# forck.live — 22–28 September 2025

> 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 2025-W39, 22–28 September 2025, 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/2025-W39
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### OpenAI — Partnering with AARP to help keep older adults safe online

- Company: OpenAI (openai.com)
- Announced: 2025-09-26T06:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/aarp-partnership-older-adults-online-safety
- Record: https://forck.live/items/456-partnering-with-aarp-to-help-keep-older-adults-safe-online
- Subject: GPT / ChatGPT / API

OpenAI and AARP are partnering to provide AI training, scam-spotting tools, and nationwide programs for older adults through OpenAI Academy and OATS's Senior Planet initiative.

### Thinking Machines Lab — Modular Manifolds

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2025-09-26
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://thinkingmachines.ai/blog/modular-manifolds/
- Record: https://forck.live/items/18570-modular-manifolds
- Subject: Thinking Machines / Inkling

Thinking Machines Lab proposes constraining neural network weight matrices to submanifolds, such as the Stiefel manifold, to improve training stability and enable co-designed optimization algorithms. The post introduces a manifold version of the Muon optimizer and defines the concept of a modular manifold for composable constraints. The work is presented as an introduction to a research area with many directions for future exploration.

### Hugging Face — Swift Transformers Reaches 1.0 – and Looks to the Future

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-26
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/swift-transformers
- Record: https://forck.live/items/1614-swift-transformers-reaches-1-0-and-looks-to-the-future
- Subject: Platform

Hugging Face released version 1.0 of swift-transformers, a Swift library for local LLM inference on Apple Silicon. The release marks API stability, introduces modular Tokenizers and Hub components, adopts modern Core ML APIs, removes dependencies, and signals future focus on MLX and agentic use cases.

### Vercel — Preventing the stampede: Request collapsing in the Vercel CDN

- Company: Vercel (vercel.com)
- Announced: 2025-09-25T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/cdn-request-collapsing
- Record: https://forck.live/items/20814-preventing-the-stampede-request-collapsing-in-the-vercel-cdn
- Subject: v0 / AI SDK / AI Gateway

Vercel added request collapsing to its CDN to prevent cache stampedes on ISR pages. When multiple requests hit the same uncached path, only one function invocation per region runs; other requests wait for the cached response. The system uses a two-level distributed lock (node-level and regional) and automatically determines cacheability through framework integration.

### OpenAI — More ways to work with your team and tools in ChatGPT

- Company: OpenAI (openai.com)
- Announced: 2025-09-25T11:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/more-ways-to-work-with-your-team
- Record: https://forck.live/items/457-more-ways-to-work-with-your-team-and-tools-in-chatgpt
- Subject: GPT / ChatGPT / API

OpenAI announces new shared projects, smarter connectors, and compliance and security updates for ChatGPT to help teams get more done.

### Google Research — Towards better health conversations: Research insights on a “wayfinding” AI agent based on Gemini

- Company: Google Research (research.google)
- Announced: 2025-09-25T10:33:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/towards-better-health-conversations-research-insights-on-a-wayfinding-ai-agent-based-on-gemini/
- Record: https://forck.live/items/1414-towards-better-health-conversations-research-insights-on-a-wayfinding-ai-agent
- Subject: Research

Google Research shares user insights from a research prototype called Wayfinding AI, based on Gemini, which proactively asks clarifying questions to help users find better health information. Studies showed users found it more helpful, relevant, and tailored than a baseline AI.

### OpenAI — Measuring the performance of our models on real-world tasks

- Company: OpenAI (openai.com)
- Announced: 2025-09-25T09:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://openai.com/index/gdpval
- Record: https://forck.live/items/458-measuring-the-performance-of-our-models-on-real-world-tasks
- Subject: GPT / ChatGPT / API

OpenAI announces GDPval, a new evaluation benchmark for measuring model performance on economically valuable tasks across 44 occupations.

### Upstage — The cracks MGAs can’t ignore: how AI Space reinforces the foundation for MGA growth

- Company: Upstage (upstage.ai)
- Announced: 2025-09-25
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.upstage.ai/blog/en/the-cracks-mgas-cant-ignore-how-ai-space-reinforces-the-foundation-for-mga-growth
- Record: https://forck.live/items/18463-the-cracks-mgas-can-t-ignore-how-ai-space-reinforces-the-foundation-for-mga
- Subject: Solar Pro / Solar Mini

Upstage introduced AI Space, a platform for managing insurance submissions from managing general agents (MGAs). The platform uses Upstage's proprietary Information Extract technology to read, extract, and organize data from multiple file formats including PDFs, Excel, Word documents, and screenshots. Upstage reports that MGAs using AI Space have achieved accuracy rates averaging 96% and reduced submission preparation time by hours per account.

### Factory AI — Factory Raises $50M Series B

- Company: Factory AI (factory.com)
- Announced: 2025-09-25
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/series-b
- Record: https://forck.live/items/17870-factory-raises-50m-series-b
- Subject: Droid

Factory announced a $50M Series B funding round at a $300M valuation, led by NEA, Sequoia Capital, J.P. Morgan, Nvidia, and others. The company's Droids, which reached #1 on Terminal Bench, are now available to anyone with any model in any interface. Enterprise customers including MongoDB, EY, Bayer, Zapier, and Clari report results such as 31x faster feature delivery and 96.1% shorter migration times.

### Factory AI — Droid: The #1 Software Development Agent on Terminal-Bench

- Company: Factory AI (factory.com)
- Announced: 2025-09-25
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://factory.com/news/terminal-bench
- Record: https://forck.live/items/17831-droid-the-1-software-development-agent-on-terminal-bench
- Subject: Droid

Factory AI's Droid agent achieved a 58.75% score on the Terminal-Bench benchmark, placing first among all agents. Factory agents occupy three of the top five positions, with Droid using Opus 4.1 (no thinking) at 58.8%, GPT-5 (medium reasoning) at 52.5%, and Sonnet 4 (no thinking) at 50.5%. The company attributes the performance to agent design rather than model choice, noting that Droid with Sonnet outperforms all other agents using Opus.

### Cua — What happens when hackathon judging is a public benchmark (Hack the North edition)

- Company: Cua (cua.ai)
- Announced: 2025-09-25
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://cua.ai/blog/hack-the-north
- Record: https://forck.live/items/16831-what-happens-when-hackathon-judging-is-a-public-benchmark-hack-the-north-edition
- Subject: Cua Driver / Lume / Cua Bench / Cua Fleets

Cua and HUD ran a computer-use agents track at Hack the North, scoring entries on a public benchmark and offering the top performer a guaranteed YC interview. The post shares the playbook for running or sponsoring such a track.

### Suno — Introducing Suno Studio

- Company: Suno (suno.com)
- Announced: 2025-09-25
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://suno.com/blog/suno-studio
- Record: https://forck.live/items/10330-introducing-suno-studio
- Subject: Music / Studio

Suno announced Suno Studio, a generative audio workstation that integrates AI-powered music generation with traditional digital audio workstation features. The tool enables users to generate unlimited stem variations, edit multitrack timelines, adjust BPM and pitch, and export stems as audio and MIDI files for use in other DAWs.

### xAI — Expanding xAI for Government with GSA OneGov

- Company: xAI (x.ai)
- Announced: 2025-09-25
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.ai/news/onegov
- Record: https://forck.live/items/3484-expanding-xai-for-government-with-gsa-onegov
- Subject: Grok

xAI is expanding its 'xAI For Government' offering in partnership with GSA OneGov to provide more accessible AI tools for the Federal Government.

### OpenAI — Introducing ChatGPT Pulse

- Company: OpenAI (openai.com)
- Announced: 2025-09-25
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-chatgpt-pulse
- Record: https://forck.live/items/459-introducing-chatgpt-pulse
- Subject: GPT / ChatGPT / API

OpenAI announced a preview of ChatGPT Pulse, a new experience that proactively delivers personalized updates based on user chats, feedback, and connected apps, initially available to Pro users on mobile.

### Google Research — AfriMed-QA: Benchmarking large language models for global health

- Company: Google Research (research.google)
- Announced: 2025-09-24T19:11:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/afrimed-qa-benchmarking-large-language-models-for-global-health/
- Record: https://forck.live/items/1415-afrimed-qa-benchmarking-large-language-models-for-global-health
- Subject: Research

Google Research presents AfriMed-QA, a benchmark question-answer dataset for evaluating LLMs on African health tasks, developed with partners across Africa. The dataset includes ~15,000 questions from 60 medical schools across 16 countries. The paper was published at ACL 2025 and won the Best Social Impact Paper Award. The dataset and evaluation code are open-sourced.

### OpenAI — ENEOS Materials brings ChatGPT Enterprise to manufacturing

- Company: OpenAI (openai.com)
- Announced: 2025-09-24T17:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/eneos-materials
- Record: https://forck.live/items/460-eneos-materials-brings-chatgpt-enterprise-to-manufacturing
- Subject: GPT / ChatGPT / API

ENEOS Materials uses ChatGPT Enterprise to speed research, improve plant design safety, and cut HR analysis time by 90%, with 80% of users reporting better workflows.

### Lovable — Introducing Lovable File Uploads

- Company: Lovable (lovable.dev)
- Announced: 2025-09-24T15:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/lovable-file-uploads
- Record: https://forck.live/items/12131-introducing-lovable-file-uploads
- Subject: Lovable / AI app builder

Lovable introduced file uploads, allowing users to drop files such as CSV, XLSX, PDF, PPTX, MP4, MP3, TTF, and OTF directly into the platform to generate apps and websites. The feature reads the file and uses it as context with the user's prompt.

### Sakana AI — ShinkaEvolve: Evolving New Algorithms with LLMs, Orders of Magnitude More Efficiently

- Company: Sakana AI (sakana.ai)
- Announced: 2025-09-24T15:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/shinka-evolve
- Record: https://forck.live/items/4708-shinkaevolve-evolving-new-algorithms-with-llms-orders-of-magnitude-more
- Subject: Sakana models

Sakana AI introduces ShinkaEvolve, an open-source evolutionary code optimization framework that uses LLMs to discover new algorithms with state-of-the-art performance and unprecedented sample efficiency, tested across four domains including circle packing, math reasoning, competitive programming, and MoE training.

### Harvey — Blue Owl Partners with Harvey on New AI Workflows for the Alternative Asset Management Industry

- Company: Harvey (harvey.ai)
- Announced: 2025-09-24T12:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/blue-owl-partners-with-harvey
- Record: https://forck.live/items/17672-blue-owl-partners-with-harvey-on-new-ai-workflows-for-the-alternative-asset
- Subject: Harvey

Harvey announced a multi-year partnership with Blue Owl Capital, under which Blue Owl signed an enterprise license to deploy Harvey's platform firm-wide and co-develop AI workflows for alternative asset management. The collaboration aims to improve efficiency and client service, and will involve Harvey in Blue Owl's internal training programs.

### Runway — Autoregressive-to-Diffusion Vision Language Models

- Company: Runway (runway.com)
- Announced: 2025-09-24T10:10:10.110000+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/research/autoregressive-to-diffusion-vlms
- Record: https://forck.live/items/7902-autoregressive-to-diffusion-vision-language-models
- Subject: Gen / Aleph

Runway presents A2D-VL 7B, a diffusion vision language model adapted from the pretrained autoregressive VLM Qwen2.5-VL for parallel diffusion decoding. The model uses block size annealing and noise level annealing to preserve base model capabilities while enabling a speed-quality trade-off through adjustable parallelism.

### OpenAI — SAP and OpenAI partner to launch sovereign ‘OpenAI for Germany’

- Company: OpenAI (openai.com)
- Announced: 2025-09-24T04:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/global-affairs/openai-for-germany
- Record: https://forck.live/items/461-sap-and-openai-partner-to-launch-sovereign-openai-for-germany
- Subject: GPT / ChatGPT / API

SAP and OpenAI announce a partnership to launch OpenAI for Germany in 2026, targeting secure and sovereign AI for the public sector in Germany.

### Cartesia — Cartesia achieves GDPR compliance

- Company: Cartesia (cartesia.ai)
- Announced: 2025-09-24
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.cartesia.ai/blog/gdpr-compliance
- Record: https://forck.live/items/8014-cartesia-achieves-gdpr-compliance
- Subject: Sonic / Ink

Cartesia announced that its text-to-speech platform is now GDPR compliant, reflecting its commitment to data protection and privacy.

### Google Research — Time series foundation models can be few-shot learners

- Company: Google Research (research.google)
- Announced: 2025-09-23T18:00:43+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/time-series-foundation-models-can-be-few-shot-learners/
- Record: https://forck.live/items/1416-time-series-foundation-models-can-be-few-shot-learners
- Subject: Research

Google Research presents TimesFM-ICF, a time-series foundation model that uses continued pre-training to learn from in-context examples at inference time, enabling few-shot forecasting without supervised fine-tuning. The work was presented at ICML 2025.

### OpenAI — OpenAI, Oracle, and SoftBank expand Stargate with five new AI datacenter sites

- Company: OpenAI (openai.com)
- Announced: 2025-09-23T14:00:00+00:00
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/five-new-stargate-sites
- Record: https://forck.live/items/462-openai-oracle-and-softbank-expand-stargate-with-five-new-ai-datacenter-sites
- Subject: GPT / ChatGPT / API

OpenAI, Oracle, and SoftBank announced five new Stargate AI datacenter sites, accelerating a $500 billion, 10-gigawatt infrastructure buildout to power next-generation AI and create tens of thousands of jobs.

### Perceptron — TensorStream

- Company: Perceptron (perceptron.inc)
- Announced: 2025-09-23
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.perceptron.inc/blog/tensorstream
- Record: https://forck.live/items/18621-tensorstream
- Subject: Perceptron Isaac / Perceptron Mk1

Perceptron introduced TensorStream, a data structure and library for organizing multimodal data (e.g., video, audio, metadata) into a causal token stream for generative causal transformers. The design uses Events tagged with modality and time descriptors, then schedules them into a sorted sequence based on a user-specified priority list. The library also handles token accounting by tracking real and virtual dimensions and index ranges through preprocessing and embedding.

### Amp — Globs in AGENTS.md

- Company: Amp (ampcode.com)
- Announced: 2025-09-23
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/globs-in-AGENTS.md
- Record: https://forck.live/items/7804-globs-in-agents-md
- Subject: Amp

Amp now allows AGENTS.md files to include other context files dynamically using glob mentions and filters, enabling more granular guidance for the agent.

### Amp — Amp Tab for All

- Company: Amp (ampcode.com)
- Announced: 2025-09-23
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/amp-tab-for-all
- Record: https://forck.live/items/7803-amp-tab-for-all
- Subject: Amp

Amp has turned on Amp Tab by default for all new Amp extension installs. Amp Tab is a free in-editor completion engine for VS Code, Windsurf, and Cursor that is fast, understands diagnostic errors, and can suggest updates across multiple files. Existing users can enable it via the command palette.

### LG AI Research — Opening the Next Chapter with New Leadership, an Interview with Honglak Lee and Woohyung Lim, Co-Heads of LG AI Research

- Company: LG AI Research (lgresearch.ai)
- Announced: 2025-09-23
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/blog/view?seq=584
- Record: https://forck.live/items/4778-opening-the-next-chapter-with-new-leadership-an-interview-with-honglak-lee-and
- Subject: EXAONE

LG AI Research announces new co-heads Honglak Lee and Woohyung Lim, who discuss their vision for Interactive AI, including Agentic AI and Physical AI, and three core technology pillars: World Models, autonomous decision-making, and AI platform technology. The post emphasizes human-centered AI and future research directions.

### Replicate — Which image editing model should I use?

- Company: Replicate (replicate.com)
- Announced: 2025-09-23
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://replicate.com/blog/compare-image-editing-models
- Record: https://forck.live/items/2477-which-image-editing-model-should-i-use
- Subject: Platform

Replicate published a blog post comparing several image editing models across tasks like object removal, front view generation, background editing, text editing, and style transfer, evaluating models from Black Forest Labs, Alibaba, Google, ByteDance, and OpenAI.

### Hugging Face — Smol2Operator: Post-Training GUI Agents for Computer Use

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-23
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/smol2operator
- Record: https://forck.live/items/1615-smol2operator-post-training-gui-agents-for-computer-use
- Subject: Platform

Hugging Face releases Smol2Operator, a post-training recipe and model that transforms SmolVLM2-2.2B-Instruct into a GUI agent for computer use, along with training recipes, data-processing tools, datasets, and a demo.

### Alibaba — Qwen3Guard: Real-time Safety for Your Token Stream

- Company: Alibaba (alibaba.com)
- Announced: 2025-09-22T20:00:00+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://qwenlm.github.io/blog/qwen3guard/
- Record: https://forck.live/items/2281-qwen3guard-real-time-safety-for-your-token-stream
- Subject: Qwen

Alibaba announces Qwen3Guard, the first safety guardrail model in the Qwen family, built on Qwen3 foundation models and fine-tuned for safety classification.

### OpenAI — CNA is transforming its newsroom with AI

- Company: OpenAI (openai.com)
- Announced: 2025-09-22T17:17:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/cna-walter-fernandez
- Record: https://forck.live/items/463-cna-is-transforming-its-newsroom-with-ai
- Subject: GPT / ChatGPT / API

OpenAI features CNA's transformation of its newsroom with AI, including insights from Editor-in-Chief Walter Fernandez on AI adoption, culture, and the future of journalism.

### Vercel — BotID uncovers hidden SEO poisoning

- Company: Vercel (vercel.com)
- Announced: 2025-09-22T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/botid-uncovers-hidden-seo-poisoning
- Record: https://forck.live/items/20815-botid-uncovers-hidden-seo-poisoning
- Subject: v0 / AI SDK / AI Gateway

Vercel's blog describes how a financial institution used BotID to discover that suspicious bot traffic was actually verified Google crawlers re-crawling indexed pages from a years-old SEO poisoning attack on their previous infrastructure, not an active threat.

### Harvey — Harvey Expands European Presence With Team in Spain

- Company: Harvey (harvey.ai)
- Announced: 2025-09-22T11:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/harvey-expands-european-presence-with-team-in-spain
- Record: https://forck.live/items/17673-harvey-expands-european-presence-with-team-in-spain
- Subject: Harvey

Harvey announced the launch of a dedicated team in Spain to support legal professionals across the Iberian market, including law firms, mid-sized practices, and in-house legal teams. The Harvey platform now includes embedded access to Spanish legislation, with Portuguese data coming soon. The expansion is part of Harvey's broader global growth strategy to meet rising demand for AI solutions in professional services throughout Europe.

### OpenAI — SchoolAI builds an AI platform that empowers teachers

- Company: OpenAI (openai.com)
- Announced: 2025-09-22T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/schoolai
- Record: https://forck.live/items/464-schoolai-builds-an-ai-platform-that-empowers-teachers
- Subject: GPT / ChatGPT / API

SchoolAI uses GPT-4.1, image generation, and TTS to deliver AI tools for teachers in over 1 million classrooms.

### OpenAI — OpenAI and NVIDIA announce strategic partnership to deploy 10 gigawatts of NVIDIA systems

- Company: OpenAI (openai.com)
- Announced: 2025-09-22T08:45:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/openai-nvidia-systems-partnership
- Record: https://forck.live/items/465-openai-and-nvidia-announce-strategic-partnership-to-deploy-10-gigawatts-of
- Subject: GPT / ChatGPT / API

OpenAI and NVIDIA announce a strategic partnership to deploy 10 gigawatts of AI datacenters powered by NVIDIA systems, with the first phase launching in 2026.

### Hugging Face — SyGra: The One-Stop Framework for Building Data for LLMs and SLMs

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-22T06:45:05+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/ServiceNow-AI/sygra-data-gen-framework
- Record: https://forck.live/items/1616-sygra-the-one-stop-framework-for-building-data-for-llms-and-slms
- Subject: Platform

ServiceNow AI announces SyGra, a low-code/no-code framework for dataset creation, transformation, and alignment for LLMs and SLMs, supporting multiple inference backends and flexible data generation.

### Genspark — Genspark Photo Genius and Voice-Powered AI Photo Magic

- Company: Genspark (genspark.ai)
- Announced: 2025-09-22
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.genspark.ai/blog/genspark-photo-genius
- Record: https://forck.live/items/18267-genspark-photo-genius-and-voice-powered-ai-photo-magic
- Subject: Genspark

### LG AI Research — How AI Sees Through Companies

- Company: LG AI Research (lgresearch.ai)
- Announced: 2025-09-22
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/blog/view?seq=586
- Record: https://forck.live/items/4779-how-ai-sees-through-companies
- Subject: EXAONE

LG AI Research's Data Intelligence Lab has developed EXAONE BI, a solution powered by the proprietary corporate analysis agent GanphA, which provides AEFS scores and daily commentary for U.S.-listed equities, and highlights the performance of the LQAI ETF.

### DeepSeek — DeepSeek-V3.1 is now DeepSeek-V3.1-Terminus

- Company: DeepSeek (deepseek.com)
- Announced: 2025-09-22
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.deepseek.com/en/news/v3-1-terminus
- Record: https://forck.live/items/4322-deepseek-v3-1-is-now-deepseek-v3-1-terminus
- Subject: DeepSeek models

DeepSeek updated its V3.1 model to V3.1-Terminus, improving language consistency, agent performance, and output stability.

### Hugging Face — Gaia2 and ARE: Empowering the community to study agents

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-22
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/gaia2
- Record: https://forck.live/items/1617-gaia2-and-are-empowering-the-community-to-study-agents
- Subject: Platform

Hugging Face introduces Gaia2, a follow-up to the GAIA benchmark for evaluating AI agents, and the open Meta Agents Research Environments (ARE) framework for running, debugging, and evaluating agents. Gaia2 includes tasks such as execution, search, ambiguity handling, adaptability, temporal reasoning, agent-to-agent collaboration, and noise tolerance. The dataset is released under CC by 4.0, and ARE under MIT license. Results comparing various models are provided.