# forck.live — 8–14 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-W37, 8–14 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-W37
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Ant Ling — Segmentation-as-Editing for Unified Multimodal AI

- Company: Ant Ling (ant-ling.com)
- Announced: 2025-09-13
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://developer.ant-ling.com/en/blogs/ming-lite-omni-1_5-seg
- Record: https://forck.live/items/22339-segmentation-as-editing-for-unified-multimodal-ai
- Subject: Ling / Ring / Ming

Ant Ling's Ming-lite-omni 1.5 model reframes segmentation as a color-editing task, improving its cIoU score on RefCOCO-val from 16% to 72.4%. The approach converts mask generation into image-to-image editing, aligning the task with generative model strengths. The model is open-weight and available on GitHub, Hugging Face, and ModelScope.

### Harvey — 5 Questions With Maggie Landers

- Company: Harvey (harvey.ai)
- Announced: 2025-09-12T14:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/5-questions-with-maggie-landers
- Record: https://forck.live/items/17675-5-questions-with-maggie-landers
- Subject: Harvey

Our VP of Talent shares what Harvey looks for in candidates and why now is the right moment to join. “5 Questions With” is our blog series where we interview senior leaders at Harvey about the things our customers and partners care about most.

### Vercel — Introducing x402-mcp: Open protocol payments for MCP tools

- Company: Vercel (vercel.com)
- Announced: 2025-09-12T13:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/introducing-x402-mcp-open-protocol-payments-for-mcp-tools
- Record: https://forck.live/items/20823-introducing-x402-mcp-open-protocol-payments-for-mcp-tools
- Subject: v0 / AI SDK / AI Gateway

Vercel introduced x402-mcp, an open protocol that integrates payments into HTTP requests using the 402 Payment Required status code, enabling programmatic payments for MCP servers and the Vercel AI SDK without prior account setup. The protocol is payment-network agnostic and currently settles as exact, one-time payments in USDC on the Base blockchain. Vercel also released an x402 AI Starter template demonstrating the protocol with Next.js, AI SDK, AI Elements, and AI Gateway.

### OpenAI — Working with US CAISI and UK AISI to build more secure AI systems

- Company: OpenAI (openai.com)
- Announced: 2025-09-12T12:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/us-caisi-uk-aisi-ai-update
- Record: https://forck.live/items/473-working-with-us-caisi-and-uk-aisi-to-build-more-secure-ai-systems
- Subject: GPT / ChatGPT / API

OpenAI shares progress on the partnership with the US CAISI and UK AISI to strengthen AI safety and security.

### Upstage — Upstage Information Extract Gets Location Coordinates Beta

- Company: Upstage (upstage.ai)
- Announced: 2025-09-12
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.upstage.ai/blog/en/information-extract-location-coordinates-beta
- Record: https://forck.live/items/18464-upstage-information-extract-gets-location-coordinates-beta
- Subject: Solar Pro / Solar Mini

Upstage added a Location Coordinates (Beta) feature to its Information Extract service, returning the page number and bounding-box coordinates for each extracted value. The feature is intended to support audit trails, compliance reviews, and contract verification by linking extracted data to its source position in the document.

### Luma — Clio Awards

- Company: Luma (lumalabs.ai)
- Announced: 2025-09-12
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lumalabs.ai/news/clio-awards
- Record: https://forck.live/items/7981-clio-awards
- Subject: Ray / Dream Machine

Luma AI and the Clio Awards announced a partnership to launch the "AI Challenge – Powered by Dream Machine," a creative brief for junior to mid-level creatives and recent graduates to submit 10-second vertical ads built with Dream Machine. Three finalists will be recognized at the Clio Creative Summit in New York City on October 20, 2025. Submissions are open through October 6.

### Baidu — ERNIE 4.5 Gets a Major Inference Speed Boost

- Company: Baidu (baidu.com)
- Announced: 2025-09-12
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ernie.baidu.com/blog/posts/plas/
- Record: https://forck.live/items/4653-ernie-4-5-gets-a-major-inference-speed-boost
- Subject: ERNIE

Baidu ERNIE 4.5 models get a major inference speed boost using a new sparse attention mechanism called PLAS, available for -Paddle versions deployed with FastDeploy.

### Hugging Face — Introducing the Palmyra-mini family: Powerful, lightweight, and ready to reason!

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-11T20:04:44+00:00
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/Writer/announcing-palmyra-mini
- Record: https://forck.live/items/1623-introducing-the-palmyra-mini-family-powerful-lightweight-and-ready-to-reason
- Subject: Platform

Writer released three new open models in the Palmyra-mini family: palmyra-mini (base), palmyra-mini-thinking-a (optimized for complex reasoning), and palmyra-mini-thinking-b (optimized for mathematical reasoning). The models have 1.5B to 1.7B parameters and are available with GGUF and MLX quantizations.

### OpenAI — A joint statement from OpenAI and Microsoft

- Company: OpenAI (openai.com)
- Announced: 2025-09-11T14:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/joint-statement-from-openai-and-microsoft
- Record: https://forck.live/items/475-a-joint-statement-from-openai-and-microsoft
- Subject: GPT / ChatGPT / API

OpenAI and Microsoft announced a new memorandum of understanding that strengthens their partnership and emphasizes a shared commitment to AI safety and innovation.

### OpenAI — Statement on OpenAI’s Nonprofit and PBC

- Company: OpenAI (openai.com)
- Announced: 2025-09-11T14:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/statement-on-openai-nonprofit-and-pbc
- Record: https://forck.live/items/474-statement-on-openai-s-nonprofit-and-pbc
- Subject: GPT / ChatGPT / API

OpenAI announces a new structure that grants equity in its Public Benefit Corporation (PBC), enabling over $100 billion in resources to advance safe and beneficial AI.

### Harvey — Enabling Document-Wide Edits in Harvey’s Word Add-In

- Company: Harvey (harvey.ai)
- Announced: 2025-09-11T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/enabling-document-wide-edits-in-harveys-word-add-in
- Record: https://forck.live/items/17676-enabling-document-wide-edits-in-harvey-s-word-add-in
- Subject: Harvey

Harvey expanded its Word Add-In to support document-wide edits on 100+ page documents with a single query, using an orchestrator-subagent architecture to overcome position bias and ensure thorough coverage. The system separates concerns by translating OOXML to natural language for model reasoning, then deterministically applying edits back to preserve styles and structure.

### Hugging Face — Tricks from OpenAI gpt-oss YOU 🫵 can use with transformers

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-11
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/faster-transformers
- Record: https://forck.live/items/1624-tricks-from-openai-gpt-oss-you-can-use-with-transformers
- Subject: Platform

Hugging Face announces upgrades to the transformers library, including zero-build kernels downloadable from the Hub, MXFP4 quantization, tensor parallelism, expert parallelism, dynamic sliding window layer and cache, continuous batching and paged attention, and faster model loading, inspired by OpenAI's GPT-OSS release. These features are designed to work across major models in transformers.

### Replit — Introducing Agent 3: Our Most Autonomous Agent Yet

- Company: Replit (replit.com)
- Announced: 2025-09-10T22:58:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replit.com/blog/introducing-agent-3-our-most-autonomous-agent-yet
- Record: https://forck.live/items/17940-introducing-agent-3-our-most-autonomous-agent-yet
- Subject: Replit Agent

Replit introduced Agent 3, an upgrade to its AI coding agent that can periodically test apps in a browser and automatically fix issues. The agent is claimed to be 10x more autonomous, 3x faster, and 10x more cost-effective than Computer Use models. Agent 3 can also generate other agents and automations, and run autonomously for up to 200 minutes. The new app creation flow offers both full-stack and frontend-only modes.

### Together AI — Fine-tune Any LLM from the Hugging Face Hub with Together AI

- Company: Together AI (together.ai)
- Announced: 2025-09-10T17:04:36+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/togethercomputer/together-ft
- Record: https://forck.live/items/1625-fine-tune-any-llm-from-the-hugging-face-hub-with-together-ai

Hugging Face and Together AI announce a new integration that allows users to fine-tune any compatible LLM from the Hugging Face Hub using Together AI's infrastructure, specifying a base model for configuration and the custom Hugging Face model to fine-tune, with bidirectional support for pulling and pushing models.

### Stability AI — Stability AI Introduces Stable Audio 2.5, the First Audio Model Built for Enterprise Sound Production at Scale

- Company: Stability AI (stability.ai)
- Announced: 2025-09-10T14:07:07+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://stability.ai/news-updates/stability-ai-introduces-stable-audio-25-the-first-audio-model-built-for-enterprise-sound-production-at-scale
- Record: https://forck.live/items/17483-stability-ai-introduces-stable-audio-2-5-the-first-audio-model-built-for
- Subject: Stable Diffusion

Stability AI released Stable Audio 2.5, an audio generation model designed for enterprise-grade sound production. The model generates tracks up to three minutes long with inference under two seconds on a GPU, supports audio inpainting for custom audio inputs, and is optimized for improved musical structure and prompt adherence. It is available via the Stability AI API, partner platforms, and on-premises enterprise licensing.

### Vercel — MongoDB Atlas is now available on the Vercel Marketplace

- Company: Vercel (vercel.com)
- Announced: 2025-09-10T13:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/mongodb-atlas-is-now-available-on-the-vercel-marketplace
- Record: https://forck.live/items/20824-mongodb-atlas-is-now-available-on-the-vercel-marketplace
- Subject: v0 / AI SDK / AI Gateway

Vercel added MongoDB Atlas to its Marketplace, allowing developers to provision a managed MongoDB database directly from the Vercel dashboard with automatic environment variable configuration and integrated billing.

### Harvey — Preview: Access Harvey, Wherever You’re Working

- Company: Harvey (harvey.ai)
- Announced: 2025-09-10T12:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/mobile-voice-announcement
- Record: https://forck.live/items/17677-preview-access-harvey-wherever-you-re-working
- Subject: Harvey

Harvey announced upcoming iOS and Android mobile apps that will extend its AI-powered legal assistant to mobile devices. The apps will include voice prompting, file upload via photo, and access to Assistant and Vault. Voice to Prompt is already available in the web app, with mobile launch planned for November.

### Thinking Machines Lab — Defeating Nondeterminism in LLM Inference

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2025-09-10T07:01:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://thinkingmachines.ai/blog/defeating-nondeterminism-in-llm-inference/
- Record: https://forck.live/items/18571-defeating-nondeterminism-in-llm-inference
- Subject: Thinking Machines / Inkling

Thinking Machines Lab published a technical analysis arguing that the common 'concurrency + floating point' hypothesis does not fully explain nondeterminism in LLM inference. The post identifies floating-point non-associativity as the root cause of numerical differences and explains that while GPU kernels are deterministic, inference server results appear nondeterministic to users. It also describes how to achieve reproducible results in LLM inference.

### Genspark — Genspark AI Browser and World''s First On-Device Free AI

- Company: Genspark (genspark.ai)
- Announced: 2025-09-10
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.genspark.ai/blog/genspark-ai-browser-and-on-device-free-ai
- Record: https://forck.live/items/18268-genspark-ai-browser-and-world-s-first-on-device-free-ai
- Subject: Genspark

### Together AI — Fine-Tuning Platform Upgrades: Larger Models, Longer Contexts, Enhanced Hugging Face Integrations

- Company: Together AI (together.ai)
- Announced: 2025-09-10
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.together.ai/blog/fine-tuning-updates-sept-2025
- Record: https://forck.live/items/2410-fine-tuning-platform-upgrades-larger-models-longer-contexts-enhanced-hugging
- Subject: Inference platform

Together AI upgraded its fine-tuning platform, adding support for over a dozen large models (including DeepSeek, Qwen, Meta, and OpenAI's gpt-oss), extending context lengths by 2x-4x (up to 131k tokens), and integrating with Hugging Face Hub for model fine-tuning.

### Together AI — Together AI welcomes Mahadev Konar as SVP for Infrastructure Engineering

- Company: Together AI (together.ai)
- Announced: 2025-09-10
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/mahadev-konar-svp-infrastructure-engineering
- Record: https://forck.live/items/2409-together-ai-welcomes-mahadev-konar-as-svp-for-infrastructure-engineering
- Subject: Inference platform

Together AI announces Mahadev Konar as SVP for Infrastructure Engineering.

### Hugging Face — Jupyter Agents: training LLMs to reason with notebooks

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-10
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/jupyter-agent-2
- Record: https://forck.live/items/1626-jupyter-agents-training-llms-to-reason-with-notebooks
- Subject: Platform

Hugging Face introduces Jupyter Agents, a project to train small LLMs to reason with notebooks for data science tasks. They fine-tune Qwen3-4B on Kaggle notebooks and evaluate on DABStep benchmark, achieving improved accuracy after scaffolding optimization.

### Supabase — OrioleDB Patent: now freely available to the Postgres community

- Company: Supabase (supabase.com)
- Announced: 2025-09-09T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/orioledb-patent-free
- Record: https://forck.live/items/21334-orioledb-patent-now-freely-available-to-the-postgres-community
- Subject: Supabase

Supabase announced that it has completed the legal transfer of OrioleDB assets and now fully owns U.S. Patent 10,325,030. The company is making a non-exclusive license of that patent available to all OrioleDB users, including proprietary forks. OrioleDB's license has been changed to Apache 2.0, and the project will continue as an open-source effort with an open contribution model.

### Vercel — A more flexible Pro plan for modern teams

- Company: Vercel (vercel.com)
- Announced: 2025-09-09T13:00:00+00:00
- Category: pricing-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/new-pro-pricing-plan
- Record: https://forck.live/items/20826-a-more-flexible-pro-plan-for-modern-teams
- Subject: v0 / AI SDK / AI Gateway

Vercel is updating its Pro plan to a flexible credit-based usage model with $20 in monthly flexible credit, $150+ for Fast Data Transfer, and $20+ for Edge Requests. The plan adds free Viewer seats for non-deploying teammates and self-serve access to Enterprise features like SAML SSO and HIPAA BAAs. Around 7% of teams with diverse usage will see bill increases, while over 100,000 paying teams will see a decrease or no change.

### Vercel — The second wave of MCP: Building for LLMs, not developers

- Company: Vercel (vercel.com)
- Announced: 2025-09-09T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/the-second-wave-of-mcp-building-for-llms-not-developers
- Record: https://forck.live/items/20825-the-second-wave-of-mcp-building-for-llms-not-developers
- Subject: v0 / AI SDK / AI Gateway

Vercel's post argues that MCP (Model Context Protocol) tools should be designed around complete user intentions rather than mirroring individual API endpoints, because LLMs lack the context and state management that developers use to chain API calls. It recommends creating single workflow tools, such as a deploy_project tool that handles the entire deployment pipeline internally, instead of exposing separate tools for each API operation.

### OpenAI — SafetyKit scales risk agents with OpenAI’s most capable models

- Company: OpenAI (openai.com)
- Announced: 2025-09-09T10:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/safetykit
- Record: https://forck.live/items/476-safetykit-scales-risk-agents-with-openai-s-most-capable-models
- Subject: GPT / ChatGPT / API

SafetyKit uses OpenAI GPT-5 to improve content moderation, compliance, and accuracy compared to legacy safety systems.

### Mistral AI — Mistral raises 1.7B€ to accelerate technological progress with AI

- Company: Mistral AI (mistral.ai)
- Announced: 2025-09-09T07:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/mistral-ai-raises-1-7-b-to-accelerate-technological-progress-with-ai/
- Record: https://forck.live/items/19886-mistral-raises-1-7b-to-accelerate-technological-progress-with-ai
- Subject: Mistral / Le Chat

Mistral AI announced a €1.7 billion fundraising round to accelerate technological progress with AI.

### Amp — Streaming JSON

- Company: Amp (ampcode.com)
- Announced: 2025-09-09
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/streaming-json
- Record: https://forck.live/items/7805-streaming-json
- Subject: Amp

The Amp CLI added two new flags, --stream-json and --stream-json-input, for structured JSON output and input when used with the --execute flag.

### Mistral AI — Mistral AI raises 1.7B€ to accelerate technological progress with AI

- Company: Mistral AI (mistral.ai)
- Announced: 2025-09-09
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://mistral.ai/news/mistral-ai-raises-1-7-b-to-accelerate-technological-progress-with-ai
- Record: https://forck.live/items/3312-mistral-ai-raises-1-7b-to-accelerate-technological-progress-with-ai
- Subject: Mistral / Le Chat

Mistral AI raises 1.7 billion euros to accelerate technological progress with AI.

### Together AI — Announcing General Availability of Together Instant Clusters, offering ready to use, self-service NVIDIA GPUs

- Company: Together AI (together.ai)
- Announced: 2025-09-09
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/together-instant-clusters-ga
- Record: https://forck.live/items/2411-announcing-general-availability-of-together-instant-clusters-offering-ready-to
- Subject: Inference platform

Together AI announces the general availability of Together Instant Clusters, an API-first, self-service GPU cluster provisioning service that allows users to quickly spin up multi-node clusters with NVIDIA GPUs for training and inference.

### Hugging Face — mmBERT: ModernBERT goes Multilingual

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-09
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/mmbert
- Record: https://forck.live/items/1627-mmbert-modernbert-goes-multilingual
- Subject: Platform

Hugging Face introduces mmBERT, a massively multilingual encoder model trained on over 3 trillion tokens across more than 1800 languages, built upon the ModernBERT architecture with novel training techniques.

### OpenAI — A People-First AI Fund: $50M to support nonprofits

- Company: OpenAI (openai.com)
- Announced: 2025-09-08T14:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/people-first-ai-fund
- Record: https://forck.live/items/477-a-people-first-ai-fund-50m-to-support-nonprofits
- Subject: GPT / ChatGPT / API

OpenAI has opened applications for a $50 million fund to support U.S. nonprofits in education, community innovation, and economic opportunity, with unrestricted grants and an October 8, 2025 deadline.

### Vercel — Critical npm supply chain attack response – September 8, 2025

- Company: Vercel (vercel.com)
- Announced: 2025-09-08T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/critical-npm-supply-chain-attack-response-september-8-2025
- Record: https://forck.live/items/20827-critical-npm-supply-chain-attack-response-september-8-2025
- Subject: v0 / AI SDK / AI Gateway

Vercel reported a supply chain attack on September 8, 2025, that compromised 18 popular npm packages including chalk, debug, and ansi-styles. The injected code intercepted cryptocurrency transactions in browsers. Vercel identified 70 teams with 76 affected projects, purged build caches, and notified customers; no Vercel customers were impacted by a subsequent DuckDB-related incident.

### Harvey — Harvey Reaches 100 Global Legal Knowledge Sources

- Company: Harvey (harvey.ai)
- Announced: 2025-09-08T13:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/100-knowledge-sources-available-in-harvey
- Record: https://forck.live/items/17678-harvey-reaches-100-global-legal-knowledge-sources
- Subject: Harvey

Harvey announced it now integrates 100 national legal and regulatory data sources, covering six continents and supranational bodies such as the EU and UN. The sources are organized into regional bundles, and customers can select the regions they need. The company noted that over half its customers are outside the US and that it has expanded office presence in several countries.

### LG AI Research — LG Aimers 7th Challenge: A Journey to Becoming an Expert

- Company: LG AI Research (lgresearch.ai)
- Announced: 2025-09-08
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/news/view?seq=582
- Record: https://forck.live/items/11000-lg-aimers-7th-challenge-a-journey-to-becoming-an-expert
- Subject: EXAONE

LG AI Research concluded the 7th season of its LG Aimers AI education program with a hackathon where 102 participants developed demand forecasting models for Konjiam Resort. The winning team, Um-nya, combined machine learning and deep learning models to create a robust baseline optimized for tabular data. A job fair with four LG affiliates was also held during the event.

### Cognition — Funding, growth, and the next frontier of AI coding agents

- Company: Cognition (cognition.com)
- Announced: 2025-09-08
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/funding-growth-and-the-next-frontier-of-ai-coding-agents
- Record: https://forck.live/items/7692-funding-growth-and-the-next-frontier-of-ai-coding-agents
- Subject: Devin / SWE

Cognition raised over $400M at a $10.2B valuation led by Founders Fund. The company acquired Windsurf, more than doubling its ARR, and combined enterprise ARR grew over 30% in the seven weeks post-acquisition. Devin ARR grew from $1M to $73M between September 2024 and June 2025.

### Replicate — Torch compile caching for inference speed

- Company: Replicate (replicate.com)
- Announced: 2025-09-08
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replicate.com/blog/torch-compile-caching
- Record: https://forck.live/items/2479-torch-compile-caching-for-inference-speed
- Subject: Platform

Replicate announces caching of torch.compile artifacts to reduce boot times for models using PyTorch, with specific models starting 2-3x faster and cold boot times reduced by 50-62%.