the OCP Global Summit, NVIDIA contributed Vera Rubin rack and compute tray designs to the open MGX architecture, updated the MGX 1.2 specification for AI factories with blind-mate liquid cooling and a 5,000A busbar, and launched open source software components for the DSX AI Factory platform, including Config Manager, Cluster Readiness Engine, Rack Management Service, and a Cluster API Provider.

### AMD — Introducing AMD Helion, Agentic AI Compute at Rack Scale

- Company: AMD (amd.com)
- Announced: 2026-10-12T13:00:00+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ir.amd.com/news-events/press-releases/detail/1301/introducing-amd-helion-agentic-ai-compute-at-rack-scale
- Record: https://forck.live/items/19828-introducing-amd-helion-agentic-ai-compute-at-rack-scale
- Subject: AMD Instinct / ROCm

AMD announced the AMD Helion agentic AI rack, an open blueprint based on an Open Compute Project specification that allows customers to customize CPU core density, performance, and power for agentic workloads. The rack supports CPU-only configurations for hyperscale deployments and integrated CPU/GPU configurations for enterprises, with plans to contribute the specification to OCP and make configurations available in November 2026.

### Tencent — You Work Across Six Apps. Your AI Should Too.

- Company: Tencent (tencent.com)
- Announced: 2026-10-12T08:49:19+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.tencent.com/you-work-across-six-apps-your-ai-should-too/
- Record: https://forck.live/items/19783-you-work-across-six-apps-your-ai-should-too
- Subject: Hunyuan

Tencent introduced WorkBuddy, an AI agent that edits documents, spreadsheets and presentations directly within the file rather than requiring copy-paste. It supports Human–AI Co-Writing in Word, Excel, PowerPoint and Tencent Docs, and can retrieve information from external tools via internal connectors or the Model Context Protocol (MCP).

### Lovable — EU inference: control where AI processes your requests

- Company: Lovable (lovable.dev)
- Announced: 2026-10-12T08:30:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/eu-inference
- Record: https://forck.live/items/19827-eu-inference-control-where-ai-processes-your-requests
- Subject: Lovable / AI app builder

Lovable introduced an EU inference setting for Enterprise customers that routes supported AI model requests to designated European endpoints, covering requests made while building in Lovable and through the built-in Lovable AI service in published apps. If no permitted endpoint can serve a request, it fails rather than being sent outside the configured boundary. The setting does not move all customer data into EU storage or establish a platform-wide data residency commitment, and some features such as video generation and voice conversations become unavailable when it is enabled.

### xAI — Introducing The X Subscription

- Company: xAI (x.ai)
- Announced: 2026-10-12
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.ai/news/x-subscription
- Record: https://forck.live/items/22304-introducing-the-x-subscription
- Subject: Grok

xAI launched The X Subscription, a single plan bundling Grok, Grok Bot, Cursor, and X Premium. The subscription includes a shared pool of Grok model usage across Grok and Cursor, with each tier providing more usage per dollar than the comparable standalone plan. Existing SuperGrok, Cursor, and X Premium subscribers can switch to the new plan with a prorated refund.

### Apple — Self-Compacting Language Model Agents

- Company: Apple (apple.com)
- Announced: 2026-10-12
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://machinelearning.apple.com/research/self-compacting-language-agents
- Record: https://forck.live/items/20061-self-compacting-language-model-agents
- Subject: Machine Learning Research

Authors Tianjian Li†, Jingyu Zhang†, William Jurayj†, Xi Wang†, Chuanyang Jin†, Mehrdad Farajtabar, Eric Nalisnick†, Daniel Khashabi† Long agent traces composed of chains of thought and tool calls accumulate stale content that anchor subsequent generations, and eventually …

### Amp — First Things First, Faster

- Company: Amp (ampcode.com)
- Announced: 2026-10-12
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/first-things-first-faster
- Record: https://forck.live/items/20059-first-things-first-faster
- Subject: Amp

Amp has added automatic intelligent ordering to its Changes pane, which reads the diff and places essential files first. Updated orders appear as a blue dot instead of rearranging files during review, and users can still opt for alphabetical ordering.

### Harvey — Harvey to Open Tokyo Office

- Company: Harvey (harvey.ai)
- Announced: 2026-10-12
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/harvey-to-open-tokyo-office
- Record: https://forck.live/items/20058-harvey-to-open-tokyo-office
- Subject: Harvey

Harvey announced plans to open a Tokyo office in early 2027 to better support its growing customer base in Japan, including law firms like Mori Hamada & Matsumoto and Nagashima Ohno & Tsunematsu, and in-house teams like Mitsubishi Corporation. The company is also developing Harvey to better support work in Japanese and has partnered with Legalscape for Japanese legal research. This expands Harvey's Asia-Pacific presence alongside teams in Sydney, Singapore, and Bengaluru.

### Apple — Minions: A Dynamic and Distributed Virtual GPU Cloud Built on Idle Apple Silicon Devices

- Company: Apple (apple.com)
- Announced: 2026-10-12
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://machinelearning.apple.com/research/distributed-gpu-cloud
- Record: https://forck.live/items/20055-minions-a-dynamic-and-distributed-virtual-gpu-cloud-built-on-idle-apple
- Subject: Machine Learning Research

Apple researchers introduced Minions, a distributed virtual GPU cloud that aggregates idle Apple Silicon GPUs across macOS devices to run large-scale machine learning inference workloads. The system aims to reduce reliance on centralized GPU infrastructure by leveraging existing hardware, providing fault tolerance and efficient resource utilization.

### Apple — Large Language Models for Machine Translation Quality Annotation: Humans and Models Are Both Challenged

- Company: Apple (apple.com)
- Announced: 2026-10-12
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://machinelearning.apple.com/research/llm-translation-quality-annotation
- Record: https://forck.live/items/20051-large-language-models-for-machine-translation-quality-annotation-humans-and
- Subject: Machine Learning Research

Apple researchers evaluated LLMs for machine translation quality annotation using MQM and ESA schemes across 70 language pairs and WMT23/WMT25 data. They found that LLM agreement with humans sometimes exceeds inter-human agreement, but both vary substantially and remain unreliable for most settings. Humans struggle with fine-grained MQM annotations and low-resource pairs, while LLMs struggle with minor errors, wrong language variants, and error span annotation.

### Amp — You've Got Comments

- Company: Amp (ampcode.com)
- Announced: 2026-10-12
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/github-listeners
- Record: https://forck.live/items/20018-you-ve-got-comments
- Subject: Amp

Amp now follows up on GitHub pull requests it opens by subscribing to comments, reviews, and CI results, which appear in the same thread. The company also introduced first-class GitHub Webhook support and the ability to spawn a new orb for each new GitHub issue or PR.

### Factory AI — Droids now have a fleet of sandboxes

- Company: Factory AI (factory.com)
- Announced: 2026-10-12
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/droids-now-have-a-fleet-of-sandboxes
- Record: https://forck.live/items/19995-droids-now-have-a-fleet-of-sandboxes
- Subject: Droid

Factory announced Environments, a new feature that lets users define reusable sandbox configurations for their Droids (cloud agents). Environments are pre-built into portable images, enabling instant onboarding and parallel execution of tasks. The feature is generally available for all plans and supports Automations, Agent Readiness scoring, and Missions.

### Shanghai AI Lab — 拓宽AI物理实验版图：上海AI实验室推出“书生SCP-Lab”，端砚自主实验能力再跃升

- Company: Shanghai AI Lab (shlab.org.cn)
- Announced: 2026-10-12
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.shlab.org.cn/news/5444305
- Record: https://forck.live/items/19644-scp-lab
- Title in English (translated by forck.live from zh): Expanding the AI Physics Experiment Landscape: Shanghai AI Lab Launches 'InternSCP-Lab', Further Elevating DuanYan's Autonomous Experiment Capabilities
- Subject: InternLM / 书生

Shanghai AI Lab released SCP-Lab, a protocol that provides a unified interface for AI agents to discover, task, and receive feedback from heterogeneous lab equipment. It uses a layered architecture and a five-level control granularity (L1–L5) to standardize access across different devices and vendors. The protocol has been validated in nine scenarios including chemical synthesis, materials optimization, and protein directed evolution.

### Cartesia — Raise Financial Services x Cartesia: Voice AI for Financial Access in India

- Company: Cartesia (cartesia.ai)
- Announced: 2026-10-12
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.cartesia.ai/blog/raise-financial-services
- Record: https://forck.live/items/19312-raise-financial-services-x-cartesia-voice-ai-for-financial-access-in-india
- Subject: Sonic / Ink

Cartesia is partnering with Raise Financial Services to provide multilingual voice AI for Raise's trading and investing platforms in India. Cartesia's Sonic text-to-speech model supports 11 Indic languages and powers Dhan's voice bot interactions, including conversational KYC. Over 75% of these interactions occur in languages other than English, with Hindi alone accounting for more than half.

### Sakana AI — Recursive Self-Improvement through Collective Intelligence

- Company: Sakana AI (sakana.ai)
- Announced: 2026-10-11T15:00:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/mass
- Record: https://forck.live/items/19304-recursive-self-improvement-through-collective-intelligence
- Subject: Sakana models

Sakana AI introduced Multi-Agent Self-Supervision (MASS), a method where a shared model uses a team of virtual subagents to solve tasks, searches for better multi-agent workflows, and selects the best ones using its own judgments. Training on the best team's executions distills collective experience back into the shared model. In two cycles using a 27B open-weights model on synthetic open-ended research tasks, the score per output token reached 1.2-1.6x the base model's level across four research benchmarks.

### GitHub — New controls and chat improvements in Copilot for JetBrains

- Company: GitHub (github.com)
- Announced: 2026-10-10T18:53:39+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://github.blog/changelog/2026-10-10-new-controls-and-chat-improvements-in-copilot-for-jetbrains
- Record: https://forck.live/items/19104-new-controls-and-chat-improvements-in-copilot-for-jetbrains
- Subject: Copilot

GitHub added enterprise controls for default Copilot agent models and MCP server startup in Copilot for JetBrains. The update also improves chat navigation, diagnostics with a Fix action, and account controls. Support for JetBrains IDE 2025.1 has ended, requiring version 2025.2 or later.

### GitHub — GitHub Copilot weekly releases — October 5

- Company: GitHub (github.com)
- Announced: 2026-10-09T22:58:23+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://github.blog/changelog/2026-10-09-github-copilot-weekly-releases-october-5
- Record: https://forck.live/items/18883-github-copilot-weekly-releases-october-5
- Subject: Copilot

GitHub added Claude Haiku 5.5 for Copilot Pro, Pro+, Max, Business, and Enterprise users. Local sandboxing for agents is now generally available in Copilot CLI, the Copilot app, and VS Code sessions using Agent Host at no extra cost. The Copilot app now supports separate GitHub accounts for licensing and repositories.

### Baseten — Project Beacon: Bringing safety signals to open models at scale

- Company: Baseten (baseten.co)
- Announced: 2026-10-09T17:01:47+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.baseten.co/blog/safe-open-models-baseten-goodfire
- Record: https://forck.live/items/21694-project-beacon-bringing-safety-signals-to-open-models-at-scale
- Subject: Baseten

Baseten and Goodfire AI announced Project Beacon, a partnership to bring proactive, inline safety controls to open model inference. The system reads a model's internal activations during generation to detect unsafe events as they happen and act before output reaches a user or tool. Baseten plans to release safety capabilities integrated with its inference platform over the next several months, starting with selected models and monitored behaviors.

### Amazon — ICYMI: What landed for AI builders in September 2026

- Company: Amazon (amazon.com)
- Announced: 2026-10-09T15:38:39+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://aws.amazon.com/blogs/machine-learning/icymi-what-landed-for-ai-builders-in-september-2026/
- Record: https://forck.live/items/18762-icymi-what-landed-for-ai-builders-in-september-2026
- Subject: Bedrock / Nova

Amazon announced several updates across Bedrock, AgentCore, and Strands. OpenAI's Astra, Sol, and Luna models became generally available on Bedrock, while Managed Agents powered by OpenAI entered public preview. Strands introduced an open source agent harness using 28% fewer tokens and a 2B-parameter decision model, Decider 2B. AgentCore's runtime improved memory management and cold start latency for serverless agents.

### Harvey — The Brief: October 2026

- Company: Harvey (harvey.ai)
- Announced: 2026-10-09T12:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/the-brief-october-2026
- Record: https://forck.live/items/17514-the-brief-october-2026
- Subject: Harvey

Harvey introduced a Connector Library, agentic search for SharePoint and Google Drive, Gmail search, PacerPro Vault sync, and Ironclad file import. It added a Motion to Dismiss workflow with LexisNexis, 130+ new legal research sources, Octus knowledge source, and custom knowledge source filters. Harvey improved Excel understanding, refreshed the review table, and made Anthropic's Fable 5.1 and OpenAI's GPT-6 Astra available. Mobile features include agents, connectors, live dictation, and localization in English (UK), German, Italian, and Portuguese.

### Ai2 — Ai2 at COLM 2026: Open research, from models to agents for science

- Company: Ai2 (allenai.org)
- Announced: 2026-10-09
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://allenai.org/blog/colm-2026-recap
- Record: https://forck.live/items/18893-ai2-at-colm-2026-open-research-from-models-to-agents-for-science
- Subject: Olmo / Molmo / Asta

This week Ai2 was at COLM 2026 in San Francisco. The work we shared reflects our wider mission: building AI research artifacts anyone can study and extend, and developing AI for science tools in collaboration with domain experts.

### Factory AI — Factory Acquires Mentlio: Building Intelligent Token Allocation for the Software Factory

- Company: Factory AI (factory.com)
- Announced: 2026-10-09
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/mentlio
- Record: https://forck.live/items/18767-factory-acquires-mentlio-building-intelligent-token-allocation-for-the
- Subject: Droid

Factory announced the acquisition of Mentlio (YC S26), a startup building in-harness model routing and context compression to optimize cost and quality of AI usage. The acquisition aims to help Factory bring efficient and intelligent autonomous software engineering to large enterprises by measuring and optimizing AI spend.

### Ai2 — Impactful scheduling for GPU clusters

- Company: Ai2 (allenai.org)
- Announced: 2026-10-09
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://allenai.org/blog/impactful-scheduling
- Record: https://forck.live/items/18749-impactful-scheduling-for-gpu-clusters
- Subject: Olmo / Molmo / Asta

Ai2's AI Infrastructure team replaced a priority-based GPU scheduler with a system using GPU time budgets, hierarchical fair-share allocation, and a time-slicing contract. The change shifted GPU time allocation from case-by-case operational decisions to a transparent administrative budgeting process, aiming to prioritize high-impact research while maintaining full cluster occupancy.

## How-tos, customer stories and company notes

_Implementation walkthroughs, changelog listings, customer stories, explainers and staffing posts: first-party and traceable, and not news of anything shipping. Here in full, and last._

### TypeSafe AI — How Jack & Jill Cut Candidate Screening Costs By 88% With Jev

- Company: TypeSafe AI (typesafe.ai)
- Announced: not stated by the source (first seen 2026-10-14T22:02:24.783063+00:00)
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://typesafe.ai/blog/jack-jill-jev-case-study
- Record: https://forck.live/items/22433-how-jack-jill-cut-candidate-screening-costs-by-88-with-jev
- Subject: Jev / System One

TypeSafe AI published a case study showing that talent marketplace Jack & Jill replaced Google's Gemini 3.1 Flash Lite with TypeSafe's Jev model for candidate screening. Jev maintained similar accuracy while reducing costs by 88% and cutting median screening time by roughly 50%. Jack & Jill projects $500K in annual savings and has expanded Jev to over 15 workflows.

### Microsoft — Microsoft named a Leader in the 2026 Gartner® Magic Quadrant™ for Server Virtualization Platforms

- Company: Microsoft (microsoft.com)
- Announced: 2026-10-14T19:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://azure.microsoft.com/en-us/blog/microsoft-named-a-leader-in-the-2026-gartner-magic-quadrant-for-server-virtualization-platforms/
- Record: https://forck.live/items/22464-microsoft-named-a-leader-in-the-2026-gartner-magic-quadrant-for-server
- Subject: Azure AI

Microsoft has been named a Leader in the 2026 Gartner Magic Quadrant for Server Virtualization Platforms. The company attributes this recognition to its approach of providing operational consistency through Azure and Azure Arc, and infrastructure consistency through Hyper-V, which helps organizations manage distributed environments.

### Amazon — Build self-improving document processing using Amazon Bedrock AgentCore

- Company: Amazon (amazon.com)
- Announced: 2026-10-14T16:24:07+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/build-self-improving-document-processing-using-amazon-bedrock-agentcore/
- Record: https://forck.live/items/22265-build-self-improving-document-processing-using-amazon-bedrock-agentcore
- Subject: Bedrock / Nova

Amazon published a guide on building an agentic intelligent document processing (IDP) system using Amazon Bedrock AgentCore. The system uses eight specialized agents across two graph-based workflows to classify, extract, validate, and learn from business documents with minimal human intervention. The post includes deployment steps using AWS CDK and prerequisites such as access to Amazon Bedrock with Amazon Nova and Anthropic Claude models.

### Amazon — Generate presentations with Amazon Bedrock AgentCore

- Company: Amazon (amazon.com)
- Announced: 2026-10-14T13:58:43+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/generate-presentations-with-amazon-bedrock-agentcore/
- Record: https://forck.live/items/22181-generate-presentations-with-amazon-bedrock-agentcore
- Subject: Bedrock / Nova

Amazon published a guide for building an AI agent system on Amazon Bedrock AgentCore that automates PowerPoint creation from enterprise documents, completing the task in under 90 seconds in internal tests. The solution uses a supervisor agent with the Strands Agents framework, Anthropic Claude Sonnet 4.6 for content, Amazon Titan Text Embeddings v2 for vector search, and Stability AI Stable Image Inpaint for slide images. It includes a RAG pipeline with Amazon OpenSearch Serverless and Amazon Bedrock Guardrails for content safety.

### Amazon — How Tamnoon built AI-powered cloud security remediation with Amazon Bedrock

- Company: Amazon (amazon.com)
- Announced: 2026-10-14T13:53:33+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/how-tamnoon-built-ai-powered-cloud-security-remediation-with-amazon-bedrock/
- Record: https://forck.live/items/22182-how-tamnoon-built-ai-powered-cloud-security-remediation-with-amazon-bedrock
- Subject: Bedrock / Nova

Tamnoon built Tami, an agentic cloud remediation orchestrator on Amazon Bedrock, that autonomously investigates and remediates cloud security findings. The platform uses four specialized AI agents for investigation, safety scoring, remediation planning, and code generation. Tamnoon reported an over 90 percent reduction in critical cloud exposure in 90 days with zero production issues.

### Amazon — Deploy and evaluate models with Amazon SageMaker JumpStart

- Company: Amazon (amazon.com)
- Announced: 2026-10-14T13:14:36+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/deploy-and-evaluate-models-with-amazon-sagemaker-jumpstart/
- Record: https://forck.live/items/22171-deploy-and-evaluate-models-with-amazon-sagemaker-jumpstart
- Subject: Bedrock / Nova

Amazon published a guide on deploying and evaluating models from SageMaker JumpStart using the SageMaker Python SDK. The post describes a three-step workflow: discover and deploy a model, run sample prompts to measure latency and generation rate, and clean up the endpoint. It also notes improvements to SageMaker JumpStart, including pre-configured endpoint deployment options and a deep link integration from Hugging Face.

### Baseten — Deploy Runway’s video models in dedicated environments on Baseten

- Company: Baseten (baseten.co)
- Announced: 2026-10-14T12:01:44+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.baseten.co/blog/deploy-runways-video-models-in-dedicated-environments-on-baseten
- Record: https://forck.live/items/22125-deploy-runway-s-video-models-in-dedicated-environments-on-baseten
- Subject: Baseten

Baseten partnered with Runway to offer Runway's Gen-4.5 and Aleph 2.0 video models in single-tenant, dedicated environments on Baseten Dedicated Inference. Each deployment runs in an isolated workload plane with dedicated GPU resources, and Baseten manages the control plane and infrastructure. The deployment is backed by a 99.99% uptime SLA and covered by Baseten's SOC 2 Type II and HIPAA compliance programs.

### OpenAI — SentinelOne cuts malware analysis from weeks to hours

- Company: OpenAI (openai.com)
- Announced: 2026-10-14T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/sentinelone
- Record: https://forck.live/items/22307-sentinelone-cuts-malware-analysis-from-weeks-to-hours
- Subject: GPT / ChatGPT / API

SentinelOne uses OpenAI Daybreak to reduce the time required for reverse-engineering sophisticated malware from four to six weeks to roughly eight hours, enabling researchers to investigate more threats and identify connections across samples. The capability is part of OpenAI's Daybreak offering for cybersecurity work. SentinelOne's Chief AI Officer notes that human accountability remains essential for architectural decisions.

### Lovable — Deploy your Lovable apps in your own AWS account

- Company: Lovable (lovable.dev)
- Announced: 2026-10-14T11:45:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/deploy-on-aws
- Record: https://forck.live/items/22118-deploy-your-lovable-apps-in-your-own-aws-account
- Subject: Lovable / AI app builder

Lovable announced Deploy on AWS, a private preview feature that lets enterprise teams run Lovable apps in their own AWS account, with all data and compute staying in a dedicated VPC. Lovable installs, manages, and updates the deployed infrastructure. Deutsche Telekom is the first customer using this deployment mode.

### OpenAI — How Block is building AI-powered security with OpenAI technology

- Company: OpenAI (openai.com)
- Announced: 2026-10-14T07:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/block-ai-powered-security
- Record: https://forck.live/items/22456-how-block-is-building-ai-powered-security-with-openai-technology
- Subject: GPT / ChatGPT / API

Block's security team uses OpenAI Daybreak to automate vulnerability discovery, detection, and response, compressing tasks from weeks to hours. The system has achieved a 50% boost in triage throughput and performs approximately 1,100 analyst-hours of investigation per quarter.

### Vercel — Findings from the $1M Vercel Sandbox challenge

- Company: Vercel (vercel.com)
- Announced: 2026-10-14T07:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/findings-from-the-1-million-dollar-vercel-sandbox-challenge
- Record: https://forck.live/items/22414-findings-from-the-1m-vercel-sandbox-challenge
- Subject: v0 / AI SDK / AI Gateway

Vercel published findings from its $1M Sandbox challenge, reporting 1,285 reports, $242,000 in bounties paid, and one Linux kernel zero-day vulnerability discovered by researcher Paulos Yibelo. The kernel defect allowed leaking host memory, crashing the host, and executing code on the host. No reports showed access to customer data.

### OpenAI — How Third Moment Research runs 10,000 agents for 36+ hours

- Company: OpenAI (openai.com)
- Announced: 2026-10-14T07:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/third-moment-research
- Record: https://forck.live/items/22220-how-third-moment-research-runs-10-000-agents-for-36-hours
- Subject: GPT / ChatGPT / API

With GPT‑6 Astra and GPT‑6.1 Sol, Codex, and Daybreak for cybersecurity, this AI lab builds agents that focus for days and improve with use.

### NVIDIA — NVIDIA Announces New Investments in the UK Startup Ecosystem

- Company: NVIDIA (nvidia.com)
- Announced: 2026-10-14T06:04:15+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blogs.nvidia.com/blog/uk-ecosystem-deployment/
- Record: https://forck.live/items/21922-nvidia-announces-new-investments-in-the-uk-startup-ecosystem
- Subject: AI platform

NVIDIA announced a five-year, $250 million compute commitment for UK agentic safety research, with 50 million annually through BOLD, and joined the Earthshot Prize as a Global Alliance Founding Partner. It detailed deployments by UK organizations using NVIDIA B300 Blackwell Ultra, Nemotron-Personas-UK, Nemotron Voice, NVIDIA Cosmos, NVIDIA BioNeMo, and NVIDIA Triton across energy, healthcare, finance, and quantum.

### Supabase — How we made the Supabase Select 2026 badge

- Company: Supabase (supabase.com)
- Announced: 2026-10-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/how-we-made-the-supabase-select-2026-badge
- Record: https://forck.live/items/22436-how-we-made-the-supabase-select-2026-badge
- Subject: Supabase

Supabase describes how it designed and produced 800 programmable badges for its Supabase Select 2026 developer conference, including the hardware choice (Pimoroni Tufty 2350 with RP2350), the firmware and software shipped, and the logistical challenges of charging, updating, and syncing all badges before the event. The post also highlights projects attendees built with the badges after the conference, such as a live flight tracker, a food-finder app, and a tiny story-generation model running entirely on the device.

### Ideogram — How to edit product photos with Ideogram 4.5

- Company: Ideogram (ideogram.ai)
- Announced: 2026-10-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://ideogram.ai/blog/ideogram-4-5-product-photo-editing
- Record: https://forck.live/items/22434-how-to-edit-product-photos-with-ideogram-4-5
- Subject: Ideogram image models

Ideogram published a guide on using Ideogram 4.5 for product photo editing, covering Precise Edit, Studio, and step-by-step examples such as changing a keyboard's background and accent keys. The post explains how to preserve product details, write effective edit instructions, and inspect results for unintended changes.

### OpenAI — The EU’s next chapter in AI is agentic

- Company: OpenAI (openai.com)
- Announced: 2026-10-14
- Category: research-paper
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/the-eus-next-chapter-in-ai-is-agentic
- Record: https://forck.live/items/22037-the-eu-s-next-chapter-in-ai-is-agentic
- Subject: GPT / ChatGPT / API

OpenAI's Economic Research report finds that among ChatGPT Enterprise customers in the EU, agentic tools Codex and ChatGPT Work now generate more than half of output tokens. Weekly active users of these tools among EU small and medium businesses grew more than eightfold since May 2026. The report also identifies a substantial gap in AI usage intensity between EU frontier firms and global frontier firms.

### Amazon — Building agents that transact with Anchor Browser and AgentCore payments

- Company: Amazon (amazon.com)
- Announced: 2026-10-13T20:10:33+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/building-agents-that-transact-with-anchor-browser-and-agentcore-payments/
- Record: https://forck.live/items/21711-building-agents-that-transact-with-anchor-browser-and-agentcore-payments
- Subject: Bedrock / Nova

Amazon announced the availability of Anchor Browser with Agentic Payments on AWS Marketplace, a solution developed with Agent Factory that integrates with Amazon Bedrock AgentCore payments. This enables AI agents to autonomously handle HTTP 402 Payment Required responses by managing wallet connectivity, spending limits, and payment processing through the x402 protocol, allowing agents to pay for premium datasets, research APIs, and paywalled content without manual intervention.

### Replit — Introducing Replit Solutions Partners

- Company: Replit (replit.com)
- Announced: 2026-10-13T17:25:04+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://replit.com/blog/introducing-replit-solutions-partners
- Record: https://forck.live/items/20579-introducing-replit-solutions-partners
- Subject: Replit Agent

Replit announced the Replit Solutions Partner Program, which connects customers with systems integrators, consultancies, and agencies that help identify opportunities, develop solutions, and get teams building on Replit. The program provides partners with training, enablement, and collaboration on customer opportunities, and offers prospective partners a certification path on Replit's enterprise features.

### Microsoft — Advancing hardware security with open-source Integrated HSM technology

- Company: Microsoft (microsoft.com)
- Announced: 2026-10-13T16:15:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://azure.microsoft.com/en-us/blog/advancing-hardware-security-with-open-source-integrated-hsm-technology/
- Record: https://forck.live/items/22169-advancing-hardware-security-with-open-source-integrated-hsm-technology
- Subject: Azure AI

Microsoft announced Project Tessera, an open-source evolution of its Azure Integrated Hardware Security Module (HSM), developed with Marvell, Oracle, and the Open Compute Project. The project makes the architecture, specifications, protocols, drivers, software, firmware, and reproducible build environments open source, with a goal of advancing toward a fully open-source silicon implementation. Tessera is designed to enforce FIPS 140-3 Level 3 key protection throughout the key lifecycle and address key management anti-patterns.

### Amazon — Natural language security queries at Orca Security using Amazon Bedrock

- Company: Amazon (amazon.com)
- Announced: 2026-10-13T16:03:31+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/natural-language-security-queries-at-orca-security-using-amazon-bedrock/
- Record: https://forck.live/items/20519-natural-language-security-queries-at-orca-security-using-amazon-bedrock
- Subject: Bedrock / Nova

Orca Security partnered with AWS to build Discovery AI, a natural language query agent on Amazon Bedrock that converts plain-text questions into Sonar queries. Compared to an initial baseline, Orca reduced query costs by 90 percent, input tokens by 93 percent, and latency by 75 percent. The system handles thousands of queries per day with 15-second response times.

### ByteDance — Seed Alignment & Preparedness team looking for global talents

- Company: ByteDance (bytedance.com)
- Announced: 2026-10-13T16:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://seed.bytedance.com/en/blog/seed-alignment-preparedness-team-looking-for-global-talents
- Record: https://forck.live/items/22102-seed-alignment-preparedness-team-looking-for-global-talents
- Subject: Seed / Doubao

ByteDance's Seed research unit announced the formation of an Alignment & Preparedness team focused on model safety and alignment. The team plans to initially recruit 10 researchers worldwide to work on safety foundations, evaluation systems, training methods, and external partnerships.

### SenseTime — 他看不见屏幕，却用 AI 为视障孩子搭了一座“光桥”

- Company: SenseTime (sensetime.com)
- Announced: 2026-10-13T16:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.sensetime.com/cn/news/news-article-1
- Record: https://forck.live/items/22039-ai
- Title in English (translated by forck.live from zh): He can't see the screen, but he used AI to build a 'bridge of light' for visually impaired children
- Subject: SenseNova / 商汤

SenseTime's case study documents how visually impaired user Cao Wenwei used the company's Raccoon Work AI assistant to build 'Guangqiao AI', a personalized accessible learning support system for visually impaired and multi-disabled students. The system, which won silver in SenseTime's OPC challenge, helps teachers and parents generate learning materials, teaching suggestions, and feedback forms. Cao also serves as a blind information accessibility supervisor, advocating for better digital accessibility.

### Amazon — How ShipStation Global built an AI IT help desk on Amazon Bedrock AgentCore

- Company: Amazon (amazon.com)
- Announced: 2026-10-13T15:54:49+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/how-shipstation-global-built-an-ai-it-help-desk-on-amazon-bedrock-agentcore/
- Record: https://forck.live/items/20520-how-shipstation-global-built-an-ai-it-help-desk-on-amazon-bedrock-agentcore
- Subject: Bedrock / Nova

ShipStation Global built Deckhand, an AI-powered IT help desk agent for Slack, on Amazon Bedrock AgentCore. The agent resolves 97.5 percent of incoming channel queries and earns a 96 percent positive feedback rating from employees. Deckhand answers routine IT questions from a curated knowledge base, reads error-message screenshots, and creates or updates Jira Service Management tickets.

### Amazon — Build long-horizon agents with Grok 4.7 on Amazon Bedrock

- Company: Amazon (amazon.com)
- Announced: 2026-10-13T15:52:20+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/build-long-horizon-agents-with-grok-4-7-on-amazon-bedrock/
- Record: https://forck.live/items/20509-build-long-horizon-agents-with-grok-4-7-on-amazon-bedrock
- Subject: Bedrock / Nova

Amazon Bedrock now offers SpaceXAI's Grok 4.7, which supports configurable reasoning effort levels (low, medium, high, xhigh) and tool calling. In tests on three long-horizon agent tasks, the model answered 67 of 72 runs correctly, with xhigh achieving 18/18 and low achieving 15/18 while using roughly one-eighth the output tokens. The model has a 500K token context window and accepts text and image input with text output.

### Amazon — Mixture-of-Experts kernel engineering on Amazon SageMaker AI

- Company: Amazon (amazon.com)
- Announced: 2026-10-13T15:17:09+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/mixture-of-experts-kernel-engineering-on-amazon-sagemaker-ai/
- Record: https://forck.live/items/20502-mixture-of-experts-kernel-engineering-on-amazon-sagemaker-ai
- Subject: Bedrock / Nova

Amazon describes kernel optimizations for fused Mixture-of-Experts models on SageMaker AI, focusing on a Swap-AB technique that transposes matrices to improve WGMMA utilization on NVIDIA Hopper GPUs during the decode phase.

### Runway — The Runway Weekly Challenge is here.

- Company: Runway (runway.com)
- Announced: 2026-10-13T14:00:58+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://x.com/runwayml/status/2110007893194154167
- Record: https://forck.live/items/20469-the-runway-weekly-challenge-is-here

Runway launched a weekly creative challenge offering 175,000 credits in prizes, with a new brief each week and community-voted winners.

### NVIDIA — NVIDIA Announces New Investments in France’s Startups

- Company: NVIDIA (nvidia.com)
- Announced: 2026-10-13T05:00:31+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blogs.nvidia.com/blog/france-ecosystem-agentic-computing/
- Record: https://forck.live/items/20198-nvidia-announces-new-investments-in-france-s-startups
- Subject: AI platform

NVIDIA highlights how companies in France are deploying agentic computing across space, financial services, the public sector, and scientific research, building on sovereign AI infrastructure and open models. Bull plans to extend European AI server production to NVIDIA Vera Rubin Ultra NVL72 systems by 2028. The Altair-1 satellite, part of a 50-satellite constellation, launched with NVIDIA Jetson AGX Orin modules running Mistral's models for real-time Earth observation. LINAGORA and Docaposte secured a framework agreement making Twake.ai, built on NVIDIA Nemotron, the sole referenced sovereign digital workplace for over 2,500 French public organizations. Mistral introduced Mistral Large 4 under private preview with open-weight performance. L'Oréal uses the NVIDIA ALCHEMI Toolkit for cosmetic formulation discovery targeting up to 100x faster results. Mirakl onboarded over 70 million products with fine-tuned models on NVIDIA GPUs, cutting catalog onboarding to under two hours. ThinkDeep and Akila built a digital-twin agent managing nearly 5,800 kilometers of roads and buildings, cutting total operating costs by 10% and targeting 30%.

### Manus — How This Boston Startup Is Giving Demolition Waste a Second Life

- Company: Manus (manus.im)
- Announced: 2026-10-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://manus.im/blog/circa-builds-customer-story
- Record: https://forck.live/items/21805-how-this-boston-startup-is-giving-demolition-waste-a-second-life
- Subject: Manus

Manus published a customer story about Circa, a Boston demolition startup that uses Manus to run a lean operation. Circa's cofounders built their website on Manus, used it to prepare grant applications and internal tools, and reported that a remote marketing teammate could produce roughly 50 social posts per week versus five through conventional workflow. The company also uses a visual intake system called Waste-Less Vision, which identifies materials from customer photos or video and prepares preliminary estimates.

### Ai2 — How Molmo 2’s open release helped an MIT researcher rethink fish tracking

- Company: Ai2 (allenai.org)
- Announced: 2026-10-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://allenai.org/blog/molmo2fish
- Record: https://forck.live/items/20637-how-molmo-2-s-open-release-helped-an-mit-researcher-rethink-fish-tracking
- Subject: Olmo / Molmo / Asta

Ai2 published a case study about MIT researcher Kai Van Brunt, who adapted the open vision-language model Molmo 2 to create Molmo2Fish, a system that tracks salmon in underwater sonar footage and can correct its predictions based on natural-language instructions. The project demonstrates how open model releases enable researchers to apply existing capabilities to specialized ecological problems.

### Suno — Behind the Scenes: Make what moves you

- Company: Suno (suno.com)
- Announced: 2026-10-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://suno.com/blog/behind-the-scenes-make-what-moves-you
- Record: https://forck.live/items/20630-behind-the-scenes-make-what-moves-you
- Subject: Music / Studio

Suno released v6, its latest music generation model. To demonstrate the model's flexibility, the company collaborated with Emmy Award-winning director and choreographer Nina McNeely, dancer Braylon Browner, and singer-songwriter/producer Banoffee on two creative projects. Each artist used v6 as the foundation for an original song, with Browner building from movement and Banoffee from genre-mixing and audio experimentation.

### Google DeepMind — Applying technology to understand and treat disease

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-10-12T17:30:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.google/innovation-and-ai/technology/health/ai-health-research/
- Record: https://forck.live/items/19969-applying-technology-to-understand-and-treat-disease
- Subject: Gemini

Google DeepMind highlights its health research efforts, including AlphaGenome Atlas for genetic variant mapping, AlphaFold for drug discovery, and clinical tools like DeepVariant and DeepSomatic. The post notes 90 peer-reviewed papers and collaborations with institutions such as Stanford and the NHS. It also describes a mammography screening system that caught 25% of missed interval cancers and a flu forecasting model ranked #1 by the CDC.

### Amazon — Deploy decision models with llama.cpp on Amazon SageMaker AI

- Company: Amazon (amazon.com)
- Announced: 2026-10-12T15:51:46+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/deploy-decision-models-with-llama-cpp-on-amazon-sagemaker-ai/
- Record: https://forck.live/items/19906-deploy-decision-models-with-llama-cpp-on-amazon-sagemaker-ai
- Subject: Bedrock / Nova

Amazon published a guide on deploying decision models using llama.cpp on Amazon SageMaker AI. The post covers two deployment paths: a standard text-generation path using the AWS DLC for llama.cpp with an OpenAI-compatible API, and a structured-decision path using a SemIf adapter that returns probabilities for declared options. The guide includes CPU and GPU deployment options and benchmarks showing GPU endpoints were faster than CPU.

### Amazon — Run multi-node inference workloads using Ray LLM Deep Learning Containers

- Company: Amazon (amazon.com)
- Announced: 2026-10-12T15:50:14+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/run-multi-node-inference-workloads-using-ray-llm-deep-learning-containers/
- Record: https://forck.live/items/19907-run-multi-node-inference-workloads-using-ray-llm-deep-learning-containers
- Subject: Bedrock / Nova

AWS announced the Ray LLM Deep Learning Container (DLC) for multi-node inference, enabling deployment of a single large language model across two GPU nodes on Amazon EKS with separate prefill and decode phases. The container bundles vLLM, Ray Serve, NVIDIA NIXL for KV-cache transfer, and the EFA stack, exposing an OpenAI-compatible API without custom application code. This setup addresses workloads such as RAG, document summarization, and agentic applications by preventing compute-heavy prefill from stalling token generation.

### Harvey — Ogier Adopts Harvey Firmwide

- Company: Harvey (harvey.ai)
- Announced: 2026-10-12
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/ogier-adopts-harvey-firmwide
- Record: https://forck.live/items/19683-ogier-adopts-harvey-firmwide
- Subject: Harvey

International and offshore law firm Ogier has adopted Harvey firmwide after a successful pilot across its 13 offices, equipping its 500-strong legal team with AI for document analysis, legal research, drafting and knowledge retrieval. The deployment is part of Ogier's long-term investment in digital transformation and responsible AI, with data protection and confidentiality as central considerations. Harvey supports cross-border complexity in Ogier's corporate, finance, funds, dispute resolution, private wealth and regulatory work.

### SenseTime — 商汤小浣熊霸榜了！月增量超 1500 万

- Company: SenseTime (sensetime.com)
- Announced: 2026-10-11T16:00:19+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.sensetime.com/cn/news/1500-20261013-1655
- Record: https://forck.live/items/21985-1500
- Title in English (translated by forck.live from zh): SenseTime Xiaohuanxiong tops the charts! Monthly growth exceeds 15 million
- Subject: SenseNova / 商汤

SenseTime's Xiaohuanxiong AI assistant ranked second in monthly visit growth for office agents and second in web monthly unique visitors and average usage time in the AI efficiency category, according to QuantumBit's September AI App Monthly Report. The assistant's monthly user growth exceeded 15 million.

### Amp — Use Your Claude Plan in Amp

- Company: Amp (ampcode.com)
- Announced: 2026-10-10
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://ampcode.com/news/use-your-claude-plan
- Record: https://forck.live/items/18971-use-your-claude-plan-in-amp
- Subject: Amp

Amp added a Claude Code mode that lets users link their existing Claude Pro or Max subscription to use the Claude Agent SDK within Amp, supporting orbs, runners, thread sharing, portals, multiplayer, and orchestration.

### Amazon — How Postman runs Agent Mode for 40 million developers on Amazon Bedrock

- Company: Amazon (amazon.com)
- Announced: 2026-10-09T15:35:02+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/how-postman-runs-agent-mode-for-40-million-developers-on-amazon-bedrock/
- Record: https://forck.live/items/18763-how-postman-runs-agent-mode-for-40-million-developers-on-amazon-bedrock
- Subject: Bedrock / Nova

Postman built Agent Mode, an AI-native interface for API testing and documentation, using Amazon Bedrock for model flexibility and inference. The team found that controlling tool sprawl and treating context as the primary bottleneck were key challenges. They dynamically scope tools per task, decouple tools from UI state, and use Amazon Bedrock Guardrails for PII redaction.

### Baseten — What makes a high-quality NVFP4 quantization?

- Company: Baseten (baseten.co)
- Announced: 2026-10-09T14:16:52+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.baseten.co/blog/what-makes-a-high-quality-nvfp4-quantization
- Record: https://forck.live/items/21695-what-makes-a-high-quality-nvfp4-quantization
- Subject: Baseten

Baseten's post explains three approaches to assigning precision formats per layer in NVFP4 quantization: heuristic-based algorithms, isolated-layer sensitivity scoring, and SaturationQuant, which incorporates saturation so scores can be summed across a mixed-precision recipe. The post also covers calibration steps and the role of block scales.

### Vercel — How Rillet ships 3× faster with AI agents on Vercel

- Company: Vercel (vercel.com)
- Announced: 2026-10-09T04:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/how-rillet-ships-3-faster-with-ai-agents-on-vercel
- Record: https://forck.live/items/20648-how-rillet-ships-3-faster-with-ai-agents-on-vercel
- Subject: v0 / AI SDK / AI Gateway

Rillet, an AI-native ERP platform, describes how its two-person frontend team uses Vercel's open-source agent framework eve to build and deploy AI agents that handle customer requests, from ticket capture to code changes and testing. The team tripled its shipping rate over three months and can ship customer-requested changes to production in as little as two hours. Rillet also uses Vercel Connect for Slack and Linear access and AI Gateway to route model requests from multiple providers.

### Cohere — Shared or dedicated inference for Embed & Rerank

- Company: Cohere (cohere.com)
- Announced: 2026-10-09
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cohere.com/blog/shared-or-dedicated-inference-for-embed-rerank
- Record: https://forck.live/items/18831-shared-or-dedicated-inference-for-embed-rerank
- Subject: Command

Cohere published a guide explaining that the choice between shared (consumption-based) and dedicated (provisioned) inference for its Embed and Rerank models depends on the application's request profile, traffic pattern, and utilization, not just on headline pricing. The guide details how request size, token volume, and reranking candidate count affect cost efficiency and latency requirements.

### Perplexity — AI Agent vs MCP? When To Use Each (& When You Need Both)

- Company: Perplexity (perplexity.ai)
- Announced: 2026-10-09
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.perplexity.ai/hub/blog/agent-vs-mcp
- Record: https://forck.live/items/18783-ai-agent-vs-mcp-when-to-use-each-when-you-need-both
- Subject: Perplexity

Perplexity's guide explains the difference between AI agents, which autonomously decide actions to achieve goals, and the Model Context Protocol (MCP), an open standard for connecting AI applications to external tools and data. It provides criteria for when to use each approach, when to combine them, and when to use predefined workflows instead.

## Press chatter — a separate record, not announcements

38 stories are being carried by outlets this tracker follows with no post from the company under them, and 2 have since been answered by an announcement. Those rows are **not** in the list above and never will be: they are press coverage, not first-party announcements, and this file holds only announcements.

Fetch https://forck.live/chatter.md for them. Attribute every claim there to the outlet that made it, treat an entry with no announcement as unconfirmed rather than as forthcoming, and never read a large outlet count as confirmation. Page: https://forck.live/chatter

## Other doors

- One story in full, with its evidence quote and the outlets named: `/items/{slug}.md` — the slug is the one in that entry's `Record` link above
- One lab only: `/company/{slug}/feed.xml`, e.g. https://forck.live/company/anthropic/feed.xml
- A scheduled or recurring brief: https://forck.live/feed.xml, everything confirmed and newest first — keep the items newer than your last run

## Weekly archive

One file per ISO week, same entry shape as above. Each week listed here has
closed, and each window is fixed, so a week's URL always names the same seven
days; a post published inside them and detected later is added when it
arrives. The issue above covers 7 calendar days rather than an ISO
week, so it straddles 2026-W41 and 2026-W42: the announcements it holds
from 2026-W41 are also in the file for 2026-W41 below, which
therefore overlaps this issue rather than sitting entirely before it. Newest first:

- https://forck.live/week/2026-W41.md — 5–11 October 2026 — 159 announcements
- https://forck.live/week/2026-W40.md — 28 September – 4 October 2026 — 194 announcements
- https://forck.live/week/2026-W39.md — 21–27 September 2026 — 178 announcements
- https://forck.live/week/2026-W38.md — 14–20 September 2026 — 135 announcements
- https://forck.live/week/2026-W37.md — 7–13 September 2026 — 146 announcements
- https://forck.live/week/2026-W36.md — 31 August – 6 September 2026 — 151 announcements