# forck.live — 29 September – 5 October 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-W40, 29 September – 5 October 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-W40
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Supabase — Supabase Series E

- Company: Supabase (supabase.com)
- Announced: 2025-10-03T19:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/supabase-series-e
- Record: https://forck.live/items/21325-supabase-series-e
- Subject: Supabase

Supabase raised $100 million in a Series E round led by Accel and Peak XV, at a $5 billion pre-money valuation, with participation from Figma and existing investors. The company plans to use the funds for employee secondaries and to accelerate hiring for open source initiatives like Multigres and OrioleDB. A $1 million community round for early customers and contributors is planned in the coming months.

### Supabase — Announcing the Supabase Remote MCP Server

- Company: Supabase (supabase.com)
- Announced: 2025-10-03T17:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/remote-mcp-server
- Record: https://forck.live/items/21328-announcing-the-supabase-remote-mcp-server
- Subject: Supabase

Supabase launched a remote MCP server supporting HTTP transport, OAuth2 authentication, and new tools including feature groups, doc search, and security/performance advisors. The remote server allows web-based AI agents like ChatGPT, Claude, and Builder.io to connect to Supabase projects via a single URL, replacing the previous stdio-based setup that required Node.js and manual tokens.

### Supabase — gm 👋 web3, welcome aboard to Sign in with Web3 (Solana, Ethereum)

- Company: Supabase (supabase.com)
- Announced: 2025-10-03T17:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/login-with-solana-ethereum
- Record: https://forck.live/items/21327-gm-web3-welcome-aboard-to-sign-in-with-web3-solana-ethereum
- Subject: Supabase

Supabase added support for Web3 logins using Ethereum and Solana, based on the Sign in with Ethereum standard (EIP-4361). The feature is available through the Supabase JavaScript SDK and can be configured via the Supabase Dashboard or CLI.

### Supabase — 1000 Y Combinator Founders Choose Supabase

- Company: Supabase (supabase.com)
- Announced: 2025-10-03T17:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://supabase.com/blog/1000-yc-companies
- Record: https://forck.live/items/21326-1000-y-combinator-founders-choose-supabase
- Subject: Supabase

Supabase announced that over 1,000 Y Combinator companies now use its platform, with 55% of the most recent YC batch choosing Supabase. The post highlights several YC-backed startups that built on Supabase and includes testimonials from founders.

### Sakana AI — Sakana AI and Daiwa Securities Group to Develop AI for Advanced Asset Consulting

- Company: Sakana AI (sakana.ai)
- Announced: 2025-10-03T13:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://sakana.ai/daiwa-securities
- Record: https://forck.live/items/4707-sakana-ai-and-daiwa-securities-group-to-develop-ai-for-advanced-asset-consulting
- Subject: Sakana models

Sakana AI and Daiwa Securities Group announced a long-term partnership to co-develop a 'Total Asset Consulting Platform' powered by Sakana AI's models, aiming to provide personalized financial advice and innovate Japan's asset management industry.

### Amp — Amp TypeScript SDK

- Company: Amp (ampcode.com)
- Announced: 2025-10-03
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/typescript-sdk
- Record: https://forck.live/items/7799-amp-typescript-sdk
- Subject: Amp

Amp released a TypeScript SDK that enables programmatic use of the Amp agent in TypeScript programs.

### Replit — Bring your Vercel project to Replit (in minutes)

- Company: Replit (replit.com)
- Announced: 2025-10-02T18:12:22+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://replit.com/blog/how-to-migrate-from-vercel
- Record: https://forck.live/items/17936-bring-your-vercel-project-to-replit-in-minutes
- Subject: Replit Agent

Replit published a guide showing how to import a Vercel project into Replit using its Agent. The process involves linking a GitHub repo, providing secrets and environment variables, and the Agent automatically configures the app with a live preview and logs.

### Google Research — A collaborative approach to image generation

- Company: Google Research (research.google)
- Announced: 2025-10-02T17:04:05+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/a-collaborative-approach-to-image-generation/
- Record: https://forck.live/items/1410-a-collaborative-approach-to-image-generation
- Subject: Research

Google Research introduces PASTA, a reinforcement learning agent that collaborates with users over multiple turns to refine text-to-image outputs by learning user preferences, enabled by a novel user simulation technique. They also release a foundational dataset of over 7,000 human rater interactions.

### Harvey — The Brief: October

- Company: Harvey (harvey.ai)
- Announced: 2025-10-02T15:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/the-brief-october-2025
- Record: https://forck.live/items/17667-the-brief-october
- Subject: Harvey

Harvey announced platform-wide performance improvements including up to 10x faster Vault uploads, 53% quicker processing, 89% lower upload latency, 6x fewer document processing errors, and 3x faster Excel processing. Vault project limits increased to 100GB or 100,000 files. New features for Harvey for Word include table filling via natural language, Magic Prompt and Prompt Library in Ask Mode, query history, and improved Playbook review. Additional capabilities include voice input, email attachment querying, iManage integration, and streamlined admin controls with opt-in feature management and updated workflow builder permissions.

### OpenAI — With GPT-5, Wrtn builds lifestyle AI for millions in Korea

- Company: OpenAI (openai.com)
- Announced: 2025-10-02T10:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/wrtn
- Record: https://forck.live/items/434-with-gpt-5-wrtn-builds-lifestyle-ai-for-millions-in-korea
- Subject: GPT / ChatGPT / API

Wrtn used GPT-5 to build a 'Lifestyle AI' app that blends productivity, creativity, and learning, reaching 6.5 million users in Korea and expanding across East Asia.

### Upstage — The responsiveness revolution: how AI Space helps MGAs deliver what brokers need

- Company: Upstage (upstage.ai)
- Announced: 2025-10-02
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.upstage.ai/blog/en/the-responsiveness-revolution-how-ai-space-helps-mgas-deliver-what-brokers-need
- Record: https://forck.live/items/18462-the-responsiveness-revolution-how-ai-space-helps-mgas-deliver-what-brokers-need
- Subject: Solar Pro / Solar Mini

Upstage's AI Space helps MGAs process insurance submissions in a single day instead of weeks by extracting, categorizing, and preparing risk information automatically, eliminating manual back-and-forth and legacy email chains.

### LG AI Research — [Paper Review] WebDancer: Towards Autonomous Information Seeking Agency

- Company: LG AI Research (lgresearch.ai)
- Announced: 2025-10-02
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.lgresearch.ai/blog/view?seq=588
- Record: https://forck.live/items/4777-paper-review-webdancer-towards-autonomous-information-seeking-agency
- Subject: EXAONE

LG AI Research reviews the WebDancer paper from Alibaba's Tongyi Lab, which proposes enhancing long reasoning in Large Reasoning Models (LRMs) by using web search tools. The paper introduces synthetic data generation methods CRAWLQA and E2HQA to help models learn exploration-based reasoning and knowledge supplementation via search.

### Replicate — IBM's Granite 4.0 is now on Replicate

- Company: Replicate (replicate.com)
- Announced: 2025-10-02
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replicate.com/blog/2025-10-02-ibm-granite-4-models
- Record: https://forck.live/items/2476-ibm-s-granite-4-0-is-now-on-replicate
- Subject: Platform

IBM's Granite 4.0 family of open-source small language models is now available on Replicate, featuring a hybrid Mamba-2/Transformer architecture, Apache 2.0 license, and a 30B parameter instruct model.

### Hugging Face — SOTA OCR with Core ML and dots.ocr

- Company: Hugging Face (huggingface.co)
- Announced: 2025-10-02
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/dots-ocr-ne
- Record: https://forck.live/items/1609-sota-ocr-with-core-ml-and-dots-ocr
- Subject: Platform

Hugging Face posts a guide explaining how to convert the 3B parameter OCR model dots.ocr to run on Apple devices using Core ML and MLX, highlighting its on-device performance and power efficiency compared to cloud-based alternatives.

### OpenAI — OpenAI announces strategic collaboration with Japan’s Digital Agency

- Company: OpenAI (openai.com)
- Announced: 2025-10-02
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/global-affairs/strategic-collaboration-with-japan-digital-agency
- Record: https://forck.live/items/435-openai-announces-strategic-collaboration-with-japan-s-digital-agency
- Subject: GPT / ChatGPT / API

OpenAI and Japan's Digital Agency announced a partnership to advance generative AI in public services, support international AI governance, and promote safe, trustworthy AI adoption worldwide.

### Google Research — Introducing interactive on-device segmentation in Snapseed

- Company: Google Research (research.google)
- Announced: 2025-10-01T17:05:30+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/introducing-interactive-on-device-segmentation-in-snapseed/
- Record: https://forck.live/items/1411-introducing-interactive-on-device-segmentation-in-snapseed
- Subject: Research

Google Research announces the Object Brush feature in Snapseed on iOS, powered by an on-device Interactive Segmenter model, enabling real-time object selection and editing via simple strokes.

### Lovable — How Lumoo used Lovable to transform the fashion industry from the inside out

- Company: Lovable (lovable.dev)
- Announced: 2025-10-01T14:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/lumoo-ai-fashion-platform
- Record: https://forck.live/items/12128-how-lumoo-used-lovable-to-transform-the-fashion-industry-from-the-inside-out
- Subject: Lovable / AI app builder

Lumoo, a fashion and retail platform built entirely on Lovable, integrates AI-native content creation tools including virtual try-on, design visualization, and localization for fashion brands. The platform has reached €700k ARR in 9 months since launch in February 2025 with 15+ customer brands including Gant and AWNR Group, and aims to reduce production costs and return rates through accurate product representation.

### Replit — Introducing Connectors

- Company: Replit (replit.com)
- Announced: 2025-10-01T04:38:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replit.com/blog/connectors
- Record: https://forck.live/items/17938-introducing-connectors
- Subject: Replit Agent

Replit launched Connectors, a feature that lets users build apps and automations integrating with external services such as Google, Dropbox, Salesforce, Linear, and Notion. Users connect a service once via one-click login, and apps can then use it without managing credentials. The launch includes first-party support for over 20 connectors, with more planned.

### Replit — Replit acquires OpenInt

- Company: Replit (replit.com)
- Announced: 2025-10-01T04:38:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://replit.com/blog/replit-acquires-openint
- Record: https://forck.live/items/17937-replit-acquires-openint
- Subject: Replit Agent

Replit has acquired OpenInt, an integration platform (iPaaS) for code generation that supports over 160 connectors. OpenInt will serve as the foundation for Replit Connectors, enabling the Replit Agent to handle complex customer integrations. The OpenInt Cloud service will wind down on December 31, 2025, and existing customers are being migrated to a self-hosted solution.

### OpenAI — Samsung and SK join OpenAI’s Stargate initiative to advance global AI infrastructure

- Company: OpenAI (openai.com)
- Announced: 2025-10-01T03:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/samsung-and-sk-join-stargate
- Record: https://forck.live/items/436-samsung-and-sk-join-openai-s-stargate-initiative-to-advance-global-ai
- Subject: GPT / ChatGPT / API

Samsung and SK join OpenAI's Stargate initiative to expand global AI infrastructure, scaling advanced memory chip production and building next-gen data centers in Korea.

### Ant Ling — Ming-UniVision: Joint Image Understanding and Generation via a Unified Continuous Tokenizer

- Company: Ant Ling (ant-ling.com)
- Announced: 2025-10-01
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://developer.ant-ling.com/en/blogs/mingtok
- Record: https://forck.live/items/22338-ming-univision-joint-image-understanding-and-generation-via-a-unified
- Subject: Ling / Ring / Ming

Ant Ling introduced Ming-UniVision, an autoregressive vision-language model built on the MingTok tokenizer, which is the first visual tokenizer based on a continuous latent space. The unified representation supports image understanding, generation, and editing within a single next-token prediction framework, achieving over 3.5× faster training convergence compared to traditional approaches.

### Ant Ling — Ming-UniAudio: Speech LLM for Joint Understanding, Generation and Editing with Unified Representation

- Company: Ant Ling (ant-ling.com)
- Announced: 2025-10-01
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://developer.ant-ling.com/en/blogs/ming-uniaudio
- Record: https://forck.live/items/22337-ming-uniaudio-speech-llm-for-joint-understanding-generation-and-editing-with
- Subject: Ling / Ring / Ming

Ant Ling introduced Ming-UniAudio, an end-to-end unified speech language model with a single LLM backbone for both understanding and generation tasks, enhanced with a Diffusion Head for high-fidelity speech synthesis. It uses MingTok-Audio, a unified continuous speech tokenizer based on a VAE framework with a causal Transformer architecture, claimed to be the first continuous speech tokenizer to integrate semantic and acoustic features. The release also includes Ming-Freeform-Audio-Edit, an instruction-guided free-form speech editing framework, and Audio-Edit-Benchmark, an open-source evaluation set for free-form speech editing tasks.

### Thinking Machines Lab — Announcing Tinker

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2025-10-01
- Category: api-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://thinkingmachines.ai/news/announcing-tinker
- Record: https://forck.live/items/18564-announcing-tinker
- Subject: Thinking Machines / Inkling

Thinking Machines Lab launched Tinker, a flexible API for fine-tuning open-weight language models, including large mixture-of-experts models like Qwen-235B-A22B. The managed service handles distributed training infrastructure and uses LoRA to share compute across runs. Tinker is in private beta and free initially, with usage-based pricing planned.

### Amp — GPT-5 Oracle

- Company: Amp (ampcode.com)
- Announced: 2025-10-01
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/gpt-5-oracle
- Record: https://forck.live/items/7800-gpt-5-oracle
- Subject: Amp

Amp announced that GPT-5 has replaced o3 as the oracle model in their product, becoming a permanent non-experimental model. The oracle can be summoned by users for complex planning and debugging tasks.

### Hugging Face — Introducing RTEB: A New Standard for Retrieval Evaluation

- Company: Hugging Face (huggingface.co)
- Announced: 2025-10-01
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/rteb
- Record: https://forck.live/items/1610-introducing-rteb-a-new-standard-for-retrieval-evaluation
- Subject: Platform

Hugging Face and collaborators introduce the beta version of the Retrieval Embedding Benchmark (RTEB), a new benchmark designed to reliably evaluate the retrieval accuracy of embedding models for real-world applications, using a hybrid strategy of open and private datasets to measure generalization.

### OpenAI — Scam operations: Online fraud networks

- Company: OpenAI (openai.com)
- Announced: 2025-10-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-scam-operations
- Record: https://forck.live/items/443-scam-operations-online-fraud-networks
- Subject: GPT / ChatGPT / API

OpenAI banned accounts linked to online fraud networks that used AI for scam scripts, impersonation, translation, and victim engagement.

### OpenAI — Cyber Operation: Russian-speaking malware tooling

- Company: OpenAI (openai.com)
- Announced: 2025-10-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-russian-speaking-malware-tooling
- Record: https://forck.live/items/442-cyber-operation-russian-speaking-malware-tooling
- Subject: GPT / ChatGPT / API

OpenAI banned accounts likely linked to Russian-speaking criminal groups that were using AI to build malware loaders, evasion layers, credential-theft scripts, and C2 infrastructure.

### OpenAI — Cyber Operation: Phishing and scripting support

- Company: OpenAI (openai.com)
- Announced: 2025-10-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-phishing-and-scripting-support
- Record: https://forck.live/items/441-cyber-operation-phishing-and-scripting-support
- Subject: GPT / ChatGPT / API

OpenAI banned accounts that were involved in malicious activities using AI to support phishing and scripting, which overlapped with PRC intelligence requirements.

### OpenAI — Cyber Operation: Korean-language malware support

- Company: OpenAI (openai.com)
- Announced: 2025-10-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-korean-language-malware-support
- Record: https://forck.live/items/440-cyber-operation-korean-language-malware-support
- Subject: GPT / ChatGPT / API

OpenAI banned Korean-language accounts that were using AI to support malware development, debugging, phishing, and credential-theft workflows.

### OpenAI — PRC-linked abuse: Surveillance and influence activity

- Company: OpenAI (openai.com)
- Announced: 2025-10-01
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-prc-linked-abuse
- Record: https://forck.live/items/439-prc-linked-abuse-surveillance-and-influence-activity
- Subject: GPT / ChatGPT / API

OpenAI banned accounts linked to China that used AI for surveillance, profiling, and research on critics and individuals.

### OpenAI — Operation “Stop News”: Recidivist influence activity

- Company: OpenAI (openai.com)
- Announced: 2025-10-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-stop-news-2025
- Record: https://forck.live/items/438-operation-stop-news-recidivist-influence-activity
- Subject: GPT / ChatGPT / API

OpenAI banned accounts linked to a Russia-origin operation, using AI to generate recidivist influence content targeting Africa and the UK.

### OpenAI — Operation “Nine–emdash Line”: Regional influence activity

- Company: OpenAI (openai.com)
- Announced: 2025-10-01
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/disrupting-malicious-uses-of-ai-nine-emdash-line
- Record: https://forck.live/items/437-operation-nine-emdash-line-regional-influence-activity
- Subject: GPT / ChatGPT / API

OpenAI banned accounts linked to a previously unreported PRC-origin operation called "Nine-emdash Line" that used AI to create regional influence content about the South China Sea, Hong Kong, and US politics.

### Google Research — AI as a research partner: Advancing theoretical computer science with AlphaEvolve

- Company: Google Research (research.google)
- Announced: 2025-09-30T16:57:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/ai-as-a-research-partner-advancing-theoretical-computer-science-with-alphaevolve/
- Record: https://forck.live/items/1412-ai-as-a-research-partner-advancing-theoretical-computer-science-with-alphaevolve
- Subject: Research

Google DeepMind researchers announce a paper demonstrating how AlphaEvolve, an LLM-based coding agent, discovered new combinatorial structures that improve inapproximability bounds for the MAX-4-CUT problem and tighten bounds on average-case hardness of certifying random graph properties.

### Harvey — Sonnet 4.5, Now Live in Harvey

- Company: Harvey (harvey.ai)
- Announced: 2025-09-30T15:30:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/sonnet-4-5-now-live-in-harvey
- Record: https://forck.live/items/17668-sonnet-4-5-now-live-in-harvey
- Subject: Harvey

Harvey has made Anthropic's Claude Sonnet 4.5 available to its customers. The model, trained with an emphasis on code generation and numerical analysis, achieved the highest score yet for the Claude family on Harvey's BigLaw Bench evaluation suite at 89.55%. Harvey is rolling out Sonnet 4.5 in the model selector to eligible clients in the US and EU.

### Supabase — Enterprise speed, enterprise standards with Bolt Cloud + Supabase

- Company: Supabase (supabase.com)
- Announced: 2025-09-30T15:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/bolt-cloud-launch
- Record: https://forck.live/items/21329-enterprise-speed-enterprise-standards-with-bolt-cloud-supabase
- Subject: Supabase

Supabase announced that every Bolt Cloud project requiring a backend will now include a complete Supabase foundation, including Postgres database, authentication, storage, edge functions, and realtime APIs. The partnership aims to combine the speed of AI-assisted development with enterprise compliance and reliability.

### Lovable — AI usage in Lovable apps through Lovable AI is FREE (with Google Gemini models) until October 6th!

- Company: Lovable (lovable.dev)
- Announced: 2025-09-30T15:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://lovable.dev/blog/free-lovable-ai
- Record: https://forck.live/items/12129-ai-usage-in-lovable-apps-through-lovable-ai-is-free-with-google-gemini-models
- Subject: Lovable / AI app builder

Lovable is offering free AI usage in published apps, powered by Google Gemini, from September 29 to October 6. Normal pricing for Lovable credits still applies, but AI inference costs are waived during this period.

### Google Research — The anatomy of a personal health agent

- Company: Google Research (research.google)
- Announced: 2025-09-30T14:41:00+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/the-anatomy-of-a-personal-health-agent/
- Record: https://forck.live/items/1413-the-anatomy-of-a-personal-health-agent
- Subject: Research

Google Research introduces a research prototype for a personal health agent (PHA) powered by LLMs, using a multi-agent architecture to provide personalized, evidence-based health insights and coaching, evaluated on a real-world dataset.

### Harvey — UChicago, Penn, Vanderbilt, WashU, Boston University, Fordham, BYU, UGA, Villanova, Baylor, SMU Join Harvey Law School Program

- Company: Harvey (harvey.ai)
- Announced: 2025-09-30T13:30:48.706000+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/uchicago-penn-washu-boston-university-vanderbilt-fordham-uga-villanova-smu-baylor-and-byu-join-harvey-law-school-program
- Record: https://forck.live/items/17669-uchicago-penn-vanderbilt-washu-boston-university-fordham-byu-uga-villanova
- Subject: Harvey

Harvey expanded its law school alliance program to include eleven additional U.S. law schools: University of Chicago, Penn, Vanderbilt, WashU, Boston University, Fordham, BYU, UGA, Villanova, Baylor, and SMU. The program provides students, faculty, and administrators access to the Harvey platform and supports co-creation of generative AI curricula during the 2025-26 academic year.

### Vercel — Towards the AI Cloud: Our Series F

- Company: Vercel (vercel.com)
- Announced: 2025-09-30T13:00:00+00:00
- Category: funding-company-news
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/series-f
- Record: https://forck.live/items/20812-towards-the-ai-cloud-our-series-f
- Subject: v0 / AI SDK / AI Gateway

Vercel announced a $300M Series F funding round led by Accel and GIC, valuing the company at $9.3 billion. The company also launched a ~$300M tender offer for certain early investors and employees. Vercel highlighted growth of its AI SDK (3M+ weekly downloads) and Next.js (500M+ downloads in 12 months), and introduced Vercel Agent in public beta with AI-powered code reviews.

### Harvey — Harvey In Practice: How to Build and Run Playbooks in Word

- Company: Harvey (harvey.ai)
- Announced: 2025-09-30T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harvey-in-practice-build-and-run-playbooks-in-word
- Record: https://forck.live/items/17670-harvey-in-practice-how-to-build-and-run-playbooks-in-word
- Subject: Harvey

Harvey published a guide showing how its Playbooks feature works in Word, including a recent upgrade that makes the tool five times faster and more accurate by generating 4.5 times the number of suggestions. The guide explains that playbooks can be uploaded in any format and converted into structured rules, and that managers can control access for different teams. New functionality includes the ability to see which clauses have been reviewed, filter to redlined text, reject redlines, and view playbook version history.

### Supabase — PostgREST 13

- Company: Supabase (supabase.com)
- Announced: 2025-09-30
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/postgrest-13-release
- Record: https://forck.live/items/21330-postgrest-13
- Subject: Supabase

PostgREST 13 introduces new features including the ability to spread to-many relationships as flat JSON arrays, automatic tsvector conversion for text and JSON columns, a maxAffected limit for update/delete operations, and new Content-Length and Proxy-Status headers for observability. It also includes breaking changes: JWT kid validation and schema validation in the search path.

### Amp — Amp CLI connects to VS Code and Neovim

- Company: Amp (ampcode.com)
- Announced: 2025-09-30
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://ampcode.com/news/cli-vscode-neovim
- Record: https://forck.live/items/7801-amp-cli-connects-to-vs-code-and-neovim
- Subject: Amp

Amp has released an update to its CLI that allows it to connect to VS Code and Neovim, enabling access to the current file, selected text, and diagnostics, as well as direct file editing through the IDE.

### Z.ai — GLM-4.6

- Company: Z.ai (z.ai)
- Announced: 2025-09-30
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://docs.z.ai/release-notes/new-released#2025-09-30
- Record: https://forck.live/items/4459-glm-4-6
- Subject: GLM

Z.ai has launched GLM-4.6, a coding model with enhanced performance on benchmarks and real-world tasks, and an expanded context window of 200K tokens.

### OpenAI — Launching Sora responsibly

- Company: OpenAI (openai.com)
- Announced: 2025-09-30
- Category: product-launch
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/launching-sora-responsibly
- Record: https://forck.live/items/446-launching-sora-responsibly
- Subject: GPT / ChatGPT / API

OpenAI announces the launch of Sora 2 and the Sora app, a video model and social creation platform built with safety as a foundation.

### OpenAI — Sora 2 is here

- Company: OpenAI (openai.com)
- Announced: 2025-09-30
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://openai.com/index/sora-2
- Record: https://forck.live/items/445-sora-2-is-here
- Subject: GPT / ChatGPT / API

OpenAI announced Sora 2, an updated video generation model that is more physically accurate, realistic, and controllable, with synchronized dialogue and sound effects, available in a new Sora app.

### OpenAI — Sora 2 System Card

- Company: OpenAI (openai.com)
- Announced: 2025-09-30
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://openai.com/index/sora-2-system-card
- Record: https://forck.live/items/444-sora-2-system-card
- Subject: GPT / ChatGPT / API

OpenAI has introduced Sora 2, a new video and audio generation model with improvements in physics, realism, audio, steerability, and style range.

### Harvey — Harvey’s Momentum in Q3

- Company: Harvey (harvey.ai)
- Announced: 2025-09-29T16:30:38.380000+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/harveys-momentum-in-q3
- Record: https://forck.live/items/17671-harvey-s-momentum-in-q3
- Subject: Harvey

Harvey reported Q3 growth including a 60% year-over-year increase in queries per user per month, 77% seat utilization, 98% gross revenue retention, and 167% net dollar retention. The company added new customers such as Latham & Watkins and Blue Owl, released Harvey for Word with 4.5x more suggestions, and opened offices in Australia and Toronto.

### Supabase — Lovable Cloud + Supabase: The Default Platform for AI Builders

- Company: Supabase (supabase.com)
- Announced: 2025-09-29T15:00:00+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://supabase.com/blog/lovable-cloud-launch
- Record: https://forck.live/items/21331-lovable-cloud-supabase-the-default-platform-for-ai-builders
- Subject: Supabase

Supabase announced that every project created in Lovable Cloud is powered by Supabase, making Supabase the default backend for Lovable's AI builders. The partnership combines Lovable's AI-first interface with Supabase's managed Postgres platform, including Auth, Storage, Edge Functions, and Realtime APIs.

### Lovable — Introducing Lovable Cloud and AI

- Company: Lovable (lovable.dev)
- Announced: 2025-09-29T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://lovable.dev/blog/lovable-cloud
- Record: https://forck.live/items/12130-introducing-lovable-cloud-and-ai
- Subject: Lovable / AI app builder

Lovable launched Lovable Cloud and Lovable AI, enabling developers to build fullstack, AI-native applications through prompting without managing infrastructure. Lovable Cloud provides a built-in backend with database, authentication, and file storage capabilities, while Lovable AI integrates Google Gemini models to add AI features such as content summarization, chatbots, and image generation.

### OpenAI — Improving support with every interaction at OpenAI

- Company: OpenAI (openai.com)
- Announced: 2025-09-29T13:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/openai-support-model
- Record: https://forck.live/items/452-improving-support-with-every-interaction-at-openai
- Subject: GPT / ChatGPT / API

OpenAI describes how it uses AI to improve customer support by reducing response times, increasing quality, and scaling to meet hypergrowth demands.

### OpenAI — Turning contracts into searchable data at OpenAI

- Company: OpenAI (openai.com)
- Announced: 2025-09-29T13:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/openai-contract-data-agent
- Record: https://forck.live/items/451-turning-contracts-into-searchable-data-at-openai
- Subject: GPT / ChatGPT / API

OpenAI built an internal system to extract contract data, reducing turnaround times and improving access for teams.

### OpenAI — Building OpenAI with OpenAI

- Company: OpenAI (openai.com)
- Announced: 2025-09-29T13:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/building-openai-with-openai
- Record: https://forck.live/items/450-building-openai-with-openai
- Subject: GPT / ChatGPT / API

OpenAI announces a new series called 'OpenAI on OpenAI' where they will share lessons on using their own technology to help other organizations.

### OpenAI — Empowering teams to unlock insights faster at OpenAI

- Company: OpenAI (openai.com)
- Announced: 2025-09-29T13:30:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/openai-research-assistant
- Record: https://forck.live/items/449-empowering-teams-to-unlock-insights-faster-at-openai
- Subject: GPT / ChatGPT / API

OpenAI's research assistant helps teams analyze support tickets and surface insights faster.

### OpenAI — Converting inbound leads into customers at OpenAI

- Company: OpenAI (openai.com)
- Announced: 2025-09-29T13:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/openai-inbound-sales-assistant
- Record: https://forck.live/items/448-converting-inbound-leads-into-customers-at-openai
- Subject: GPT / ChatGPT / API

OpenAI describes how they built an internal AI sales assistant to qualify leads and help sales teams convert customer interest.

### OpenAI — Driving sales productivity and customer success at OpenAI

- Company: OpenAI (openai.com)
- Announced: 2025-09-29T13:30:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/openai-gtm-assistant
- Record: https://forck.live/items/447-driving-sales-productivity-and-customer-success-at-openai
- Subject: GPT / ChatGPT / API

OpenAI describes building an internal AI sales assistant to automate account research, meeting preparation, product knowledge, and customer follow-up.

### Vercel — Collaborating with Anthropic on Claude Sonnet 4.5 to power intelligent coding agents

- Company: Vercel (vercel.com)
- Announced: 2025-09-29T13:00:00+00:00
- Category: availability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/collaborating-with-anthropic-on-claude-sonnet-4-5
- Record: https://forck.live/items/20813-collaborating-with-anthropic-on-claude-sonnet-4-5-to-power-intelligent-coding
- Subject: v0 / AI SDK / AI Gateway

Vercel made Anthropic's Claude Sonnet 4.5 available on its AI Gateway and AI SDK, and released an open-source Coding Agent Platform template that defaults to using the model. Internal evaluations showed the model outperformed its predecessor on Next.js builds, linting, and feature implementation tasks.

### OpenAI — Combating online child sexual exploitation & abuse

- Company: OpenAI (openai.com)
- Announced: 2025-09-29T03:00:00+00:00
- Category: safety-policy-update
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/combating-online-child-sexual-exploitation-abuse
- Record: https://forck.live/items/454-combating-online-child-sexual-exploitation-abuse
- Subject: GPT / ChatGPT / API

OpenAI describes its efforts to combat online child sexual exploitation and abuse through strict usage policies, advanced detection tools, and industry collaboration.

### OpenAI — Introducing parental controls

- Company: OpenAI (openai.com)
- Announced: 2025-09-29T03:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-parental-controls
- Record: https://forck.live/items/453-introducing-parental-controls
- Subject: GPT / ChatGPT / API

OpenAI is rolling out parental controls and a new parent resource page to help families guide how ChatGPT works in their homes.

### Thinking Machines Lab — LoRA Without Regret

- Company: Thinking Machines Lab (thinkingmachines.ai)
- Announced: 2025-09-29
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://thinkingmachines.ai/blog/lora/
- Record: https://forck.live/items/18569-lora-without-regret
- Subject: Thinking Machines / Inkling

Thinking Machines Lab conducted experiments comparing LoRA to full fine-tuning (FullFT) across supervised fine-tuning and reinforcement learning. They found that LoRA matches FullFT in sample efficiency and ultimate performance when key details are correct, such as applying LoRA to all weight matrices including MLP and MoE layers. LoRA underperforms on very large datasets that exceed its capacity and is less tolerant of large batch sizes than FullFT.

### Amp — Now Using Claude Sonnet 4.5

- Company: Amp (ampcode.com)
- Announced: 2025-09-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/claude-sonnet-4-5
- Record: https://forck.live/items/7802-now-using-claude-sonnet-4-5
- Subject: Amp

Amp has switched its primary model to Claude Sonnet 4.5, the latest release from Anthropic. The company reports improvements in coding tasks such as reduced sycophancy, higher success rates for file edits, and improved reasoning during debugging. The switch is automatic for users who update to the latest release of Amp's VS Code extension or CLI.

### Cognition — Rebuilding Devin for Claude Sonnet 4.5: Lessons and Challenges

- Company: Cognition (cognition.com)
- Announced: 2025-09-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/devin-sonnet-4-5-lessons-and-challenges
- Record: https://forck.live/items/7691-rebuilding-devin-for-claude-sonnet-4-5-lessons-and-challenges
- Subject: Devin / SWE

Cognition rebuilt Devin for Claude Sonnet 4.5, resulting in a 2x faster and 12% better performance on Junior Developer Evals. The new version is available in Agent Preview, while the old version remains available. The post details observations about the model's awareness of its context window, note-taking behavior, and parallel tool execution.

### Cognition — Announcing Devin Agent Preview with Sonnet 4.5

- Company: Cognition (cognition.com)
- Announced: 2025-09-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://cognition.com/blog/devin-agent-preview-sonnet-4-5
- Record: https://forck.live/items/7690-announcing-devin-agent-preview-with-sonnet-4-5
- Subject: Devin / SWE

Cognition announced that Claude Sonnet 4.5 is now available in Devin, delivering an 18% improvement in planning performance and a 12% improvement in end-to-end eval scores over the previous model. The model's new behaviors required changes to Devin's architecture.

### DeepSeek — Introducing DeepSeek-V3.2-Exp

- Company: DeepSeek (deepseek.com)
- Announced: 2025-09-29
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.deepseek.com/en/news/v3-2-exp
- Record: https://forck.live/items/4321-introducing-deepseek-v3-2-exp
- Subject: DeepSeek models

DeepSeek announces DeepSeek-V3.2-Exp, an experimental model built on V3.1-Terminus that introduces DeepSeek Sparse Attention (DSA) for faster and more efficient training and inference on long contexts. The model is now live on App, Web, and API. API prices are reduced by more than 50%. Benchmarks show V3.2-Exp performs on par with V3.1-Terminus. V3.1-Terminus remains available via a temporary API until October 15, 2025. Key GPU kernels are released as open source in TileLang and CUDA.

### Hugging Face — VibeGame: Exploring Vibe Coding Games

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-29
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/vibegame
- Record: https://forck.live/items/1612-vibegame-exploring-vibe-coding-games
- Subject: Platform

Hugging Face publishes a blog post by Dylan Ebert introducing VibeGame, a high-level declarative game engine built on three.js, rapier, and bitecs, designed specifically for AI-assisted game development. The post discusses 'vibe coding' games, problems with context management, and compares platforms like Roblox, Unity, and the web stack.

### Hugging Face — Accelerating Qwen3-8B Agent on Intel® Core™ Ultra with Depth-Pruned Draft Models

- Company: Hugging Face (huggingface.co)
- Announced: 2025-09-29
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/intel-qwen3-agent
- Record: https://forck.live/items/1611-accelerating-qwen3-8b-agent-on-intel-core-ultra-with-depth-pruned-draft-models
- Subject: Platform

Intel and Hugging Face demonstrate accelerating Qwen3-8B agent inference on Intel Core Ultra using speculative decoding with a depth-pruned draft model (Qwen3-0.6B), achieving ~1.4× speedup over baseline.

### OpenAI — Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol

- Company: OpenAI (openai.com)
- Announced: 2025-09-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/buy-it-in-chatgpt
- Record: https://forck.live/items/455-buy-it-in-chatgpt-instant-checkout-and-the-agentic-commerce-protocol
- Title in English (translated by forck.live from en): Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol
- Subject: GPT / ChatGPT / API

OpenAI introduces early features for agentic commerce in ChatGPT, enabling shopping interactions involving people, AI agents, and businesses.