✷ The archive · 11 labs · 3,412 dispatches
The labs
No more opening a dozen official blogs every morning. First-hand releases from OpenAI, Anthropic, DeepMind and the rest, each with its substance pulled out.
OpenAI Announces Harvey Powered by GPT-6 Astra for Enhanced Legal Drafting
OpenAI released Harvey, a legal‑workflow AI platform upgraded with GPT‑6 Astra, which improves context handling, document formatting, and integrates lawyer preferences to produce higher‑quality drafts.
OpenAI GPT-6 Astra boosts Invideo color grading 3x
OpenAI announced that its GPT-6 Astra model enables Invideo to improve color‑grading success rates threefold, dramatically speeding up complex video edits while keeping editors in control.
Sam Altman’s UN Security Council Remarks on AI Governance
OpenAI CEO Sam Altman told the UN Security Council that AI must stay under human control and be governed through international cooperation, warning that unchecked AI could concentrate power or become uncontrollable.
OpenAI Introducing MentalHealthBench
OpenAI has released MentalHealthBench, an open benchmark co-created with over 80 global mental health experts to evaluate how AI systems handle a full spectrum of mental health conversations, from everyday stress to emergencies.
ChatGPT Ads expands to Southeast Asia and Taiwan
OpenAI has expanded ChatGPT Ads to Indonesia, Malaysia, the Philippines, Singapore, Thailand, Vietnam, and Taiwan, bringing the service to over 60 countries.
Airbnb expands access to GPT-6 Astra and OpenAI frontier models
Airbnb has entered a new agreement with OpenAI to provide its engineering and product teams with broader access to frontier models, including GPT-6 Astra, to accelerate feature shipping and product development.
Grab and OpenAI launch GO Forward with AI programme for Southeast Asian partners
OpenAI and Grab announced the GO Forward with AI programme, a two‑year initiative to train 30,000 driver, delivery and merchant partners across Southeast Asia in practical AI skills using ChatGPT and related tools.
Claude discovers array-associated reverse transcriptases (ART)
Anthropic announced that its Claude AI autonomously identified a novel enzyme system called array-associated reverse transcriptases (ART), a CRISPR‑like repeat array linked to a reverse transcriptase in bacteriophages, demonstrating AI‑driven hypothesis generation in biology.
OpenAI GPT-6 Prompt Caching Updates
OpenAI has introduced an improved prompt caching system for GPT-6 that increases cache hit rates and provides developers with new monitoring and optimization tools.
OpenAI introduces GPT-6 Sol and Luna
OpenAI announced GPT-6 Sol and GPT-6 Luna, cheaper, faster variants of GPT-6 Astra that cut API prices by 50% and deliver strong performance across professional work, factuality, coding, and computer use.
GPT-6 Astra: Reducing Research Time and Cost for Parallel
OpenAI announces that Parallel has reduced research task time and cost by 50% using GPT-6 Astra, achieving high-quality results with fewer search calls and fewer tokens.
Jun Kim joins Hugging Face to support oMLX and the MLX community
Hugging Face has hired Jun Kim, the creator and maintainer of oMLX, to accelerate the development of local AI tools optimized for Apple Silicon.
Hugging Face Transformers GGUF Support
Hugging Face has added support for running GGUF quantized models directly in the transformers library, enabling efficient local inference on Apple Silicon using ggml kernels.
OpenAI Priorities and Principles for Effective Third-Party Assessments
OpenAI announced a set of four priority areas and accompanying principles to guide independent third‑party safety assessments of frontier AI models, aiming to improve rigor, transparency, and accountability.
EvalEval and UK AISI Release Open Evaluation Cards for Frontier LLM Benchmarks
EvalEval and the UK AI Security Institute (AISI) have published verified Evaluation Cards for five major benchmarks covering six frontier LLMs, enabling reproducible and transparent model evaluation.
vllm-metal v0.28.0 release notes / what's new
vllm-metal brings vLLM's scheduler, paged KV cache, and OpenAI-compatible server to Apple Silicon, enabling efficient concurrent serving of LLMs via MLX and Metal.
Reducing vLLM startup time with uv bytecode compilation
vLLM improved container startup performance by shipping precompiled Python bytecode using the UV_COMPILE_BYTECODE=1 flag during image build, resulting in significant time savings.
How SpaceXSI Uses Grok Bot for Customer Support: 175% More Tickets, No New Hires
xAI published a first-party case study describing how SpaceXSI uses Grok Bot to absorb a 175% increase in support tickets without new hires, claiming per-resolution costs of $0.20 to $0.30 versus the $1 to $4 charged by traditional AI support tools.
LLM Compression by Block Removal with Constrained Binary Optimization
Hugging Face introduced a new block‑removal pruning method that formulates transformer block selection as a constrained binary optimization (Ising) problem, achieving up to 23 MMLU points improvement over prior techniques at 50% depth compression.
OpenAI Announces Independent Mathematics Advisory Group
OpenAI has formed an independent advisory group of leading mathematicians to guide the responsible communication and deployment of its rapidly advancing math‑solving AI models.
Higgsfield AI launches GPT-6 Astra video features for rapid ad creation
Higgsfield AI announced that its integration of OpenAI's GPT-6 Astra enables single‑engineer development of new video ad creation tools in a day, dramatically speeding up creative workflows for small businesses.
OpenAI Building Standards for the Next Phase of AI
OpenAI proposes the development of global technical standards for frontier AI, specifically targeting recursive self-improvement (RSI) to ensure safety and human control as AI systems automate AI research.
OpenAI Academy Expanded Learning Paths
OpenAI has expanded its Academy with specialized learning paths for developers, leaders, educators, and college students to help organizations build shared AI habits and technical skills.
vLLM PD Serving of Qwen3.8-2.4T achieves 5K TPS/GPU and 180 gen tok/s/user
vLLM announced that its PD serving of the Qwen3.8-2.4T model reaches 5,000 total token throughput per GPU and 180 generated tokens per user, demonstrating a full pareto frontier for high‑throughput and low‑latency scenarios.
Hugging Face tokenizers v1 release notes
Hugging Face has released a release candidate for tokenizers v1, which achieves 3 to 30 times faster encoding speeds than v0.23 through SIMD-based splitting, a word cache, and allocation-free merge loops.
V7 Go and OpenAI Model Integration for Institutional Memory
V7 Go leverages OpenAI's GPT-5.6 and GPT-6 Astra models to create a Context Graph that provides AI agents with structured institutional memory for complex enterprise workflows.
xAI Grok 4.7 Release Notes
xAI has released Grok 4.7, a model optimized for long-running coding and professional knowledge work with improved safeguards and price-performance competitiveness.
Qwen-Image-2.1 release notes / what's new
Qwen-Image-2.1 is a unified 7B parameter image model that integrates text-to-image generation, native transparency support, and advanced image editing capabilities into a single efficient architecture.
OpenAI Introduces Australian Youth Safety Blueprint
OpenAI has launched the Australian Youth Safety Blueprint, a six-pillar roadmap designed to protect young people aged 13-17 using AI through enhanced safeguards, literacy, and accountability.
Qwen3.8-LiveTranslate model release
Qwen announced Qwen3.8‑LiveTranslate, a real‑time simultaneous interpretation model that reduces average lagging to 2.3 seconds, adds speaker separation, bilingual screen output, and long‑context disambiguation across 60 input languages.
Anthropic and Accenture Partner on Embedded Evaluation for Frontier AI
Anthropic is partnering with Accenture to implement embedded evaluation, granting independent evaluators deep internal access to verify safety commitments and model development.
xAI Grok Voice Transcribe 2.0: Speech-to-Text Model Release Notes and What's New
xAI released Grok Voice Transcribe 2.0 on September 18, 2026, a speech-to-text model it says is twice as accurate as version 1.0 at the same price, ranking first among 32 streaming models on the Artificial Analysis leaderboard.
vLLM Scaling Multi-GPU Video Captioning with PyNvVideoCodec
vLLM now supports NVIDIA hardware video decoding via PyNvVideoCodec, removing CPU bottlenecks and more than doubling throughput for multi-GPU video captioning tasks on H100 GPUs.
Qwen3.8-Omni-Flash Release: Omnimodal Agentic Model with 1M-Token Context
Qwen announced the Qwen3.8-Omni-Flash model, a next‑generation omnimodal LLM with a 1 million‑token context window, agentic capabilities for audio‑visual tasks, and dramatically lower audio‑visual API costs.
Cooley GO Public: Accelerating IPO Workflows with ChatGPT Work
Law firm Cooley has developed GO Public, a proprietary agentic AI product built on ChatGPT Work to automate the synthesis of IPO data and accelerate the preparation of capital markets transactions.
Anthropic Life Sciences Verification Program
Anthropic has launched the Life Sciences Verification Program (LSVP) to provide verified life science professionals with more permissive safeguards for biology-related work using Mythos, Opus, and Sonnet models.
Claude accelerates open‑source biomolecular models 4× faster and enables >10k‑token predictions on a single GPU
Anthropic announced that Claude optimized more than 30 open‑source biomolecular models, delivering roughly 4× speed‑ups, a low‑memory “Big” mode for >10,000‑token systems on one GPU, and a $1 M protein‑design competition.
OpenAI Astra for Law Release
OpenAI has introduced Astra for Law, a legal-specialized foundation combining GPT-6 Astra with a massive legal search index and professional-grade privacy controls for law firms and legal tech companies.
OpenAI Model Misalignment Reporting Framework
OpenAI has introduced a systematic framework to expedite the disclosure of model misalignment instances, accompanied by six initial reports of unexpected model behaviors.
OpenAI launches Older Adults AI Skills Jam to teach ChatGPT use
OpenAI announced the Older Adults AI Skills Jam, a free in‑person program that teaches seniors how to use ChatGPT safely and effectively, expanding AI accessibility for an aging population.
OpenAI Reimagining Advertising with AI
OpenAI is introducing Sponsored Agents, AI-powered campaign management tools, and integrations with HubSpot and Shopify to build an AI-native advertising platform.
OpenAI Admin Console Analytics: Connecting AI Usage to Business Value
OpenAI announced new analytics features in the ChatGPT Admin Console that combine usage, cost, task, and outcome data to help admins quantify AI’s business impact and guide investment decisions.
Hex integrates GPT-6 Astra for advanced data visualization and analysis
Hex has integrated GPT-6 Astra to enable the creation of complex, interactive data artifacts and improve analytical judgment in data reporting.
OpenAI Unlocking New Ways of Working Report Shows Cross‑Occupation AI Use Becomes Recurring
OpenAI’s new research report finds that workers increasingly incorporate cross‑occupation AI tasks into regular workflows, indicating early job expansion before title changes.
Mistral AI partners with Mozilla to power Firefox Smart Window with open, private multilingual models
Mistral AI announced a partnership with Mozilla to integrate its open-weight, multilingual models into Firefox Smart Window, delivering privacy‑first, locally‑fine‑tuned AI assistance in the browser.
Memory in Grok Build: xAI Adds Persistent Project Notes to Its Coding Agent
xAI's Grok Build now has memory: after every completed turn it writes durable notes on conventions, decisions, and project facts, then reads the relevant notes in later sessions, with /memory to browse and /dream to organize them into topic files.
Gemini 3.8 Live and 3.8 Live Extended Thinking Release
Google DeepMind has introduced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, advanced live dialogue models designed for real-time voice agents with enhanced reasoning and visual grounding.
ALTK-Evolve Consistency Guidelines for Agent Reliability
Hugging Face and IBM Research introduce a Consistency Analyzer and consistency guidelines to reduce the gap between average agent accuracy and reliable, repeatable success.
vLLM Kimi-K3 DSpark Speculative Decoding Implementation
vLLM has implemented a DSpark speculator for the 2.8T-parameter Kimi K3 model, increasing math reasoning interactivity from 110 to 435 tok/s/user and delivering up to 3.5x higher output throughput.
vLLM and Novita AI Release Chord: Faster INT4 MoE Kernels for Kimi K2.x
Novita AI open‑sourced Chord, a W4A16 MoE CUDA operator that speeds up INT4‑quantized Kimi K2.x serving by up to 2.15× over public Humming kernels.