# forck.live — 9–15 June 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-W24, 9–15 June 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-W24 - The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md - Catalogue: https://forck.live/llms.txt ## Announcements ### Genspark — Seeing AGI (2) – What Should I Do Now? - Company: Genspark (genspark.ai) - Announced: 2025-06-13 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.genspark.ai/blog/seeing-agi-impact2 - Record: https://forck.live/items/18280-seeing-agi-2-what-should-i-do-now - Subject: Genspark ### Stability AI — Stable Diffusion 3.5 Models Optimized with TensorRT Deliver 2X Faster Performance and 40% Less Memory on NVIDIA RTX GPUs - Company: Stability AI (stability.ai) - Announced: 2025-06-12T21:21:17+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://stability.ai/news-updates/stable-diffusion-35-models-optimized-with-tensorrt-deliver-2x-faster-performance-and-40-less-memory-on-nvidia-rtx-gpus - Record: https://forck.live/items/17487-stable-diffusion-3-5-models-optimized-with-tensorrt-deliver-2x-faster - Subject: Stable Diffusion Stability AI, in collaboration with NVIDIA, released TensorRT-optimized versions of Stable Diffusion 3.5 Large and Medium. The optimizations deliver up to 2.3x faster generation on SD3.5 Large and 1.7x faster on SD3.5 Medium, while reducing VRAM requirements by 40%. The models are available under the Stability AI Community License for commercial and non-commercial use. ### Vercel — Building efficient MCP servers - Company: Vercel (vercel.com) - Announced: 2025-06-12T13:00:00+00:00 - Category: developer-tool-release - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/building-efficient-mcp-servers - Record: https://forck.live/items/20863-building-efficient-mcp-servers - Subject: v0 / AI SDK / AI Gateway Vercel's MCP adapter now supports Streamable HTTP, the recommended transport in the March 2025 MCP specification, replacing the less efficient SSE. The adapter also supports SSE with the option to disable it. One MCP server deployed on Vercel that switched to Streamable HTTP cut CPU usage in half despite continued user growth. ### Hugging Face — How Long Prompts Block Other Requests - Optimizing LLM Performance - Company: Hugging Face (huggingface.co) - Announced: 2025-06-12T08:00:13+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/tngtech/llm-performance-blocked-by-long-prompts - Record: https://forck.live/items/1676-how-long-prompts-block-other-requests-optimizing-llm-performance - Subject: Platform Optimization of LLM performance, specifically how long prompts block other requests, and describes a vLLM update that implements parallel prefills with a limit on long prompt requests to reduce time-to-first-token for short prompts. ### Genspark — WORLD FIRST: Genspark Introduces Parallel Tool Calls for AI Agents! - Company: Genspark (genspark.ai) - Announced: 2025-06-12 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.genspark.ai/blog/genspark-parallel-tool-calls - Record: https://forck.live/items/18281-world-first-genspark-introduces-parallel-tool-calls-for-ai-agents - Subject: Genspark ### Cognition — Don’t Build Multi-Agents - Company: Cognition (cognition.com) - Announced: 2025-06-12 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://cognition.com/blog/dont-build-multi-agents - Record: https://forck.live/items/7700-don-t-build-multi-agents - Subject: Devin / SWE Cognition argues against multi-agent architectures for building AI agents, advocating instead for single-threaded linear agents with context compression. The principles of context engineering include sharing full agent traces and avoiding conflicting implicit decisions. ### Hugging Face — Featherless AI on Hugging Face Inference Providers 🔥 - Company: Hugging Face (huggingface.co) - Announced: 2025-06-12 - Category: availability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/inference-providers-featherless - Record: https://forck.live/items/1678-featherless-ai-on-hugging-face-inference-providers - Subject: Platform Featherless AI is now a supported Inference Provider on the Hugging Face Hub, offering serverless inference with a large catalog of open-source models. ### Hugging Face — Learn the Hugging Face Kernel Hub in 5 Minutes - Company: Hugging Face (huggingface.co) - Announced: 2025-06-12 - Category: infrastructure-release - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/hello-hf-kernels - Record: https://forck.live/items/1677-learn-the-hugging-face-kernel-hub-in-5-minutes - Subject: Platform Hugging Face announced the Kernel Hub, a platform for loading pre-compiled, optimized compute kernels directly from the Hub, simplifying the process of using custom kernels for model acceleration. ### OpenAI — Bringing the magic of AI to Mattel’s iconic brands - Company: OpenAI (openai.com) - Announced: 2025-06-12 - Category: partnership-acquisition - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/mattels-iconic-brands - Record: https://forck.live/items/540-bringing-the-magic-of-ai-to-mattel-s-iconic-brands - Subject: GPT / ChatGPT / API OpenAI and Mattel are partnering to integrate AI into iconic brands such as Barbie and Hot Wheels, aiming to enhance creative development, streamline workflows, and create new ways for fans to engage. ### NVIDIA — Post-Training Isaac GR00T N1.5 for LeRobot SO-101 Arm - Company: NVIDIA (nvidia.com) - Announced: 2025-06-11T18:27:22+00:00 - Category: model-update - Coverage: not counted - Announcement: yes - Group: models - Source: https://huggingface.co/blog/nvidia/gr00t-n1-5-so101-tuning - Record: https://forck.live/items/1679-post-training-isaac-gr00t-n1-5-for-lerobot-so-101-arm NVIDIA announced the availability of Isaac GR00T N1.5, the first major update to the open foundation model for humanoid robots, and provided a tutorial on fine-tuning it for the LeRobot SO-101 arm. ### Sakana AI — Text-to-LoRA: Instant Transformer Adaption - Company: Sakana AI (sakana.ai) - Announced: 2025-06-11T15:00:00+00:00 - Category: research-paper - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://sakana.ai/text-to-lora - Record: https://forck.live/items/4715-text-to-lora-instant-transformer-adaption - Subject: Sakana models Sakana AI introduces Text-to-LoRA, a hypernetwork that generates task-specific LoRA adapters from a text description, presented at ICML2025. ### Vercel — Designing and building the Vercel Ship conference platform - Company: Vercel (vercel.com) - Announced: 2025-06-11T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/designing-and-building-the-vercel-ship-conference-platform - Record: https://forck.live/items/20864-designing-and-building-the-vercel-ship-conference-platform - Subject: v0 / AI SDK / AI Gateway Vercel documented the design and development of its Ship 2025 conference platform, covering the use of generative AI tools (Flux, Google Veo 2, Runway, Ideogram) to generate over 15,000 images and videos during visual exploration, and the use of v0 for prototyping the site and registration flow. The team also implemented fluid motion and interaction using ray-marching rendering techniques. ### Ant Ling — Ming-Omni: A Unified Multimodal Model for Perception and Generation - Company: Ant Ling (ant-ling.com) - Announced: 2025-06-11 - Category: open-weight-release - Coverage: not counted - Announcement: yes - Group: models - Source: https://developer.ant-ling.com/en/blogs/ming-omni - Record: https://forck.live/items/22352-ming-omni-a-unified-multimodal-model-for-perception-and-generation - Subject: Ling / Ring / Ming Ant Ling released Ming-lite-omni, a unified multimodal model with 2.8 billion activated parameters that processes images, text, audio, and video and supports speech and image generation. The model is built on the Ling MoE architecture with modality-specific routers and integrates an audio decoder and Ming-Lite-Uni for generation. The authors state it is the first open-source model to match GPT-4o in modality support, and all code and model weights are released. ### Ant Ling — Open Source LLM Development 2025: Landscape, Trends and Insights - Company: Ant Ling (ant-ling.com) - Announced: 2025-06-11 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://developer.ant-ling.com/en/blogs/llm-landscape-2025 - Record: https://forck.live/items/22348-open-source-llm-development-2025-landscape-trends-and-insights - Subject: Ling / Ring / Ming Ant Ling's analysis of the open-source LLM development ecosystem as of May 2025 identifies 135 projects across 19 technical domains, with AI surpassing Cloud Native as the most influential tech domain on GitHub in 2023. Key trends include the rapid replication of closed-source tools as open-source clones, the abandonment of many AI tools (1,232 of 5,079 recorded), and the rise of agent frameworks, efficient inference engines (vLLM, SGLang), and reinforcement learning frameworks (Verl, OpenRLHF, AReaL). ### Mistral AI — Mistral Compute - Company: Mistral AI (mistral.ai) - Announced: 2025-06-11 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://mistral.ai/news/mistral-compute - Record: https://forck.live/items/3321-mistral-compute - Subject: Mistral / Le Chat Mistral AI announces Mistral Compute. ### Hugging Face — Introducing Training Cluster as a Service - a new collaboration with NVIDIA - Company: Hugging Face (huggingface.co) - Announced: 2025-06-11 - Category: product-launch - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://huggingface.co/blog/nvidia-training-cluster - Record: https://forck.live/items/1680-introducing-training-cluster-as-a-service-a-new-collaboration-with-nvidia - Subject: Platform Hugging Face and NVIDIA announced Training Cluster as a Service, a new offering that provides accessible GPU clusters for research organizations to train AI models. ### Replit — Import Figma and Other Design Tools into Agent with Replit Import - Company: Replit (replit.com) - Announced: 2025-06-10T15:50:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://replit.com/blog/import - Record: https://forck.live/items/17965-import-figma-and-other-design-tools-into-agent-with-replit-import - Subject: Replit Agent Replit launched Replit Import, allowing users to import designs from Figma, Lovable, and Bolt into Replit and transform them into applications using Replit Agent. The feature includes high-fidelity visual conversion, intelligent backend creation, and production deployment, along with built-in security screening. ### Vercel — How we’re adapting SEO for LLMs and AI search - Company: Vercel (vercel.com) - Announced: 2025-06-10T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/how-were-adapting-seo-for-llms-and-ai-search - Record: https://forck.live/items/20865-how-we-re-adapting-seo-for-llms-and-ai-search - Subject: v0 / AI SDK / AI Gateway Vercel discusses how AI search interfaces like ChatGPT and Google's AI Overviews are changing content discovery, noting that ChatGPT now refers around 10% of new Vercel signups, up from 4.8% the previous month. The post outlines strategies for adapting SEO to favor clarity, depth, and structure over traditional keyword and backlink tactics. ### Mistral AI — Magistral - Company: Mistral AI (mistral.ai) - Announced: 2025-06-10T12:00:00+00:00 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://mistral.ai/news/magistral/ - Record: https://forck.live/items/7364-magistral - Subject: Mistral / Le Chat Mistral AI announced Magistral, its first reasoning model, released in two variants: Magistral Small (24B parameters, open-source under Apache 2.0) and Magistral Medium (enterprise version). The model excels in domain-specific, transparent, and multilingual reasoning, with Magistral Medium achieving 73.6% on AIME2024. ### Supabase — Announcing Multigres: Vitess for Postgres - Company: Supabase (supabase.com) - Announced: 2025-06-10 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://supabase.com/blog/multigres-vitess-for-postgres - Record: https://forck.live/items/21355-announcing-multigres-vitess-for-postgres - Subject: Supabase Supabase announced that Sugu, the co-creator of Vitess, is joining the team to build Multigres, a new proxy for scaling Postgres databases. Multigres aims to provide sharding, connection pooling, query routing, and high availability for Postgres, with a focus on compatibility and a gradual on-ramp from simple pooling to petabyte-scale sharding. The project will be open source under the Apache 2 license. ### Genspark — Introducing Genspark AI Browser — Lightning Fast, Ad-Blocking, Full Agentic, Autopilot Mode, MCP Store, and Super Agent That Does Everything for You! - Company: Genspark (genspark.ai) - Announced: 2025-06-10 - Category: not stated - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://www.genspark.ai/blog/genspark-ai-browser - Record: https://forck.live/items/18282-introducing-genspark-ai-browser-lightning-fast-ad-blocking-full-agentic - Subject: Genspark ### Cua — From Lume to Containerization: Our Journey Meets Apple's Vision - Company: Cua (cua.ai) - Announced: 2025-06-10 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://cua.ai/blog/lume-to-containerization - Record: https://forck.live/items/16840-from-lume-to-containerization-our-journey-meets-apple-s-vision - Subject: Cua Driver / Lume / Cua Bench / Cua Fleets Cua, the company behind VM management tools for Apple Silicon, published a blog post analyzing Apple's new Containerization framework announced at WWDC. The post compares Apple's container approach with Cua's own tools Lume and Lumier, noting that Apple's framework builds on the same Virtualization foundation Cua has been using. Cua highlights potential future GPU support via a symbol called _VZPCIDeviceConfiguration in macOS Tahoe. ### Cartesia — Introducing Ink: speech-to-text models for real-time conversation - Company: Cartesia (cartesia.ai) - Announced: 2025-06-10 - Category: new-model - Coverage: not counted - Announcement: yes - Group: models - Source: https://www.cartesia.ai/blog/introducing-ink-speech-to-text - Record: https://forck.live/items/8017-introducing-ink-speech-to-text-models-for-real-time-conversation - Subject: Sonic / Ink Cartesia introduced Ink, a family of streaming speech-to-text models, with the debut model Ink-Whisper optimized for low-latency transcription in conversational settings. ### Amp — Agents for the Agent - Company: Amp (ampcode.com) - Announced: 2025-06-10 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://ampcode.com/agents-for-the-agent - Record: https://forck.live/items/7833-agents-for-the-agent - Subject: Amp Amp introduced generic subagents that can be spawned by the main agent, each with their own context window, allowing parallel task delegation. The feature is now available and leverages Claude Sonnet 4. ### Replicate — How to prompt Veo 3 for the best results - Company: Replicate (replicate.com) - Announced: 2025-06-10 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://replicate.com/blog/using-and-prompting-veo-3 - Record: https://forck.live/items/2488-how-to-prompt-veo-3-for-the-best-results - Subject: Platform This is a guide on how to write effective prompts for Google's Veo 3 video generation model, covering visual elements, audio prompting, and character consistency. ### Lovable — Lovable: Secure Vibe Coding - Company: Lovable (lovable.dev) - Announced: 2025-06-09T17:30:00+00:00 - Category: capability-change - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://lovable.dev/blog/secure-vibe-coding - Record: https://forck.live/items/12140-lovable-secure-vibe-coding - Subject: Lovable / AI app builder Lovable added three security features to its no-code app-building platform: a Security Scan that checks for issues before publishing, an AI-powered Security Reviewer that analyzes the entire app for vulnerabilities and suggests fixes, and Automatic API Key Protection that blocks around 1200 private API keys from being hardcoded per day. ### Vercel — Building secure AI agents - Company: Vercel (vercel.com) - Announced: 2025-06-09T13:00:00+00:00 - Category: not stated - Coverage: not counted - Announcement: no - Group: routine - Source: https://vercel.com/blog/building-secure-ai-agents - Record: https://forck.live/items/20866-building-secure-ai-agents - Subject: v0 / AI SDK / AI Gateway Vercel published a guide on building secure AI agents, focusing on prompt injection risks. It recommends assuming total compromise of the prompt, scoping tools to the caller's authority, and treating model output as untrusted. The post also introduces two open-source packages: harden-react-markdown and markdown-to-markdown-sanitizer. ### OpenAI — Scaling security with responsible disclosure - Company: OpenAI (openai.com) - Announced: 2025-06-09T10:00:00+00:00 - Category: safety-policy-update - Coverage: not counted - Announcement: yes - Group: announcements - Source: https://openai.com/index/scaling-coordinated-vulnerability-disclosure - Record: https://forck.live/items/541-scaling-security-with-responsible-disclosure - Subject: GPT / ChatGPT / API OpenAI announces a new Outbound Coordinated Disclosure Policy for responsibly reporting vulnerabilities in third-party software, focusing on integrity, collaboration, and proactive security.