# forck.live — 12–18 May 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-W20, 12–18 May 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-W20 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Supabase — Simplifying back-end complexity with Supabase Data APIs - Company: Supabase (supabase.com) - Announced: 2025-05-17 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/simplify-backend-with-data-api - Record: https://forck.live/items/21358-simplifying-back-end-complexity-with-supabase-data-apis - Subject: Supabase Supabase's guide explains how its platform automatically generates REST and GraphQL APIs from Postgres database schemas, eliminating the need to manually build and maintain middleware. It covers auto-generated endpoints for CRUD operations, custom endpoints via Database Functions and Edge Functions, and security through Row Level Security (RLS). The post includes a customer example where Quilia reduced development time by 75% using the Supabase Data API. ### Runway — Introducing the Gen-4 Image API - Company: Runway (runway.com) - Announced: 2025-05-16T10:10:10.110000+00:00 - Category: api-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://runway.com/news/introducing-runway-api-for-gen-4-images - Record: https://forck.live/items/7908-introducing-the-gen-4-image-api - Subject: Gen / Aleph Runway made Gen-4 Image, including References, available via the Runway API, enabling integration into apps, products, platforms, and websites at a cost of $0.08 per generated image. ### OpenAI — Addendum to o3 and o4-mini system card: Codex - Company: OpenAI (openai.com) - Announced: 2025-05-16T08:00:00+00:00 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://openai.com/index/o3-o4-mini-codex-system-card-addendum - Record: https://forck.live/items/560-addendum-to-o3-and-o4-mini-system-card-codex - Subject: GPT / ChatGPT / API OpenAI describes Codex, a cloud-based coding agent powered by codex-1, a version of o3 optimized for software engineering through reinforcement learning on real-world coding tasks. ### Replicate — NVIDIA H100 GPUs are here - Company: Replicate (replicate.com) - Announced: 2025-05-16 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replicate.com/blog/nvidia-h100-gpus-are-here - Record: https://forck.live/items/2494-nvidia-h100-gpus-are-here - Subject: Platform Replicate announces that NVIDIA H100 GPUs are now available on their platform, along with multi-GPU configurations of A100 and L40S GPUs that were previously only available in deployments. Pricing details are provided for each configuration. ### Replit — Replit: The Safest Place for Vibe Coding - Company: Replit (replit.com) - Announced: 2025-05-15T16:09:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/safe-vibe-coding - Record: https://forck.live/items/17962-replit-the-safest-place-for-vibe-coding - Subject: Replit Agent Replit announced several new security features for its platform, including making Replit Auth the default authentication provider, launching pre-deployment security scanning with Semgrep, blocking the Agent from modifying key system files, and redirecting pasted API keys to the Secrets tool. The company also introduced App History, a rebuilt version rollback interface that shows screenshots, allows database state rollback, and offers previews for checkpoints within the past 7 days. ### Vercel — How Fern delivers 6M+ monthly views and 80% faster docs with Vercel - Company: Vercel (vercel.com) - Announced: 2025-05-15T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/how-fern-delivers-6m-monthly-views-and-80-faster-docs-with-vercel - Record: https://forck.live/items/20874-how-fern-delivers-6m-monthly-views-and-80-faster-docs-with-vercel - Subject: v0 / AI SDK / AI Gateway Fern, a multi-tenant documentation platform, reports that migrating to Next.js App Router on Vercel reduced page load times by 50-80% and cut deploy times from 30 minutes to five minutes. The platform serves 6 million+ monthly page views and 1 million+ unique visitors, with 63% month-over-month growth. ### TII — Falcon-Edge: A series of powerful, universal, fine-tunable 1.58bit language models. - Company: TII (tii.ae) - Announced: 2025-05-15T12:00:00+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://falcon-lm.github.io/blog/falcon-edge/ - Record: https://forck.live/items/18416-falcon-edge-a-series-of-powerful-universal-fine-tunable-1-58bit-language-models - Subject: Falcon LLM TII released the Falcon-Edge series, a collection of 1.58-bit language models based on the BitNet architecture, available in 1B and 3B parameter sizes with base and instruction-tuned variants. The models use a novel pre-training paradigm that produces both non-quantized and quantized variants from a single training process, and they are designed for efficient deployment on edge devices. ### Genspark — Introducing the World''s First Full Agentic Download Agent and Genspark AI Drive! - Company: Genspark (genspark.ai) - Announced: 2025-05-15 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.genspark.ai/blog/genspark-download-agent-and-ai-drive - Record: https://forck.live/items/18286-introducing-the-world-s-first-full-agentic-download-agent-and-genspark-ai-drive - Subject: Genspark ### Cartesia — Introducing Organizations and Dashboards - Company: Cartesia (cartesia.ai) - Announced: 2025-05-15 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.cartesia.ai/blog/introducing-organizations-and-dashboards - Record: https://forck.live/items/8018-introducing-organizations-and-dashboards - Subject: Sonic / Ink Cartesia introduced two new features for its platform: Organizations, which enables teams to share API keys, voices, and billing under one account, and Dashboards, which provides real-time visibility into usage, concurrency, and WebSocket connections. ### Luma — Luma is Partnering with HUMAIN to Accelerate the Arrival of Multimodal AGI - Company: Luma (lumalabs.ai) - Announced: 2025-05-15 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lumalabs.ai/news/luma-joins-forces-with-humain - Record: https://forck.live/items/7985-luma-is-partnering-with-humain-to-accelerate-the-arrival-of-multimodal-agi - Subject: Ray / Dream Machine Luma announced a strategic partnership with Saudi Arabia's HUMAIN AI covering compute, go-to-market, and investment to accelerate multimodal AGI. As part of the collaboration, Luma is deploying its Ray2 video generation model on HUMAIN's infrastructure. ### Amp — Amp Is Now Available To Everyone — Here's How I Use It - Company: Amp (ampcode.com) - Announced: 2025-05-15 - Category: availability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://ampcode.com/how-i-use-amp - Record: https://forck.live/items/7843-amp-is-now-available-to-everyone-here-s-how-i-use-it - Subject: Amp Amp made its frontier agent available to everyone without a waitlist. The agent writes 70-80% of code and uses Claude Sonnet 4 under the hood. ### Replicate — Run 30,000+ LoRAs on Hugging Face with Replicate - Company: Replicate (replicate.com) - Announced: 2025-05-15 - Category: capability-change - Coverage: not counted - Announcement: no - Group: routine - Source: https://replicate.com/blog/replicate-on-huggingface - Record: https://forck.live/items/2495-run-30-000-loras-on-hugging-face-with-replicate - Subject: Platform LoRAs can now run directly on the Hugging Face Hub using Replicate for inference, via a small update to Hugging Face's inference client that routes requests to Replicate's flux-dev-lora model. ### Hugging Face — The Transformers Library: standardizing model definitions - Company: Hugging Face (huggingface.co) - Announced: 2025-05-15 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/transformers-model-definition - Record: https://forck.live/items/1697-the-transformers-library-standardizing-model-definitions - Subject: Platform Hugging Face announces that the Transformers library will act as a central standard for model definitions across the ML ecosystem, aiming to increase interoperability among training frameworks, inference engines, and local runtimes. They plan to simplify model contributions, deprecate redundant components, and enable day-0 support for new architectures in downstream tools like vLLM, SGLang, TGI, llama.cpp, and MLX. ### Stability AI — Stability AI and Arm Collaborate to Release Stable Audio Open Small, Enabling Real-World Deployment for On-Device Audio Generation - Company: Stability AI (stability.ai) - Announced: 2025-05-14T14:58:02+00:00 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://stability.ai/news-updates/stability-ai-and-arm-release-stable-audio-open-small-enabling-real-world-deployment-for-on-device-audio-control - Record: https://forck.live/items/17489-stability-ai-and-arm-collaborate-to-release-stable-audio-open-small-enabling - Subject: Stable Diffusion Stability AI and Arm released Stable Audio Open Small, a 341 million parameter text-to-audio model optimized to run entirely on Arm CPUs. The model can generate up to 11 seconds of audio on a smartphone in less than 8 seconds. It is available for free under the Stability AI Community License for commercial and non-commercial use. ### Vercel — How Consensys rebuilt MetaMask.io with Vercel and Next.js - Company: Vercel (vercel.com) - Announced: 2025-05-14T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/how-consensys-rebuilt-metamask-io-with-vercel-and-next-js - Record: https://forck.live/items/20875-how-consensys-rebuilt-metamask-io-with-vercel-and-next-js - Subject: v0 / AI SDK / AI Gateway Consensys migrated MetaMask.io from Static Site Generation on GitHub Pages to Next.js with Server-Side Rendering hosted on Vercel. The new architecture uses Contentful as a CMS, Vercel Functions for dynamic endpoints, and Incremental Static Regeneration for live data like token prices. The migration improved uptime, deployment speed, SEO outcomes, and team collaboration across engineering, design, and content. ### OpenAI — AI powers Expedia’s marketing evolution - Company: OpenAI (openai.com) - Announced: 2025-05-14T10:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://openai.com/index/expedia-jochen-koedijk - Record: https://forck.live/items/562-ai-powers-expedia-s-marketing-evolution - Subject: GPT / ChatGPT / API OpenAI published a conversation with Expedia Group's Chief Marketing Officer. ### Amp — Raising an Agent, Episode 5 - Company: Amp (ampcode.com) - Announced: 2025-05-14 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://ampcode.com/podcast - Record: https://forck.live/items/7844-raising-an-agent-episode-5 - Subject: Amp ### Hugging Face — Improving Hugging Face Model Access for Kaggle Users - Company: Hugging Face (huggingface.co) - Announced: 2025-05-14 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/kaggle-integration - Record: https://forck.live/items/1698-improving-hugging-face-model-access-for-kaggle-users - Subject: Platform Hugging Face and Kaggle announce an integration that allows users to navigate between Hugging Face model pages and Kaggle notebooks, automatically generate Hugging Face model pages from Kaggle notebooks, and discover models and community examples on Kaggle. ### Harvey — Expanding Harvey's Model Offerings - Company: Harvey (harvey.ai) - Announced: 2025-05-13T13:30:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.harvey.ai/blog/expanding-harveys-model-offerings - Record: https://forck.live/items/17710-expanding-harvey-s-model-offerings - Subject: Harvey Harvey is incorporating and optimizing foundation models from Anthropic and Google into its platform, allowing auto-routing to the best model for legal tasks and giving users a model selector for manual choice. The change is driven by improved baseline legal reasoning in general models and the recognition that no single model is best for all legal tasks. Harvey also plans to evaluate and integrate additional foundation model providers over time. ### Vercel — Updated v0 pricing - Company: Vercel (vercel.com) - Announced: 2025-05-13T13:00:00+00:00 - Category: pricing-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://vercel.com/blog/updated-v0-pricing - Record: https://forck.live/items/20876-updated-v0-pricing - Subject: v0 / AI SDK / AI Gateway Vercel updated v0 pricing, switching from fixed message counts to token-based metering. Usage is now measured in input and output tokens that convert to credits. Free users receive $5 in monthly included credits, Premium users $20, and Team plan users $30. ### Cua — Trajectory Viewer for cua - Company: Cua (cua.ai) - Announced: 2025-05-13 - Category: developer-tool-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://cua.ai/blog/trajectory-viewer - Record: https://forck.live/items/16843-trajectory-viewer-for-cua - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua released a Trajectory Viewer, a GUI tool that lets developers explore saved trajectories from Cua computer agent runs to debug agents, analyze failure cases, and collect training data. The tool runs in the browser and allows users to step through each action and observation. ### Hugging Face — Blazingly fast whisper transcriptions with Inference Endpoints - Company: Hugging Face (huggingface.co) - Announced: 2025-05-13 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/fast-whisper-endpoints - Record: https://forck.live/items/1699-blazingly-fast-whisper-transcriptions-with-inference-endpoints - Subject: Platform Hugging Face announced a new deployment option for OpenAI Whisper on Inference Endpoints, achieving up to 8x speed improvement and leveraging community open-source technologies. ### Vercel — The spring 2025 cohort of Vercel’s Open Source Program - Company: Vercel (vercel.com) - Announced: 2025-05-12T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/spring25-oss-program - Record: https://forck.live/items/20877-the-spring-2025-cohort-of-vercel-s-open-source-program - Subject: v0 / AI SDK / AI Gateway Vercel announced the spring 2025 cohort of its Open Source Program, which provides maintainers with $3,600 in Vercel credits, partner perks, and a dedicated Slack space. The cohort includes 18 projects such as Vecto3d, Script Kit, Lenis, Fumadocs, Shadcn-table, Morphic, Zola, Valibot, shadcn-form, jsPDF, LobeChat, Milkdown, KokonutUI, nuqs, Nemo, RetroUI, Tweakcn, CourseLit, and Comp AI. ### OpenAI — Introducing HealthBench - Company: OpenAI (openai.com) - Announced: 2025-05-12T10:30:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/healthbench - Record: https://forck.live/items/563-introducing-healthbench - Subject: GPT / ChatGPT / API OpenAI introduces HealthBench, a new evaluation benchmark for AI in healthcare that evaluates models in realistic scenarios, built with input from over 250 physicians to provide a shared standard for model performance and safety in health. ### Hugging Face — Vision Language Models (Better, faster, stronger) - Company: Hugging Face (huggingface.co) - Announced: 2025-05-12 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://huggingface.co/blog/vlms-2025 - Record: https://forck.live/items/1700-vision-language-models-better-faster-stronger - Subject: Platform This blog post surveys recent developments in Vision Language Models (VLMs) over the past year, covering new model trends such as any-to-any models, reasoning models, small yet capable models, and specialized capabilities. It highlights specific models including Qwen 2.5 Omni, MiniCPM-o 2.6, Janus-Pro-7B, QVQ-72B-preview, Kimi-VL-A3B-Thinking, SmolVLM, SmolVLM2, and gemma3-4b-it, and discusses new benchmarks like MMT-Bench and MMMU-Pro.