# forck.live — 28 July – 3 August 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-W31, 28 July – 3 August 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-W31
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### TII — 📚 3LM: A Benchmark for Arabic LLMs in STEM and Code

- Company: TII (tii.ae)
- Announced: 2025-08-01T14:25:21+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/tiiuae/3lm-benchmark
- Record: https://forck.live/items/1645-3lm-a-benchmark-for-arabic-llms-in-stem-and-code

Hugging Face announces the release of 3LM, a benchmark for evaluating Arabic large language models on STEM and code generation, consisting of three datasets: Native STEM, Synthetic STEM, and translated code benchmarks (HumanEval-ar, MBPP-ar). The post reports evaluation results for over 40 models.

### Harvey — 5 Questions With Siva Gurumurthy

- Company: Harvey (harvey.ai)
- Announced: 2025-08-01T14:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.harvey.ai/blog/5-questions-with-siva-gurumurthy
- Record: https://forck.live/items/17693-5-questions-with-siva-gurumurthy
- Subject: Harvey

Harvey's CTO Siva Gurumurthy describes the company's engineering focus on building a core platform layer, navigating global legal data requirements, creating an enterprise platform vision, and refining AI models through customer feedback. He highlights Harvey's scale as one of the world's largest document processing systems, where a single enterprise customer may upload 200,000 to over 1 million documents for AI agents to process.

### Vercel — A new wave of software, shipped on Vercel

- Company: Vercel (vercel.com)
- Announced: 2025-08-01T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/shipped-on-vercel
- Record: https://forck.live/items/20846-a-new-wave-of-software-shipped-on-vercel
- Subject: v0 / AI SDK / AI Gateway

Vercel announced the Shipped on Vercel showcase, highlighting real production apps built on its platform by teams such as Browserbase, Chatbase, Cursor, and OpenAI. Submissions are open for short windows, typically 48 hours, and selected teams will be featured.

### Genspark — Multi-Agent Orchestration: Shipped. On-Demand Ephemeral Agent Creation: Next

- Company: Genspark (genspark.ai)
- Announced: 2025-08-01
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.genspark.ai/blog/genspark-multiagent-orchestration
- Record: https://forck.live/items/18273-multi-agent-orchestration-shipped-on-demand-ephemeral-agent-creation-next
- Subject: Genspark

product Kay, Co-founder and CTO of Genspark

### LG AI Research — AI for industrial sites evolves into the EXAONE Ecosystem! LG AI Talk Concert 2025

- Company: LG AI Research (lgresearch.ai)
- Announced: 2025-08-01
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.lgresearch.ai/news/view?seq=578
- Record: https://forck.live/items/11002-ai-for-industrial-sites-evolves-into-the-exaone-ecosystem-lg-ai-talk-concert
- Subject: EXAONE

LG AI Research held the LG AI Talk Concert 2025, where it announced the EXAONE Ecosystem, an integrated solution covering AI models, data, services, and infrastructure. The event featured the unveiling of EXAONE 4.0, a hybrid model with 32B and 1.2B parameter variants, and EXAONE 4.0 VL, a multimodal model that achieved top performance on the ChartQA benchmark. Also introduced was EXAONE Path 2.0, a histopathological image analysis model available on the NVIDIA MONAI platform.

### LG AI Research — AI for industrial sites evolves into the EXAONE Ecosystem! LG AI Talk Concert 2025

- Company: LG AI Research (lgresearch.ai)
- Announced: 2025-08-01
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.lgresearch.ai/blog/view?seq=578
- Record: https://forck.live/items/4782-ai-for-industrial-sites-evolves-into-the-exaone-ecosystem-lg-ai-talk-concert
- Subject: EXAONE

LG AI Research announced new models EXAONE Deep, EXAONE 4.0, EXAONE 4.0 VL, and EXAONE Path 2.0 as part of the EXAONE Ecosystem at the LG AI Talk Concert 2025.

### Mistral AI — Unlocking the potential of vision language models on satellite imagery through fine-tuning

- Company: Mistral AI (mistral.ai)
- Announced: 2025-08-01
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://mistral.ai/news/unlocking-potential-vision-language-models-satellite-imagery-fine-tuning
- Record: https://forck.live/items/3315-unlocking-the-potential-of-vision-language-models-on-satellite-imagery-through
- Subject: Mistral / Le Chat

Mistral AI discusses fine-tuning vision language models for satellite imagery.

### Replicate — How to prompt Veo 3 with images

- Company: Replicate (replicate.com)
- Announced: 2025-08-01
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://replicate.com/blog/veo-3-image
- Record: https://forck.live/items/2481-how-to-prompt-veo-3-with-images
- Subject: Platform

A blog post on Replicate describing how to use image input with Veo 3 to preserve style, handle typography, and achieve more creative control. Key tricks include style preservation, text animation, and using image input for precise results.

### OpenAI — Figma uses AI to transform digital design

- Company: OpenAI (openai.com)
- Announced: 2025-08-01
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/figma-david-kossnick
- Record: https://forck.live/items/512-figma-uses-ai-to-transform-digital-design
- Subject: GPT / ChatGPT / API

OpenAI highlights how Figma is using AI, including the tool Figma Make, to transform digital design workflows for designers, developers, and non-technical creators.

### Vercel — AI SDK 5

- Company: Vercel (vercel.com)
- Announced: 2025-07-31T13:00:00+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/ai-sdk-5
- Record: https://forck.live/items/20848-ai-sdk-5
- Subject: v0 / AI SDK / AI Gateway

Vercel released AI SDK 5, a major update to its open-source TypeScript/JavaScript AI toolkit. The release introduces fully typed chat integrations for React, Svelte, Vue, and Angular, with separate UI and model message types and customizable data parts for streaming arbitrary data.

### Vercel — Vercel Open Source Program: Summer cohort

- Company: Vercel (vercel.com)
- Announced: 2025-07-31T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://vercel.com/blog/summer-2025-oss-program
- Record: https://forck.live/items/20847-vercel-open-source-program-summer-cohort
- Subject: v0 / AI SDK / AI Gateway

In April, we launched the Vercel Open Source Program , a developer initiative that gives maintainers the resources, credits, and support they need to ship faster and scale confidently, starting with the spring 2025 cohort . We're now honored to announce the summer 2025 cohort.

### Krea — Releasing Open Weights for FLUX.1 Krea

- Company: Krea (krea.ai)
- Announced: 2025-07-31
- Category: open-weight-release
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://www.krea.ai/blog/flux-krea-open-source-release
- Record: https://forck.live/items/15094-releasing-open-weights-for-flux-1-krea
- Subject: Krea 2 / AI creative platform

Krea released open weights for FLUX.1-Krea [dev], a guidance-distilled checkpoint compatible with FLUX.1-dev. The model was trained with Black Forest Labs to prioritize aesthetics and photorealism, aiming to reduce the 'AI look' in generated images.

### Replicate — Open source video is back

- Company: Replicate (replicate.com)
- Announced: 2025-07-31
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://replicate.com/blog/wan-22
- Record: https://forck.live/items/2482-open-source-video-is-back
- Subject: Platform

Replicate, in partnership with Pruna AI, released optimized versions of the open source video generation model Wan 2.2, offering fast image-to-video and text-to-video at low prices, with API access.

### Hugging Face — Implementing MCP Servers in Python: An AI Shopping Assistant with Gradio

- Company: Hugging Face (huggingface.co)
- Announced: 2025-07-31
- Category: developer-tool-release
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://huggingface.co/blog/gradio-vton-mcp
- Record: https://forck.live/items/1646-implementing-mcp-servers-in-python-an-ai-shopping-assistant-with-gradio
- Subject: Platform

Hugging Face announces that Gradio now supports the Model Context Protocol, allowing Python developers to create MCP servers that connect LLMs to AI models and Spaces. The post provides a tutorial for building an AI shopping assistant using Gradio, the IDM-VTON try-on model, and VS Code's AI chat.

### OpenAI — Introducing Stargate Norway

- Company: OpenAI (openai.com)
- Announced: 2025-07-31
- Category: infrastructure-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/introducing-stargate-norway
- Record: https://forck.live/items/513-introducing-stargate-norway
- Subject: GPT / ChatGPT / API

OpenAI announced Stargate Norway, its first AI data center initiative in Europe under the OpenAI for Countries program, as part of its infrastructure platform Stargate.

### Amp — Amp Tab 30% Faster

- Company: Amp (ampcode.com)
- Announced: 2025-07-30
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://ampcode.com/news/amp-tab-30-percent-faster
- Record: https://forck.live/items/7824-amp-tab-30-faster
- Subject: Amp

Amp announced that response times of Amp Tab, its in-editor completion engine, are now 30% faster, with up to 50% improvements during peak usage. The improvement was achieved through infrastructure optimization with Baseten, switching to TensorRT-LLM as the inference engine, implementing KV caching with speculative decoding, and using a modified lookahead decoding algorithm with improved n-gram candidate selection and variable-length speculations.

### Mistral AI — Announcing Codestral 25.08 and the Complete Mistral Coding Stack for Enterprise

- Company: Mistral AI (mistral.ai)
- Announced: 2025-07-30
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://mistral.ai/news/codestral-25-08
- Record: https://forck.live/items/3316-announcing-codestral-25-08-and-the-complete-mistral-coding-stack-for-enterprise
- Subject: Mistral / Le Chat

Mistral AI announced Codestral 25.08, an update to its coding model, along with a complete Mistral Coding Stack for enterprise use.

### OpenAI — Three lessons for creating a sustainable AI advantage

- Company: OpenAI (openai.com)
- Announced: 2025-07-30
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://openai.com/index/intercom
- Record: https://forck.live/items/514-three-lessons-for-creating-a-sustainable-ai-advantage
- Subject: GPT / ChatGPT / API

Three lessons for creating a sustainable AI advantage, based on Intercom's experience building a scalable AI platform for customer support.

### Harvey — A More Unified Harvey Experience

- Company: Harvey (harvey.ai)
- Announced: 2025-07-29T15:30:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.harvey.ai/blog/a-more-unified-harvey-experience
- Record: https://forck.live/items/17694-a-more-unified-harvey-experience
- Subject: Harvey

Harvey announced a major update to its legal AI platform, introducing a unified interface that routes queries to the best data, workflow, or product surface via model orchestration. The update includes a simplified homepage with personalized suggestions and merges drafting functionality into the main Assistant interface so users can draft and chat in one place.

### Vercel — Join the v0 Ambassador Program

- Company: Vercel (vercel.com)
- Announced: 2025-07-29T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/join-the-v0-ambassador-program
- Record: https://forck.live/items/20849-join-the-v0-ambassador-program
- Subject: v0 / AI SDK / AI Gateway

Vercel announced the v0 Ambassador Program, a beta initiative to sponsor community members who build and share content about v0. Ambassadors receive monthly v0 credits, early access to features, swag, and referral earnings in exchange for creating weekly content and providing feedback.

### OpenAI — Introducing study mode in ChatGPT

- Company: OpenAI (openai.com)
- Announced: 2025-07-29T10:00:00+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://openai.com/index/chatgpt-study-mode
- Record: https://forck.live/items/515-introducing-study-mode-in-chatgpt
- Subject: GPT / ChatGPT / API

OpenAI announces study mode in ChatGPT, a new learning experience that guides students through problems with questions, scaffolding, and feedback.

### Genspark — Less Control, More Tools: How Genspark Built a Super Agent That Scales

- Company: Genspark (genspark.ai)
- Announced: 2025-07-29
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.genspark.ai/blog/genspark-less-control-more-tools
- Record: https://forck.live/items/18274-less-control-more-tools-how-genspark-built-a-super-agent-that-scales
- Subject: Genspark

insights Kay, Co-founder and CTO of Genspark

### Amp — Thread Forking in VS Code

- Company: Amp (ampcode.com)
- Announced: 2025-07-29
- Category: capability-change
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://ampcode.com/news/thread-forking
- Record: https://forck.live/items/7825-thread-forking-in-vs-code
- Subject: Amp

Amp announced a thread forking feature in VS Code that allows users to create a new thread containing all prior messages, with a backlink to the original. An update later removed the explicit fork action, instructing users to create a new thread and mention the original by URL or ID.

### Together AI — VirtueGuard: Enterprise-Grade AI Security and Safety Now on Together AI

- Company: Together AI (together.ai)
- Announced: 2025-07-29
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/virtueguard
- Record: https://forck.live/items/2418-virtueguard-enterprise-grade-ai-security-and-safety-now-on-together-ai
- Subject: Inference platform

Together AI announces the availability of VirtueGuard, an AI security and safety guardrail model developed by Virtue AI, now integrated into the Together AI platform. VirtueGuard offers an 8ms response time, covers 12 risk categories, and can be deployed with a single API parameter.

### Hugging Face — Introducing Trackio: A Lightweight Experiment Tracking Library from Hugging Face

- Company: Hugging Face (huggingface.co)
- Announced: 2025-07-29
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://huggingface.co/blog/trackio
- Record: https://forck.live/items/1647-introducing-trackio-a-lightweight-experiment-tracking-library-from-hugging-face
- Subject: Platform

Hugging Face introduces Trackio, an open-source, free experiment tracking Python library that provides a local dashboard and integration with Hugging Face Spaces for sharing. It is a drop-in replacement for wandb.

### Sakana AI — 最先端のAI技術をビジネスへ：Sakana AI、Applied Teamメンバーインタビュー

- Company: Sakana AI (sakana.ai)
- Announced: 2025-07-28T15:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://sakana.ai/applied-team-interview-2025
- Record: https://forck.live/items/4710-ai-sakana-ai-applied-team
- Title in English (translated by forck.live from ja): Bringing cutting-edge AI technology to business: Interview with Sakana AI Applied Team members
- Subject: Sakana models

Sakana AI announces the launch of its Applied Team (Business Development Division) in 2025, with a focus on social implementation of cutting-edge generative AI technology in the finance, defense and public sectors. The team currently has about 35 members and is hiring.

### Vercel — Fluid: How we built serverless servers

- Company: Vercel (vercel.com)
- Announced: 2025-07-28T13:00:00+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://vercel.com/blog/fluid-how-we-built-serverless-servers
- Record: https://forck.live/items/20850-fluid-how-we-built-serverless-servers
- Subject: v0 / AI SDK / AI Gateway

Vercel described the infrastructure behind Fluid compute, which powers over 45 billion weekly requests and saves customers up to 95% by never charging CPU rates for idle time. The system uses a secure TCP-based tunnel to stream responses from AWS Lambda, and a compute-resolver service to route requests efficiently. Active CPU pricing was introduced at Vercel Ship 2025 for more cost-effective compute.

### Runway — Runway and IMAX® Partner on Exclusive Presentation of the 2025 AI Film Festival

- Company: Runway (runway.com)
- Announced: 2025-07-28T10:10:10.110000+00:00
- Category: partnership-acquisition
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://runway.com/news/runway-imax-aiff-presentation
- Record: https://forck.live/items/7904-runway-and-imax-partner-on-exclusive-presentation-of-the-2025-ai-film-festival
- Subject: Gen / Aleph

Runway and IMAX have partnered to screen the finalist films from the 2025 AI Film Festival exclusively in IMAX at 10 select locations across the United States. The screenings will take place from August 17 to August 20, with tickets now on sale.

### Z.ai — GLM-4.5 Series

- Company: Z.ai (z.ai)
- Announced: 2025-07-28
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://docs.z.ai/release-notes/new-released#2025-07-28
- Record: https://forck.live/items/4462-glm-4-5-series
- Subject: GLM

Z.ai launched GLM-4.5, a new native agentic LLM with doubled parameter efficiency and strong reasoning, coding, and agentic capabilities, and one-click compatibility with the Claude Code framework.

### Together AI — Together Evaluations: Benchmark Models for Your Tasks

- Company: Together AI (together.ai)
- Announced: 2025-07-28
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/introducing-together-evaluations
- Record: https://forck.live/items/2419-together-evaluations-benchmark-models-for-your-tasks
- Subject: Inference platform

Together AI announces an early preview of Together Evaluations, a platform for benchmarking LLM response quality using LLM-as-a-judge with three modes: classify, score, and compare.