✷ The archive · 11 labs · 3,414 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.
PRX Data Strategy: Scaling Pre-training with VLM Re-captioning and Mosaic Streaming
Photoroom details the data pipeline for PRX, emphasizing the use of long VLM-generated captions, a hybrid Lance and Mosaic Data Shards storage strategy, and high-quality JPEG encoding to optimize a 7B parameter model.
vLLM × HPC-Ops: High-Performance Attention and MoE Backends from Tencent Hunyuan
vLLM now includes HPC-Ops attention and MoE backends optimized for NVIDIA Hopper H20, delivering up to 2.95× attention speedup and 1.59× MoE speedup, cutting TTFT by about 24% and TPOT by about 17% on Hy3.
Hugging Face Kernels Major Updates
Hugging Face announced major updates to its Kernels project, introducing a new kernel repository type, trusted publishers and code signing, revamped CLIs, broader framework support, and foundations for agentic kernel development.
Anthropic discovers a global workspace (J-space) in Claude language models
Anthropic announced the discovery of a emergent “J-space” in Claude that functions as a global workspace for internal, reportable reasoning, enabling silent thought monitoring and new safety tools.
Alberta Government Uses Claude Code to Scan and Fix 466M Lines of Code in Hours
The Government of Alberta deployed Anthropic's Claude Code (Opus and Sonnet) to scan 466 million lines of code across 3,400 repositories in 20 hours, automatically generate fixes, and establish continuous AI‑driven security reviews, dramatically reducing a task that would have taken years.
xAI Grok Voice Update: 21 New Flagship Voices
xAI has released 21 new multilingual flagship voices for Grok Voice, alongside improvements to the original five voices and the introduction of the Grok Voice Agent Builder.
Google DeepMind and A24 Research Partnership
Google DeepMind and A24 have entered a research partnership to integrate AI innovation directly into the creative process to develop new filmmaking workflows and techniques.
Leanstral 1.5 Release: State‑of‑the‑art Formal Verification Model
Mistral AI released Leanstral 1.5, a 119 B‑parameter, Apache‑2.0 model that saturates miniF2F, solves 587 of 672 PutnamBench problems, and discovers previously unknown bugs in open‑source code.
Claude Fable 5 Cybersecurity Safeguards and Jailbreak Framework
Anthropic has detailed the safety classifiers used to prevent cybersecurity misuse in Claude Fable 5 and proposed a new Cyber Jailbreak Severity (CJS) framework to standardize how AI jailbreaks are assessed.
Anthropic Restores Access to Claude Fable 5 and Mythos 5
Anthropic announced that Claude Fable 5 and Mythos 5 are back online after U.S. export controls were lifted, detailing new safety classifiers, an industry jailbreak framework, and expanded government collaboration.
Claude Fable 5 and Claude Mythos 5 launch – state-of-the-art Mythos-class model with safety safeguards and tiered access
Anthropic released Claude Fable 5, a general‑use Mythos‑class 1 model with record performance across software engineering, vision, and science, and Claude Mythos 5, a lifted‑safeguard version for trusted cyber‑defenders and future biology partners.
BAIR 2026 Graduate Showcase
The Berkeley Artificial Intelligence Research (BAIR) Lab announced its class of 2026 Ph.D. graduates, highlighting research across robotics, large language models, AI safety, and healthcare.
Hugging Face and Cerebras Real-Time Voice AI with Gemma 4
Hugging Face and Cerebras have developed an open, cascaded speech-to-speech pipeline using Gemma 4 31B to enable natural, low-latency voice AI interactions.
vLLM-Omni Optimizations for Qwen3-Omni-30B-A3B-Instruct Serving
vLLM-Omni serves Qwen3-Omni-30B-A3B-Instruct via a three-stage pipeline (Thinker, Talker, Code2Wav) and improves throughput and latency using stage decomposition, CUDA Graphs, async chunk handoffs, async output, stage replicas, and hot‑path cleanup.
xAI Voice Agent Builder Beta Release
xAI has launched the Voice Agent Builder in beta, a no-code platform for creating production-ready voice agents powered by Grok Voice.
ScarfBench: Benchmarking AI Agents for Enterprise Java Framework Migration
IBM Research introduces ScarfBench, an open benchmark to evaluate AI agents' ability to migrate enterprise Java applications across Spring, Jakarta EE, and Quarkus frameworks.
Nano Banana 2 Lite and Gemini Omni Flash Release
Google DeepMind has released Nano Banana 2 Lite for high-speed, cost-efficient image generation and Gemini Omni Flash for high-quality video generation and conversational editing.
Why Specialization Is Inevitable in AI Systems
Based on the 2026 research by Goldfeder et al., specialization is a structural necessity for AI performance because finite resources make concentrated capacity more effective than broad generality.
OpenAI Signals: Analysis of Global ChatGPT Adoption Trends
OpenAI Signals data reveals that ChatGPT usage is deepening and expanding globally, with users increasing their daily message volume by 50% and doubling the variety of tasks performed six months after signup.
Anthropic Frontier Red Team: Stress-Testing AI Capabilities
Anthropic's Frontier Red Team conducts evidence-based analysis on AI's implications for cybersecurity, national security, and autonomous systems to anticipate future risks.
Hugging Face Community Evals and Every Eval Ever (EEE) Integration
Hugging Face has integrated Community Evals with the Every Eval Ever (EEE) project to enable cross-posting of standardized evaluation results and direct linking between model pages and detailed technical records.
OpenAI Core Dump Epidemiology: Fixing an 18-Year-Old libunwind Bug
OpenAI identified and resolved two distinct crash populations in its Rockset data infrastructure, including a silent hardware failure and a rare 18-year-old race condition in GNU libunwind.
OpenAI Introduces GeneBench-Pro for Computational Biology Reasoning
OpenAI has released GeneBench-Pro, a research-level benchmark designed to evaluate the high-level scientific judgment and iterative analysis capabilities of AI agents in computational biology.
Inside Genebench-Pro: Case Studies in Genomic AI Benchmarking
OpenAI's Genebench-Pro provides a specialized benchmark for evaluating AI models on complex genomic tasks, featuring ten detailed case studies across somatic oncology, functional genomics, and population genetics.
Claude Science AI Workbench Release
Anthropic has launched Claude Science, an AI workbench that integrates scientific tools, manages scalable compute, and provides auditable research artifacts to accelerate scientific discovery.
DiScoFormer: One transformer for density and score, across distributions
DiScoFormer is a new transformer-based model that estimates both the density and score of a distribution from a set of data points in a single forward pass without requiring retraining for new distributions.
OpenAI Mapping Europe’s AI Workforce Opportunity Report 2026
OpenAI’s 2026 report extends its AI Jobs Transition Framework to the EU, finding that about 12% of EU employment may grow with AI, 14% faces higher near‑term automation potential, 27% is likely to reorganize, and 47% sees less immediate change.
vLLM Semantic Router: Enhancing Model Performance via Micro-Agent Collaboration
vLLM introduces the Semantic Router, a serving-layer primitive that transforms a single model API call into a bounded collaboration of micro-agents to outperform frontier models on complex benchmarks.
Ollama 0.31: Faster Gemma 4 Performance via Multi-Token Prediction
Ollama 0.31 introduces multi-token prediction for Gemma 4 on Apple Silicon, increasing token generation speed by nearly 90% on coding-agent benchmarks without altering model output.
HP Inc. and OpenAI Strategic Partnership
HP Inc. is scaling its strategic partnership with OpenAI using the OpenAI Frontier platform to deploy AI agents and workflows across customer support, security, and software development.
GPT-5.6 Sol preview: capabilities, safeguards, and availability
OpenAI previewed GPT-5.6 Sol, its flagship next‑generation model, alongside Terra and Luna, highlighting stronger coding, biology, and cybersecurity capabilities with enhanced safeguards and a limited release to trusted partners.
Hugging Face Jobs: Deploying vLLM Servers with a Single Command
Hugging Face now allows users to deploy private, OpenAI-compatible vLLM endpoints on its infrastructure using a single command via HF Jobs, providing a pay-per-second alternative for testing, evaluations, and batch generation.
Anthropic Economic Index Cadences Report – June 2026
Anthropic’s June 2026 Economic Index report reveals that Claude usage now follows daily and weekly work cadences, produces higher‑value artifacts that consume more compute, and that users who automate more tasks are surprisingly more optimistic about AI’s impact on their jobs.
OpenAI Codex Agentic AI Adoption and Impact
OpenAI reports a shift in knowledge work from short chatbot interactions to long-horizon agentic tasks via Codex, with non-developer adoption growing up to 189x since August 2025.
Gemini 3.5 Flash Computer Use Integration
Google DeepMind has integrated computer use as a native tool in Gemini 3.5 Flash, enabling agents to see, reason, and take action across browser, mobile, and desktop environments.
NVIDIA NeMo AutoModel Accelerates Fine-Tuning of Mixture-of-Experts Models
NVIDIA NeMo AutoModel provides 3.4-3.7x higher training throughput and 29-32% lower GPU memory usage for fine-tuning Mixture-of-Experts models by building on Hugging Face Transformers v5 with Expert Parallelism, DeepEP, and TransformerEngine kernels, requiring only a one-line import change.
Mistral AI introduces granular admin controls, scoped API keys, and multi‑account connectors
Mistral AI announced new connector features—including GA admin controls, scoped API keys, multi‑account connectors, a Debugger, and integrations in Workflows and Vibe Code—to improve security and reliability for enterprise AI workloads.
OpenAI and Broadcom Unveil Jalapeño LLM Inference Chip
OpenAI and Broadcom have introduced Jalapeño, a custom-designed AI accelerator optimized specifically for LLM inference to improve performance per watt and reduce compute costs.
Hugging Face FFASR Leaderboard: Benchmarking Far-Field ASR
Hugging Face and Treble Technologies have launched the FFASR Leaderboard, the first open community-driven benchmark to quantify the performance gap between near-field and far-field Automatic Speech Recognition (ASR) in realistic acoustic environments.
GPT-5 Pro in Immunology: Solving T-Cell Specialization Mysteries
Immunologist Derya Unutmaz used GPT-5 Pro to solve a three-year-old mystery regarding how deoxyglucose affects T-cell specialization, demonstrating the model's ability to generate novel biological insights and predict unpublished experimental outcomes.
OpenAI and the Appia Foundation: Establishing Shared Standards for Advanced AI
OpenAI has helped found the Appia Foundation to develop open, modular specifications that translate international AI standards into practical assessment criteria to ensure interoperability across organizations and jurisdictions.
Mistral OCR 4 release notes / what's new
Mistral AI has released Mistral OCR 4, a high-performance document parsing model supporting 170 languages and providing structured outputs including bounding boxes, block classification, and confidence scores.
Qwen-AgentWorld release: language world model for seven domains and its impact on general agents
Qwen releases Qwen‑AgentWorld, a language world model that simulates seven agent environments and improves general agents via controllable simulation and unified next‑state prediction.
vLLM-Omni TTS Inference Engineering
vLLM-Omni engineered TTS inference for models like Qwen3-TTS, VoxCPM2, Higgs Audio V3, and Fish Speech S2 Pro by applying model-specific optimizations that decouple latency and throughput bottlenecks, significantly improving audio throughput and reducing end-to-end latency.
Experimenting with the Cross-Origin Storage API in Transformers.js
Hugging Face shows how Transformers.js can use the experimental Cross-Origin Storage API to cache model and Wasm resources by hash, eliminating duplicate downloads across origins.
Omio Conversational Travel and AI-Native Operations
Omio is leveraging OpenAI models to transition from search-based travel planning to conversational commerce and reducing product development time to approximately 20% of previous levels.
Hugging Face huggingface_hub Release Automation
Hugging Face has transitioned from a 4-6 week release cycle to a weekly cadence for huggingface_hub by implementing an AI-driven, human-in-the-loop CI/CD pipeline using open-source tools and open-weights models.
PP-OCRv6 release: 50-Language OCR from 1.5M to 34.5M Parameters
PaddlePaddle has released PP-OCRv6, a scalable OCR model family supporting 50 languages with parameter counts ranging from 1.5M to 34.5M.
Daybreak: Tools for securing every organization in the world
OpenAI announced an expansion of Daybreak, releasing an updated Codex Security plugin, the full version of GPT‑5.5‑Cyber, a cyber partner program, and the Patch the Planet initiative to help defenders find, validate, and patch vulnerabilities at machine speed.
OpenAI Patch the Planet Initiative
OpenAI has launched Patch the Planet, a Daybreak initiative in partnership with Trail of Bits to use AI-assisted security research and human review to identify and patch vulnerabilities in critical open-source software.