# forck.live — 13–19 July 2026

> 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 one issue of the weekly archive: every confirmed
first-party announcement of ISO week 2026-W29, 13–19 July 2026, newest
first. The window never moves, so this URL always names the same seven days —
but a post published inside them and detected later is added when it arrives,
so a closed week can still gain entries.

Every entry is one first-party announcement, and carries the same eight fields
in this order before any others:

    Company · Announced · Category · Coverage · Announcement · Group · Source · Record

`Source` is the company's own page the announcement was traced to and `Record`
is this site's page for it. Both are safe to cite.

`Coverage` is a number of outlets, or `not counted` — which means this record
reached no count of its own, never that nobody wrote about it.

`Category` is this tracker's own analysis, and is authoritative wherever it is
stated. `not stated` means either that nothing has analysed the post yet or
that the analysis placed it outside this taxonomy; neither is a verdict about
the announcement.

`Announcement` is whether the post is news of something the company did, rather
than a how-to, a customer story, a changelog listing or a staffing note.

`Group` is which of this record's groups the entry falls in, carried on the
entry itself so it holds where the groups are headings and where they are not.
One of: `models`, `covered`, `announcements`, `routine`.

Anything after `Record` is held detail that only some announcements carry.

- This week's page: https://forck.live/week/2026-W29
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### ByteDance — From Speech to Audio Creation | Introducing the Seed Audio 1.0 Audio Creation Model

- Company: ByteDance (bytedance.com)
- Announced: 2026-07-19T16:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://seed.bytedance.com/en/blog/from-speech-to-audio-creation-introducing-the-seed-audio-1-0-audio-creation-model
- Record: https://forck.live/items/4660-from-speech-to-audio-creation-introducing-the-seed-audio-1-0-audio-creation
- Subject: Seed / Doubao

ByteDance announces Seed Audio 1.0, an audio creation model that generates speech, sound effects, ambience, and other scene-level audio elements within one unified framework. It supports 20+ languages, can generate up to two minutes of audio in a single pass, and offers prompt-level timing control at 100 ms intervals. The model is now available through BytePlus.

### Amp — Subscriptions, At Last

- Company: Amp (ampcode.com)
- Announced: 2026-07-18
- Category: pricing-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/subscriptions
- Record: https://forck.live/items/7727-subscriptions-at-last
- Subject: Amp

Amp launched Amp Subscriptions in Beta, offering monthly pricing with cheaper and more predictable costs, including bonus 2x usage for orbs and agents for subscribers who join by July 19. Users can still pay-as-you-go, and beyond subscription limits they must link a ChatGPT subscription or add paid credits.

### NVIDIA — NVIDIA Vera Rubin Maximizes Intelligence per Dollar for Post-Training Workloads — a Key Metric for Agentic AI

- Company: NVIDIA (nvidia.com)
- Announced: 2026-07-17T15:00:14+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://blogs.nvidia.com/blog/nvidia-vera-rubin-post-training-intelligence-per-dollar/
- Record: https://forck.live/items/1443-nvidia-vera-rubin-maximizes-intelligence-per-dollar-for-post-training
- Subject: AI platform

NVIDIA announces the Vera Rubin platform, codesigned to maximize intelligence per dollar for agentic post-training workloads, and highlights the Nemotron 3 Ultra model and NeMo libraries for post-training infrastructure.

### Google DeepMind — Introducing Gemini 3.5 Flash Cyber

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-07-17T15:00:11+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://deepmind.google/blog/introducing-gemini-3-5-flash-cyber/
- Record: https://forck.live/items/1205-introducing-gemini-3-5-flash-cyber
- Subject: Gemini

Google DeepMind introduces Gemini 3.5 Flash Cyber, a lightweight cybersecurity model fine-tuned from 3.5 Flash to find, validate, and patch vulnerabilities. It is available via a limited-access pilot program to governments and trusted partners through CodeMender.

### Harvey — The Brief: July 2026

- Company: Harvey (harvey.ai)
- Announced: 2026-07-17T15:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/the-brief-july-2026
- Record: https://forck.live/items/17541-the-brief-july-2026
- Subject: Harvey

Harvey added agentic features including a Thread Experience for handing off multi-step work, the ability to edit review tables in-line and attach files as column context, email notifications for long-running reviews, organization-wide context for in-house teams, audio transcription, PowerPoint and Excel generation in custom workflows, a Playbook Creation Agent, and integrations with Microsoft 365 Copilot and Cowork. The update also includes dictation in Word and Outlook add-ins, expanded Android and iOS features, French localization, new models (Claude Sonnet 5 and GPT-5.6 Sol), over 245 new legal research sources, advanced Vault controls, internal Space admin management, workspace Playbook management, taxonomy updates in Library, and a new Harvey Academy certification for AI document analysis.

### Runway — Inside Our Approach to Building Runway Agent

- Company: Runway (runway.com)
- Announced: 2026-07-17T12:38:12.854000+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://runway.com/news/engineering/inside-building-runway-agent
- Record: https://forck.live/items/7866-inside-our-approach-to-building-runway-agent
- Subject: Gen / Aleph

Runway launched Runway Agent 2.0, an agentic video creation tool that helps users turn a single idea into finished creative work. The product was built with input from Runway's in-house creative team and is designed to support marketers, cinematic creators, and others. Runway Agent 2.0 was ranked first among six agents in the Arc 1.0 benchmark by Physion Labs.

### OpenAI — A scorecard for the AI age

- Company: OpenAI (openai.com)
- Announced: 2026-07-17T10:00:00+00:00
- Category: not stated
- Coverage: 1 outlet
- Announcement: no
- Group: covered
- Source: https://openai.com/index/a-scorecard-for-the-ai-age
- Record: https://forck.live/items/24-a-scorecard-for-the-ai-age
- Subject: GPT / ChatGPT / API

Sarah Friar, CFO of OpenAI, introduces a practical AI scorecard to measure ROI through useful work, cost per successful task, dependability, and return on compute.

### Manus — How Manus became “James,” the AI chief of staff behind Ascendea’s 90× output shift

- Company: Manus (manus.im)
- Announced: 2026-07-17
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://manus.im/blog/Ascendea-James-Customer-Story
- Record: https://forck.live/items/14705-how-manus-became-james-the-ai-chief-of-staff-behind-ascendea-s-90-output-shift
- Subject: Manus

Ascendea, a UK firm building AI systems for small businesses, uses the Manus agent as the top layer of a three-tier AI workforce. Manus turned scattered process knowledge into 303 SOPs in one day, a task Ascendea's CTO estimated would have taken two people about 14 weeks manually, and now handles complex work such as strategy documents, technical architecture, and M&A prospecting workflows.

### Amp — From Agent to Agent

- Company: Amp (ampcode.com)
- Announced: 2026-07-17
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/from-agent-to-agent
- Record: https://forck.live/items/7728-from-agent-to-agent
- Subject: Amp

Amp now allows users to ask their agents to spawn other agents, send messages and files, and coordinate work across threads, orbs, and machines.

### Cursor — Improvements to Cursor in Slack

- Company: Cursor (cursor.com)
- Announced: 2026-07-17
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cursor.com/changelog/slack-improvements
- Record: https://forck.live/items/2591-improvements-to-cursor-in-slack
- Subject: Cursor

Cursor in Slack now shares a plan before starting, supports multi-repo environments, and can work across channels and threads.

### Databricks — Your AI is ready. Your data foundation probably isn’t

- Company: Databricks (databricks.com)
- Announced: 2026-07-16T22:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.databricks.com/blog/your-ai-ready-your-data-foundation-probably-isnt
- Record: https://forck.live/items/17504-your-ai-is-ready-your-data-foundation-probably-isn-t
- Subject: Mosaic AI

Cushman & Wakefield built an enterprise AI core over four years by embedding technologists in business units under a product operating model, prioritizing trust and human behavior. Using Databricks, including Genie for natural-language data governance and quality checks, the company cut idea-to-outcome timelines from months to days.

### Replit — The Self-Driving Company

- Company: Replit (replit.com)
- Announced: 2026-07-16T17:01:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://replit.com/blog/self-driving-company
- Record: https://forck.live/items/17892-the-self-driving-company
- Subject: Replit Agent

Replit reports that internal AI agents, integrated across engineering and other teams, led to a 5.8× increase in lines of code contributed from January to June 2025, with a 2.9× increase per engineer when controlling for hiring. Code review latency remained flat, PR reversion rates and incident counts stayed steady, and project completion rates rose. The company describes this as the beginning of a 'self-driving company' where agents handle tasks such as incident investigation, code review, and support triage, while humans set goals and make tradeoffs.

### OpenAI — Why teens deserve access to safe AI

- Company: OpenAI (openai.com)
- Announced: 2026-07-16T16:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/why-teens-deserve-access-safe-ai
- Record: https://forck.live/items/25-why-teens-deserve-access-to-safe-ai
- Subject: GPT / ChatGPT / API

OpenAI is adding age-appropriate protections, learning tools, parental controls, and expert partnerships to ChatGPT to make it safer for teens.

### AMD — AMD Appoints Alan Smith as Newest Corporate Fellow

- Company: AMD (amd.com)
- Announced: 2026-07-16T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://ir.amd.com/news-events/press-releases/detail/1290/amd-appoints-alan-smith-as-newest-corporate-fellow
- Record: https://forck.live/items/15039-amd-appoints-alan-smith-as-newest-corporate-fellow
- Subject: AMD Instinct / ROCm

AMD appointed Alan Smith as a Corporate Fellow, recognizing his contributions to GPU architecture and technical leadership that have advanced AMD AI computing.

### Hugging Face — Newer Models, Same Advantage

- Company: Hugging Face (huggingface.co)
- Announced: 2026-07-16T11:49:48+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/Dharma-AI/newer-models-same-advantages
- Record: https://forck.live/items/1454-newer-models-same-advantage
- Subject: Platform

DharmaOCR, a specialized OCR model for Brazilian Portuguese, continues to outperform newer general-purpose models Mistral OCR4 and Unlimited-OCR on a Portuguese benchmark, demonstrating that domain specialization provides a measurable advantage even against more recent architectures.

### Harvey — Y Combinator-Backed Benchmark Joins Harvey

- Company: Harvey (harvey.ai)
- Announced: 2026-07-16T11:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/y-combinator-backed-benchmark-joins-harvey
- Record: https://forck.live/items/17542-y-combinator-backed-benchmark-joins-harvey
- Subject: Harvey

Harvey announced its acquisition of Benchmark, a decision infrastructure platform for asset management. Benchmark's co-founders and team will join Harvey's product and engineering organization. This is Harvey's third acquisition of 2026, following a record Q2 in which Harvey added more than $100 million in net-new ARR.

### Google DeepMind — Our approach to bioresilience

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-07-16T09:30:42+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/our-approach-to-bioresilience/
- Record: https://forck.live/items/1206-our-approach-to-bioresilience
- Subject: Gemini

Google DeepMind and Isomorphic Labs publish a statement on their joint approach to bioresilience and AI models, without introducing any new product, model, or capability.

### OpenAI — How Codex became a collaborator for OpenAI’s creative team

- Company: OpenAI (openai.com)
- Announced: 2026-07-16T07:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/codex-collaborator-creative-team
- Record: https://forck.live/items/26-how-codex-became-a-collaborator-for-openai-s-creative-team
- Subject: GPT / ChatGPT / API

OpenAI's creative team uses Codex to build custom creative tools, accelerate ideation, and prototype faster with context-aware AI.

### Supabase — Unified Logs is now in open beta

- Company: Supabase (supabase.com)
- Announced: 2026-07-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/unified-logs-open-beta
- Record: https://forck.live/items/21280-unified-logs-is-now-in-open-beta
- Subject: Supabase

Supabase announced that Unified Logs is now in open beta. The feature aggregates logs from all project services—including the API gateway, Postgres, Auth, Storage, PostgREST, Realtime, and connection poolers—into a single searchable view. Users can filter by log type, level, status, method, or pathname, perform free-text searches, and share filter links via the URL. A live tail mode streams new events, and a histogram shows log volume over time with error and warning highlights. The detail panel displays a request's journey across services, with raw JSON available in a separate tab. Unified Logs is rolling out to all projects during the beta period, and users can revert to the classic Logs Explorer at any time.

### Isomorphic Labs — Our approach to bioresilience

- Company: Isomorphic Labs (isomorphiclabs.com)
- Announced: 2026-07-16
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.isomorphiclabs.com/articles/our-approach-to-bioresilience
- Record: https://forck.live/items/19132-our-approach-to-bioresilience
- Subject: Drug Design Engine / AlphaFold

Isomorphic Labs and Google DeepMind published their joint approach to bioresilience, stating that frontier AI models and the scientific advances they enable are needed to help society become more resilient to events like future outbreaks.

### H Company — Computer-use Agents launch as a managed API

- Company: H Company (hcompany.ai)
- Announced: 2026-07-16
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://hcompany.ai/newsroom/computer-use-agents-api
- Record: https://forck.live/items/18364-computer-use-agents-launch-as-a-managed-api
- Subject: Holo

H Company launched Computer-use Agents, a managed API for agents that perform tasks on computers. The agents are powered by Holo3 and Holo3.1 models, which achieved 80.4% on OSWorld-Verified. The API includes browser and desktop agents, with SDKs in TypeScript and Python, and supports MCP server, CLI, and REST API integrations.

### Moonshot AI — PerceptionBench

- Company: Moonshot AI (moonshot.ai)
- Announced: 2026-07-16
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.kimi.ai/blog/perception-bench
- Record: https://forck.live/items/4622-perceptionbench
- Subject: Kimi

Moonshot AI releases PerceptionBench, a benchmark for evaluating atomic visual perception in multimodal large language models, with 3,000 verified questions across 10 atomic perceptual categories, and open-sources the dataset and evaluation code.

### Moonshot AI — Kimi K3

- Company: Moonshot AI (moonshot.ai)
- Announced: 2026-07-16
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.kimi.ai/blog/kimi-k3
- Record: https://forck.live/items/4621-kimi-k3
- Subject: Kimi

Moonshot AI introduces Kimi K3, a 2.8T-parameter open model with native vision and a 1-million-token context window, available today on Kimi platforms and API, with weights to be released by July 2026. It demonstrates strong performance in coding, knowledge work, and reasoning, outperforming several other models in benchmarks while still trailing the most powerful proprietary models like Claude Fable 5 and GPT 5.6 Sol.

### xAI — Automations in Grok

- Company: xAI (x.ai)
- Announced: 2026-07-16
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.ai/news/grok-automations
- Record: https://forck.live/items/3442-automations-in-grok
- Subject: Grok

xAI announced a new automation feature for Grok that allows users to define a job and have it run on a schedule or triggered by an email, with results reported back.

### xAI — Introducing Grok 4.5

- Company: xAI (x.ai)
- Announced: 2026-07-16
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://x.ai/news/grok-4-5
- Record: https://forck.live/items/3441-introducing-grok-4-5
- Subject: Grok

xAI announced Grok 4.5, described as their smartest model built for coding, agentic tasks, and knowledge work.

### Cohere — Cohere and the University of Toronto partner to advance responsible AI adoption at scale

- Company: Cohere (cohere.com)
- Announced: 2026-07-16
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cohere.com/blog/cohere-university-of-toronto-announcement
- Record: https://forck.live/items/3267-cohere-and-the-university-of-toronto-partner-to-advance-responsible-ai
- Subject: Command

Cohere and the University of Toronto announced a multi-year partnership to support responsible AI adoption at the university, with Cohere's North platform serving as an orchestration layer for U of T's enterprise-wide AI platform.

### Together AI — What does 99.9% uptime mean for inference?

- Company: Together AI (together.ai)
- Announced: 2026-07-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/99-9-uptime-for-inference
- Record: https://forck.live/items/2334-what-does-99-9-uptime-mean-for-inference
- Subject: Inference platform

Together AI discusses the engineering challenges and requirements for achieving 99%, 99.9%, and 99.99% inference uptime, explaining how each tier demands a specific architecture to survive different failure domains such as node failures, data center failures, and regional outages. The post details their approach to observability, health checks, multi-facility deployment, and infrastructure ownership.

### Hugging Face — Security incident disclosure — July 2026

- Company: Hugging Face (huggingface.co)
- Announced: 2026-07-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/security-incident-july-2026
- Record: https://forck.live/items/1455-security-incident-disclosure-july-2026
- Subject: Platform

Hugging Face disclosed a security incident where an autonomous AI agent system intruded into their production infrastructure, accessed internal datasets and credentials, and was detected and analyzed using AI-assisted tools. The incident did not affect public models or datasets, and they recommend rotating access tokens as a precaution. The source text is a security incident disclosure, not an AI product or model announcement. The only model mentioned (zai-org/GLM-5.2) is referenced in the context of forensic analysis, not as a new model release or update. Therefore, the category is 'other' and no model names are included in models_affected as per the instructions to only list model names that appear in the source text as announcements or updates, which is not the case here.

### OpenAI — How Cars24 scales conversations and builds faster with OpenAI

- Company: OpenAI (openai.com)
- Announced: 2026-07-16
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/cars24
- Record: https://forck.live/items/27-how-cars24-scales-conversations-and-builds-faster-with-openai
- Subject: GPT / ChatGPT / API

Cars24 uses OpenAI-powered voice and chat agents to handle over 1 million monthly conversation minutes, recover 12% of lost leads, and implement agentic workflows across the company.

### NVIDIA — NVIDIA Introduces New Jetson Thor Computers to Advance Mainstream Robotics and Edge AI

- Company: NVIDIA (nvidia.com)
- Announced: 2026-07-15T23:00:54+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://blogs.nvidia.com/blog/jetson-thor-robotics-edge-ai-agent/
- Record: https://forck.live/items/1445-nvidia-introduces-new-jetson-thor-computers-to-advance-mainstream-robotics-and
- Subject: AI platform

NVIDIA introduced the Cosmos 3 Edge model, a 4-billion-parameter lightweight model compatible with NVIDIA Thor platforms, designed for embodied systems to enable real-time vision analysis and on-device robot policy.

### Databricks — Inkling model from Thinking Machines Lab now on Databricks

- Company: Databricks (databricks.com)
- Announced: 2026-07-15T18:30:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.databricks.com/blog/inkling-thinking-machines-lab-now-databricks
- Record: https://forck.live/items/17505-inkling-model-from-thinking-machines-lab-now-on-databricks
- Subject: Mosaic AI

Databricks announced that Inkling, an open-weights model from Thinking Machines Lab, is now available on the Databricks platform through Unity AI Gateway. The model is optimized for coding, agentic reasoning, and multimodal inputs, and can be governed with Unity AI Gateway's security, permissions, cost controls, and observability. Databricks is a day zero launch partner for Thinking Machines Lab.

### Google Research — Towards demystifying the creativity of diffusion models

- Company: Google Research (research.google)
- Announced: 2026-07-15T18:06:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/towards-demystifying-the-creativity-of-diffusion-models/
- Record: https://forck.live/items/1323-towards-demystifying-the-creativity-of-diffusion-models
- Subject: Research

Google Research published a paper showing that the creativity of diffusion models is a mathematical consequence of neural networks learning a smoothed version of the score function, which causes interpolation between training data points.

### Databricks — Data-Native AI Agents: Why Agents Must Move to Your Data

- Company: Databricks (databricks.com)
- Announced: 2026-07-15T17:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.databricks.com/blog/data-native-ai-agents-why-agents-must-move-your-data
- Record: https://forck.live/items/17506-data-native-ai-agents-why-agents-must-move-to-your-data
- Subject: Mosaic AI

Databricks published an architectural argument that AI agents should run inside the data platform rather than in a separate AI stack. The post identifies penalties of external agents—governance gaps, latency, cost fragmentation, and observability loss—and contrasts them with data-native agents that embed policy enforcement into query planning. It describes Databricks' own integrated stack of Unity Catalog, AI Search, MLflow, Lakebase, and AI Gateway as an example of the data-native approach.

### Hugging Face — Model Routing Is Simple. Until It Isn’t.

- Company: Hugging Face (huggingface.co)
- Announced: 2026-07-15T17:27:01+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/ibm-research/model-routing-is-simple-until-it-isnt
- Record: https://forck.live/items/1456-model-routing-is-simple-until-it-isn-t
- Subject: Platform

Challenges in model routing for agentic systems, showing that actual cost depends on caching, difficulty is often invisible at routing time, and latency is affected by infrastructure. The authors propose treating routing as an optimization problem rather than classification, and present results on AppWorld Test Challenge showing a cost-accuracy frontier with configurations achieving 84% accuracy at $93 and 83s latency, a 21% cost reduction and 9% latency reduction compared to running Opus alone with only a 4% accuracy drop.

### Meta — Exploring Hierarchical Interest Representation For Meta Ads Deep Funnel Optimization

- Company: Meta (meta.com)
- Announced: 2026-07-15T17:00:52+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://engineering.fb.com/2026/07/15/ai-research/exploring-hierarchical-interest-representation-for-meta-ads-deep-funnel-optimization/
- Record: https://forck.live/items/1421-exploring-hierarchical-interest-representation-for-meta-ads-deep-funnel
- Subject: Llama / infrastructure

Meta announces a research area called Hierarchical Interest Representation, an upstream representation layer that learns unified embeddings for ads entities (users, advertisers, products, services) using transformer-based graph learning and multimodal knowledge enrichment to improve deep funnel ad optimization across Meta's recommendation ecosystem, including the Generative Ads Model (GEM), Andromeda, and the Adaptive Ranking Model.

### Sakana AI — Sakana AI Teams With NVIDIA to Advance Open Model Innovation from Japan

- Company: Sakana AI (sakana.ai)
- Announced: 2026-07-15T15:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/nvidia-open-model-innovation
- Record: https://forck.live/items/4676-sakana-ai-teams-with-nvidia-to-advance-open-model-innovation-from-japan
- Subject: Sakana models

Sakana AI is collaborating with NVIDIA to integrate NVIDIA's open model stack, including Nemotron, into Sakana Fugu, a multi-agent orchestration system, to advance open model innovation from Japan.

### Microsoft — Azure Databricks delivers proven business value

- Company: Microsoft (microsoft.com)
- Announced: 2026-07-15T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://azure.microsoft.com/en-us/blog/azure-databricks-delivers-proven-business-value/
- Record: https://forck.live/items/2447-azure-databricks-delivers-proven-business-value
- Subject: Azure AI

Microsoft commissioned a Forrester Total Economic Impact study of Azure Databricks, finding a 331% ROI, $58.1M net present value, and less than 6 months payback over three years for a composite organization.

### OpenAI — The US is advancing AI safety through state and federal action

- Company: OpenAI (openai.com)
- Announced: 2026-07-15T12:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/advancing-ai-safety-through-state-and-federal-action
- Record: https://forck.live/items/28-the-us-is-advancing-ai-safety-through-state-and-federal-action
- Subject: GPT / ChatGPT / API

OpenAI proposes a 'reverse federalism' strategy for AI governance, leveraging state laws to create a national framework.

### NVIDIA — NVIDIA and Japan Bring Full-Stack AI and Robotics to Every Industry

- Company: NVIDIA (nvidia.com)
- Announced: 2026-07-15T10:51:37+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blogs.nvidia.com/blog/japan-ecosystem-2026/
- Record: https://forck.live/items/1446-nvidia-and-japan-bring-full-stack-ai-and-robotics-to-every-industry
- Subject: AI platform

NVIDIA and its partners in Japan are showcasing the AI ecosystem's latest advancements, including events like Build-a-Claw and the Physical AI Initiative, with a focus on robotics and AI in manufacturing, but no specific new product, model, or API is announced in the source text provided in this excerpt. The post highlights meetings with supply chain companies and CEO discussions, but does not detail any new product or model launch. The category is 'other' because the text is a general ecosystem update, not a product launch, model release, or other specific announcement type from the taxonomy. The models_affected field is empty because no specific model names are mentioned in the source text. The license, context_window, and pricing_note are left empty as they are not stated. The date is left empty because the source text does not provide a single publication date for the entire blog post. The assessments array includes only the most relevant fields based on the taxonomy and the source text. The assessment for 'new_model' is false because no new model is announced. The assessment for 'model_update' is false because no model update is mentioned. The assessment for 'capability_change' is false because no new capability within an existing product is described. The assessment for 'availability_change' is false because no change in availability (e.g., new regions, plans) is stated. The assessment for 'pricing_change' is false because no pricing change is mentioned. The assessment for 'deprecation' is false because no deprecation is announced. The assessment for 'safety_policy_update' is false because no safety policy change is mentioned. The assessment for 'partnership_acquisition' is false because while partnerships are discussed, no specific new partnership or acquisition is announced; the text describes existing partnerships and meetings. The assessment for 'funding_company_news' is false because no funding news is mentioned. The assessment for 'research_paper' is false because no research paper is announced. The assessment for 'product_launch' is false because no new product is launched. The assessment for 'new_model' is false. The assessment for 'model_update' is false. The assessment for 'open_weight_release' is false because no open-weight release is mentioned. The assessment for 'api_release' is false because no API release is mentioned. The assessment for 'capability_change' is false. The assessment for 'availability_change' is false. The assessment for 'pricing_change' is false. The assessment for 'deprecation' is false. The assessment for 'safety_policy_update' is false. The assessment for 'partnership_acquisition' is false. The assessment for 'funding_company_news' is false. The assessment for 'research_paper' is false. The assessment for 'product_launch' is false. The most specific category applicable is 'other' because the source text is a general ecosystem update without a specific announcement type from the taxonomy. The evidence_excerpt is a verbatim span from the first paragraph of the source text. The summary is a concise description of the content. All other fields are left empty as per the rules. The models_affected array is empty because no model names appear in the source text. The date is left empty because the source text does not provide a single date for the entire blog post; it contains multiple dates for events but no overall publication date. The license, context_window, and pricing_note are empty because they are not stated. The assessments array is empty because none of the specific assessment categories apply. The JSON is constructed according to the schema. The field 'company' is not explicitly in the schema but the schema includes 'company' as a top-level field? The provided schema in the instructions includes: company, title, url, date, category, models_affected, summary, evidence_excerpt, license, context_window, pricing_note, and assessments. So we include company as

### OpenAI — GPT-Red: Unlocking Self-Improvement for Robustness

- Company: OpenAI (openai.com)
- Announced: 2026-07-15T10:00:00+00:00
- Category: capability-change
- Coverage: 1 outlet
- Announcement: yes
- Group: covered
- Source: https://openai.com/index/unlocking-self-improvement-gpt-red
- Record: https://forck.live/items/29-gpt-red-unlocking-self-improvement-for-robustness
- Subject: GPT / ChatGPT / API

OpenAI introduces GPT-Red, an automated red teaming system that uses self-play to improve AI safety, alignment, and prompt injection robustness.

### Lovable — Your Lovable app now works inside ChatGPT and Claude

- Company: Lovable (lovable.dev)
- Announced: 2026-07-15T07:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/agent-integrations
- Record: https://forck.live/items/12070-your-lovable-app-now-works-inside-chatgpt-and-claude
- Subject: Lovable / AI app builder

Lovable announced that apps built on its platform can now be integrated into ChatGPT, Claude, and other AI tools through Model Context Protocol (MCP) servers. Users can enable agent integrations on any publicly published Lovable app, and Lovable will host and maintain the MCP server, allowing AI assistants to access app functionality directly on behalf of authorized users.

### Vercel — How Speechify serves 500,000 dynamic pages to 60 million users on Vercel

- Company: Vercel (vercel.com)
- Announced: 2026-07-15T04:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/how-speechify-serves-50000-dynamic-pages-to-60-million-users-on-vercel
- Record: https://forck.live/items/20686-how-speechify-serves-500-000-dynamic-pages-to-60-million-users-on-vercel
- Subject: v0 / AI SDK / AI Gateway

Copy link to heading Speechify on Vercel 500,000+ pages served across 40+ languages Cut costs 50% by auto-scaling with Fluid compute Zero user impact on bad deploys with Instant Rollbacks Speechify started as a tool for people with dyslexia.

### Thinking Machines Lab — Inkling: Our Open-Weights Model

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2026-07-15
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://thinkingmachines.ai/news/introducing-inkling
- Record: https://forck.live/items/18556-inkling-our-open-weights-model
- Subject: Thinking Machines / Inkling

Thinking Machines Lab released Inkling, an open-weights Mixture-of-Experts transformer with 975B total parameters (41B active), supporting a 1M-token context window and pretrained on 45 trillion tokens of text, images, audio, and video. The model reasons natively over text, images, and audio, supports controllable thinking effort, and is available for fine-tuning on Tinker. A preview of Inkling-Small, a lighter 12B active parameter model, was also shared.

### Gradium — Smaller, clearer, better: Phonon Multilingual beats Magpie and NeuTTS Nano in French, German, and Spanish

- Company: Gradium (gradium.ai)
- Announced: 2026-07-15
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://gradium.ai/blog/phonon-multilingual-smaller-clearer-better
- Record: https://forck.live/items/18311-smaller-clearer-better-phonon-multilingual-beats-magpie-and-neutts-nano-in
- Subject: Gradium TTS / Phonon

Gradium's 100M-parameter on-device TTS model Phonon produces up to 3.5x fewer word errors than NVIDIA's Magpie TTS in French, German, and Spanish, at 3.6x fewer parameters, and leads other on-device voice-cloning models on word error rate and speaker similarity in every tested language. Phonon clones a reference voice across English, French, German, Spanish, and Portuguese.

### LG AI Research — [CVPR 2026] Beyond Text: How VinQA Incorporates Visual Elements into Responses for Multimodal Document QA

- Company: LG AI Research (lgresearch.ai)
- Announced: 2026-07-15
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/blog/view?seq=661
- Record: https://forck.live/items/4742-cvpr-2026-beyond-text-how-vinqa-incorporates-visual-elements-into-responses
- Subject: EXAONE

LG AI Research presents VinQA, a multimodal document QA dataset and evaluation framework presented at CVPR 2026. VinQA focuses on generating long-form answers that strategically position grounding visual elements within text, rather than appending images. The dataset includes ~130,000 pages and 42,700 QAs across seven domains, with single-page, multi-page, multi-document, and unanswerable questions. The study introduces M-GroSE, a multimodal evaluation metric. Experiments show that fine-tuning Qwen2.5-VL-7B on VinQA significantly improves its performance, approaching commercial models like GPT-4.1 and Claude 3.5 Sonnet.

### xAI — Grok Build is Now Open Source

- Company: xAI (x.ai)
- Announced: 2026-07-15
- Category: developer-tool-release
- Coverage: 1 outlet
- Announcement: yes
- Group: covered
- Source: https://x.ai/news/grok-build-open-source
- Record: https://forck.live/items/3443-grok-build-is-now-open-source
- Subject: Grok

xAI has open-sourced Grok Build, the harness behind their coding agent and TUI.

### Cohere — The total cost of AI ownership

- Company: Cohere (cohere.com)
- Announced: 2026-07-15
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cohere.com/blog/the-total-cost-of-ai-ownership
- Record: https://forck.live/items/3270-the-total-cost-of-ai-ownership
- Subject: Command

### Together AI — New in Together GPU Clusters: Reliability and control for production GPU clusters

- Company: Together AI (together.ai)
- Announced: 2026-07-15
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/new-in-together-gpu-clusters-reliability-and-control-for-production-gpu-clusters
- Record: https://forck.live/items/2336-new-in-together-gpu-clusters-reliability-and-control-for-production-gpu-clusters
- Subject: Inference platform

Together AI announces new reliability and operational control features for Together GPU Clusters, including passive health checks, auto node repair, and a rebuilt Slurm-on-Kubernetes stack (Slinky 1.0) to improve failure detection, automated recovery, and cluster management for production workloads.

### Together AI — Together AI brings Thinking Machines Lab’s new model Inkling on day 0

- Company: Together AI (together.ai)
- Announced: 2026-07-15
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.together.ai/blog/together-ai-brings-thinking-machines-labs-new-model-inkling-on-day-0
- Record: https://forck.live/items/2335-together-ai-brings-thinking-machines-lab-s-new-model-inkling-on-day-0
- Subject: Inference platform

Thinking Machines Lab released Inkling, a new multimodal mixture-of-experts model that accepts text, image, and audio inputs and produces text outputs. It features controllable reasoning effort, a 975B total parameters with 40B active per token, 1M token context window, and architectural innovations including query-conditioned relative attention, short causal convolutions, and a shared expert sink. Together AI offers access through serverless and OpenAI-compatible APIs.

### Hugging Face — Introducing Real World VoiceEQ: Measuring the human quality of voice AI

- Company: Hugging Face (huggingface.co)
- Announced: 2026-07-15
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/real-world-voiceeq
- Record: https://forck.live/items/1458-introducing-real-world-voiceeq-measuring-the-human-quality-of-voice-ai
- Subject: Platform

Hugging Face and HumeAI introduce Real World VoiceEQ, a benchmark for evaluating the human quality of voice AI across 40+ models, 15+ dimensions, and 60+ metrics, based on over 1 million human ratings.

### Hugging Face — Welcome Inkling by Thinking Machines

- Company: Hugging Face (huggingface.co)
- Announced: 2026-07-15
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/thinkingmachines-inkling
- Record: https://forck.live/items/1457-welcome-inkling-by-thinking-machines
- Subject: Platform

Thinking Machines Lab released Inkling, a large open multimodal model with ~1T parameters and 1M context window, natively accepting image, text, and audio inputs. It also released Inkling-Small with 276B total and 12B active parameters, and Inkling-Small-NVFP4 variant. The models are available on Hugging Face with day-0 support in transformers, SGLang, vLLM, and llama.cpp.

### Google DeepMind — Reconstructing Pelé’s “lost” goal

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-07-14T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/reconstructing-peles-lost-goal/
- Record: https://forck.live/items/1302-reconstructing-pel-s-lost-goal
- Subject: Gemini

Google DeepMind released a mini-documentary showing how its AI technology reconstructed Pelé's 1959 lost goal.

### Harvey — Harvey Opens Chicago Office

- Company: Harvey (harvey.ai)
- Announced: 2026-07-14T12:45:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/harvey-opens-chicago-office
- Record: https://forck.live/items/17543-harvey-opens-chicago-office
- Subject: Harvey

Harvey opened an office in Chicago to be closer to law firms and legal teams in the region, building on its existing work with firms such as DLA Piper, Dentons US, and Foley & Lardner, as well as in-house teams at Avant, CCC Intelligent Solutions, and DRW.

### OpenAI — How to manage AI investments in the agentic era

- Company: OpenAI (openai.com)
- Announced: 2026-07-14T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/managing-ai-investments-in-agentic-era
- Record: https://forck.live/items/30-how-to-manage-ai-investments-in-the-agentic-era
- Subject: GPT / ChatGPT / API

This is a guide for enterprises on managing AI investments in the agentic era, focusing on measuring useful work per dollar, improving efficiency, and scaling high-value workflows.

### Inception — Mercury 2: the first reasoning model fast enough to pick up the phone

- Company: Inception (inceptionlabs.ai)
- Announced: 2026-07-14
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.inceptionlabs.ai/blog/mercury-2-the-first-reasoning-model-fast-enough-to-pick-up-the-phone
- Record: https://forck.live/items/18519-mercury-2-the-first-reasoning-model-fast-enough-to-pick-up-the-phone
- Subject: Mercury

Inception Labs released Mercury 2, described as the world's first reasoning diffusion language model, which decodes over 1,000 tokens per second on standard NVIDIA GPUs. The model is designed for realtime voice agent applications, reducing reasoning latency from seconds to under 300 milliseconds. It offers a reasoning_effort knob with four settings and, according to the company, beats GPT 4.1 on instruction-following benchmarks while being faster.

### Manus — Introducing PowerPoint Slides on Manus

- Company: Manus (manus.im)
- Announced: 2026-07-14
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://manus.im/blog/manus-ppt-slides
- Record: https://forck.live/items/14706-introducing-powerpoint-slides-on-manus
- Subject: Manus

Manus now generates native .pptx files directly, with editable charts, structured tables, and AI-generated visuals, instead of converting from a web format. The feature is available in Beta for users on the 1.6 and Max models.

### Suno — sad alex on songwriting, short-form, and creative autonomy

- Company: Suno (suno.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://suno.com/blog/sad-alex
- Record: https://forck.live/items/10318-sad-alex-on-songwriting-short-form-and-creative-autonomy
- Subject: Music / Studio

Suno published an interview with singer-songwriter sad alex discussing how she uses Suno as a creative tool for songwriting, including solving specific production problems like switching vocal genders or sourcing instrument parts, while emphasizing the importance of maintaining creative autonomy and human authorship in her work.

### Amp — Secrets of the Orb

- Company: Amp (ampcode.com)
- Announced: 2026-07-14
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/secrets-of-the-orb
- Record: https://forck.live/items/7729-secrets-of-the-orb
- Subject: Amp

Amp announced that orbs can now prove their identity using OIDC, enabling service providers to trust Amp's identity and grant permissions without injecting secrets. The post includes examples and commands for minting tokens.

### Cognition — One Year of Building Together

- Company: Cognition (cognition.com)
- Announced: 2026-07-14
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://cognition.com/blog/one-year-of-building-together
- Record: https://forck.live/items/7645-one-year-of-building-together
- Subject: Devin / SWE

Cognition released SWE-1.7, described as the most capable and efficient model it has trained to date, and launched Devin Desktop, unifying its products under one brand. The company also introduced Devin Review, Devin CLI, and other updates.

### OpenAI — Sales workflows with ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/chatgpt-work/how-sales-teams-use-codex
- Record: https://forck.live/items/4145-sales-workflows-with-chatgpt-work
- Subject: GPT / ChatGPT / API

OpenAI published a guide on using ChatGPT Work for sales workflows, including a sales plugin that integrates with tools like Salesforce and HubSpot.

### OpenAI — Data science workflows with ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/chatgpt-work/how-data-science-teams-use-codex
- Record: https://forck.live/items/4144-data-science-workflows-with-chatgpt-work
- Subject: GPT / ChatGPT / API

OpenAI Academy published a guide on how data science teams can use ChatGPT Work to turn dashboards, metric definitions, and raw data into review-ready analysis assets, noting that the workflows were previously in the Codex app.

### OpenAI — How sales teams use ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/chatgpt-work/how-sales-teams-use-codex
- Record: https://forck.live/items/3713-how-sales-teams-use-chatgpt-work
- Subject: GPT / ChatGPT / API

OpenAI Academy published a guide on how sales teams can use ChatGPT Work to create pipeline briefs, meeting packs, and other sales artifacts, with a sales plugin that integrates with CRM and communication tools like Salesforce, HubSpot, and Slack.

### OpenAI — How data science teams use ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/chatgpt-work/how-data-science-teams-use-codex
- Record: https://forck.live/items/3712-how-data-science-teams-use-chatgpt-work
- Subject: GPT / ChatGPT / API

OpenAI Academy published a guide on how data science teams can use ChatGPT Work to turn questions, dashboards, and raw data into review-ready analysis assets.

### Cohere — Tiny Aya in the wild: What people are building with open multilingual AI

- Company: Cohere (cohere.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cohere.com/blog/tiny-aya-in-the-wild
- Record: https://forck.live/items/3271-tiny-aya-in-the-wild-what-people-are-building-with-open-multilingual-ai
- Subject: Command

What people are building with the Tiny Aya open multilingual AI model.

### Anthropic — Anthropic commits $10 million to Canadian AI research

- Company: Anthropic (anthropic.com)
- Announced: 2026-07-14
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.anthropic.com/news/canadian-ai-research
- Record: https://forck.live/items/3261-anthropic-commits-10-million-to-canadian-ai-research
- Subject: Claude

### Anthropic — Introducing Claude for Teachers

- Company: Anthropic (anthropic.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: 3 outlets
- Announcement: no
- Group: covered
- Source: https://www.anthropic.com/news/claude-for-teachers
- Record: https://forck.live/items/3260-introducing-claude-for-teachers
- Subject: Claude

Anthropic introduced Claude for Teachers.

### OpenAI — How sales teams use ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/codex-for-work/how-sales-teams-use-codex
- Record: https://forck.live/items/32-how-sales-teams-use-chatgpt-work
- Subject: GPT / ChatGPT / API

Sales teams can use ChatGPT Work to create various sales-related documents from real work inputs.

### OpenAI — How data science teams use ChatGPT Work

- Company: OpenAI (openai.com)
- Announced: 2026-07-14
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/academy/codex-for-work/how-data-science-teams-use-codex
- Record: https://forck.live/items/31-how-data-science-teams-use-chatgpt-work
- Subject: GPT / ChatGPT / API

Data science teams can use ChatGPT Work to create various work outputs such as root-cause briefs, impact readouts, KPI memos, scoped analyses, and dashboard specs from real work inputs.

### Meta — Modernizing the Meta Ads Service With an Open-Source Kernel Scheduler

- Company: Meta (meta.com)
- Announced: 2026-07-13T16:00:50+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://engineering.fb.com/2026/07/13/ml-applications/modernizing-the-meta-ads-service-with-an-open-source-kernel-scheduler/
- Record: https://forck.live/items/1422-modernizing-the-meta-ads-service-with-an-open-source-kernel-scheduler
- Subject: Llama / infrastructure

Meta used sched_ext to build a custom scheduling policy for the Ads service, resulting in a 28% reduction in ads retrieval stage tail latency, 3.28 MW power savings, and a 1.1% increase in ads ranked.

### Microsoft Research — Verifying Rust cryptography in SymCrypt, from standards to code

- Company: Microsoft Research (microsoft.com)
- Announced: 2026-07-13T16:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.microsoft.com/en-us/research/blog/verifying-rust-cryptography-in-symcrypt-from-standards-to-code/
- Record: https://forck.live/items/2458-verifying-rust-cryptography-in-symcrypt-from-standards-to-code
- Subject: Research / Phi

Microsoft Research announces the release of formally verified cryptographic code in SymCrypt, using Rust, Lean, Aeneas, and AI agents for proof automation, initially for SHA-3 and ML-KEM.

### Google DeepMind — Empowering India’s next generation of innovators with ATL Saathi

- Company: Google DeepMind (deepmind.google)
- Announced: 2026-07-13T12:37:28+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://deepmind.google/blog/empowering-indias-next-generation-of-innovators-with-atl-saathi/
- Record: https://forck.live/items/1207-empowering-india-s-next-generation-of-innovators-with-atl-saathi
- Subject: Gemini

Google and AIM launched ATL Saathi, a Gemini-powered AI tool for Indian educators in robotics labs.

### fal — From clay 3d render to a real action short: a 3D-to-AI pipeline with total control

- Company: fal (fal.ai)
- Announced: 2026-07-13T12:03:25+00:00
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://blog.fal.ai/from-clay-3d-render-to-a-real-action-short-a-3d-to-ai-pipeline-with-total-control/
- Record: https://forck.live/items/7992-from-clay-3d-render-to-a-real-action-short-a-3d-to-ai-pipeline-with-total
- Subject: fal platform

fal released the open-source 3DREAL Strong v2 adapter, a LoRA for LTX 2.3 that turns grey 3D blockouts into photoreal footage with exact composition and camera control. The adapter is available on Hugging Face and a hosted endpoint runs it.

### Meta — Teachers and Local Businesses Win as Meta Expands Louisiana Data Center

- Company: Meta (meta.com)
- Announced: 2026-07-13T09:30:24+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://about.fb.com/news/2026/07/teachers-local-businesses-win-as-meta-expands-louisiana-data-center/
- Record: https://forck.live/items/9919-teachers-and-local-businesses-win-as-meta-expands-louisiana-data-center
- Subject: Meta AI

Meta is expanding its Louisiana data center with an investment of more than $50 billion in the Richland Parish region, one of the largest investments in AI infrastructure in the world. The expansion includes a $5 million donation to Louisiana Delta Community College for scholarships, teacher bonuses, support for local businesses, and an energy agreement expected to save Entergy Louisiana customers more than $2 billion over 20 years.

### Vercel — Open-weight models surge to 29% of volume, price per token flattens

- Company: Vercel (vercel.com)
- Announced: 2026-07-13T07:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/ai-gateway-production-index-july-2026
- Record: https://forck.live/items/20687-open-weight-models-surge-to-29-of-volume-price-per-token-flattens
- Subject: v0 / AI SDK / AI Gateway

Vercel's AI Gateway Production Index for July 2026 reports that gateway token volume grew 29% month over month and spend grew 27%, while the price per token was flat after rising almost 20% in May. Open-weight models ran 29% of gateway tokens, up from 11% in April, on under 4% of spend. DeepSeek reached 22.6% of token volume, in third place and less than two points behind Google, and GLM 5.2 broke into the gateway's top models by volume two weeks after release.

### Ai2 — What building Shippy taught us about building agents

- Company: Ai2 (allenai.org)
- Announced: 2026-07-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://allenai.org/blog/shippy-deep-dive
- Record: https://forck.live/items/18545-what-building-shippy-taught-us-about-building-agents
- Subject: Olmo / Molmo / Asta

Ai2's Skylight team published a technical deep-dive on Shippy, an AI agent for maritime domain awareness. The post describes Shippy's architecture, which separates a system prompt ("soul"), markdown-based skills, and configuration, and explains how a purpose-built CLI reduces API complexity to improve reliability. The team shares lessons applicable to Ai2's other environmental platforms.

### Genspark — How a Finance Intern Used Genspark AI to Turn Public Records Into Real Assets Opportunities

- Company: Genspark (genspark.ai)
- Announced: 2026-07-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.genspark.ai/blog/how-to-use-genspark-for-finance
- Record: https://forck.live/items/18251-how-a-finance-intern-used-genspark-ai-to-turn-public-records-into-real-assets
- Subject: Genspark

### Cua — Computer-Use 2.0 at AI Engineer World's Fair

- Company: Cua (cua.ai)
- Announced: 2026-07-13
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cua.ai/blog/computer-use-2-ai-engineer-worlds-fair
- Record: https://forck.live/items/16818-computer-use-2-0-at-ai-engineer-world-s-fair
- Subject: Cua Driver / Lume / Cua Bench / Cua Fleets

Cua's talk at AI Engineer World's Fair presented Computer-Use 2.0, describing a shift in which computer-use becomes a tool inside coding agents like Claude Code and Codex, and general agents like Hermes, OpenClaw, and Pi. The session covered three connected layers: Cua Driver for consistent desktop-window operation, Cua-Bench for evaluating desktop-task completion, and Cua Fleets for provisioning and scaling computers for agent workloads.

### Manus — Introducing Auto-Publish: Build Once, Ship Continuously

- Company: Manus (manus.im)
- Announced: 2026-07-13
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://manus.im/blog/manus-auto-publish
- Record: https://forck.live/items/14707-introducing-auto-publish-build-once-ship-continuously
- Subject: Manus

Manus introduced Auto-Publish, a toggle in the publish popover that automatically deploys every successful build to the user's public URL. The feature is off by default, available to all users on web, iOS, and Android starting today, and can be turned off at any time. It is designed for iteration sessions where users have already decided to ship, and it can be combined with queued messages to deploy changes without user presence.

### Cognition — Devin is Now FedRAMP High In-Process, Unlocking Autonomous AI Engineering for Federal Agencies

- Company: Cognition (cognition.com)
- Announced: 2026-07-13
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/devin-fedramp-high-in-process
- Record: https://forck.live/items/7647-devin-is-now-fedramp-high-in-process-unlocking-autonomous-ai-engineering-for
- Subject: Devin / SWE

Cognition's entire platform, including Devin Cloud, has achieved FedRAMP High In-Process status and is listed on the FedRAMP Marketplace, enabling federal agencies such as the U.S. Army, Navy, and NASA JPL to use the autonomous AI software engineer for tasks like security vulnerability remediation, legacy code modernization, and cloud migration.

### Cognition — Making Fable Cheaper Than Opus

- Company: Cognition (cognition.com)
- Announced: 2026-07-13
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://cognition.com/blog/making-fable-cheaper-than-opus
- Record: https://forck.live/items/7646-making-fable-cheaper-than-opus
- Subject: Devin / SWE

Cognition replaced Opus 4.8 with Fable 5 in Devin, and despite Fable costing twice as much per token, the overall bill decreased when using the Fusion architecture with a sidekick. In 3,000 evaluation sessions on FrontierCode 1.1, Fable + Sidekick cost $1.86 per run versus $2.04 for Opus + Sidekick, while scoring higher (60.7 vs 54.6). The cost savings come from Fable's management style: it delegates more effectively, taking fewer turns and consuming less context.