Mistral releases improved Mistral Large 4 open weights for coding and cybersecurity
Mistral AI released open weights for an improved Mistral Large 4 (Le Chonk), its 1-trillion-parameter multimodal model previewed earlier this month. Mistral reports a 48% Artificial Analysis Coding Agent Index score with 47% fewer tokens per task than GLM-5.3, alongside three reasoning modes. The Mistral Ecosystem Open-Weight license permits use, modification, fine-tuning, and commercialization, but businesses operating Model-as-a-Service offerings must obtain a separate commercial license before commercial use.
Model releases and updates
Mistral Large 4 arrives on Hugging Face with multimodal capabilities and open weights
Mistral Large 4, also called Le Chonk, is a multilingual MoE model that accepts text and images and generates text, with about 1T total parameters and a 1M-token context window. Hugging Face details its FP8 and NVFP4 variants, cybersecurity benchmarks, and launch-day inference and fine-tuning support. The weights use the Mistral Ecosystem Open-Weight License, which limits model-as-a-service offerings.
Product releases and updates
宝玉 · XReported by 1 other sourcesOpenAI doubles Pro and Business Premium usage limits and cancels Pro 200 reduction
OpenAI is doubling all usage limits for Pro 100/200/500 and Business Premium effective today, according to @thsottiaux, while canceling the planned Pro 200 reduction. Baoyu calculates allowances relative to Plus at 10x for Pro 100, 20x for both existing and newer Pro 200 subscribers, and 50x for Pro 500; no price change was mentioned. Extended limits for dots will continue for now, with dot usage remaining separate from default usage; no extension end date was specified.
Introducing AnySpeechRecognizer and AnyDecisionModel for Speech and Decisions on Apple Platforms
The authors announce two Swift packages for Apple platforms, AnySpeechRecognizer and AnyDecisionModel. AnySpeechRecognizer mirrors Apple's speech APIs while adding local Parakeet and Qwen3-ASR engines and speaker diarization for up to four speakers; AnyDecisionModel returns typed answers and probabilities for yes-or-no, multiple-choice, and rating-scale questions through local or hosted backends. Both support downloading open models from the Hugging Face Hub, and the reported speech evaluations cover macOS only, with no iPhone testing yet.
Gemini App · XReported by 1 other sourcesGemini adds document, PDF, presentation, and spreadsheet generation and editing
You can now generate, preview, and edit documents, PDFs, presentations, spreadsheets, and more directly in Gemini. Turn handwritten notes into a PDF study guide, manually tweak content, highlight sections for Gemini to edit text or visuals, and ask follow-up questions.
Industry
宝玉 · XReported by 1 other sourcesHow Claude’s self-assessment flags potential misuse after responses
Baoyu explains Anthropic’s self-assessment system, which has Claude assess conversations for potential misuse after responding. Anthropic says it adds less than 2% in inference compute, incurs no customer charges, and collects no additional user data; traffic under zero data retention agreements is currently excluded. In sampled Fable 5 traffic, it detected 78% of conversations flagged by a separate evaluator at a 1% false-positive rate. Flags support subsequent review rather than immediate blocking, and an individual flag does not directly trigger account suspension.
Research
Epoch AI · XEpoch AI finds GLM 5.3 can autonomously exploit Firefox’s patch gap
Epoch AI reports that GLM 5.3 autonomously built successful exploits for six recent Firefox versions using public security fixes not yet shipped to users, costing about $1,000 per compromised version. The exploits could steal browser passwords and authentication cookies or read local files after a user opened one webpage. Every successful attack also used a previously unpublished vulnerability that Mozilla patched in its October 27 Firefox 159.0 release; Epoch AI has not confirmed whether the attacks still succeed without it.
The DecoderExperts adopt AI agents’ randomly assigned preferences in Anthropic study
An Anthropic experiment with 50 causal inference experts found that participants adopted an AI agent’s randomly assigned preferences in 86 percent of four analytical decisions. Those choices shaped productivity estimates and subsequent recommendations, while most participants felt the final recommendation was their own. The agent’s direct influence on the final recommendation was not statistically significant. AI users also finished faster and made fewer serious errors, but the small, self-selected comparison group limits causal conclusions about those benefits.
腾讯混元 · XTencent Hunyuan shares Hy-Nomas technical report, open sources Hy-Nomas-Bench
Tencent Hunyuan says the Hy-Nomas technical report was released last Friday: https://t.co/QoPTiJKsrP Hy-Nomas-Bench is also open sourced for reproducible research: https://t.co/3tCUyLl0f9 The team also recreates Genie 4 examples with Hy-Nomas and shares a full comparison video, Hy-Nomas vs Genie 4. Try it: https://t.co/TffftO8Xjt https://t.co/wzfGd4pCqh
Tips and perspectives
How DeepInfer Coordinates Asynchronous Scheduling for DeepSeek-V4.1-Flash Inference
DeepInfer, the inference engine behind DeepSeek-V4.1-Flash, overlaps CPU scheduling with GPU computation to reduce decoding idle time. Leiphone explains how its asynchronous scheduler handles in-flight tokens, sequence state corrections, deferred resource release, and coordination across expert-parallel GPUs. The open release currently supports offline inference, with some online infrastructure unavailable; DeepSeek has not separately published complete end-to-end speedup measurements isolating the scheduler’s contribution.
AWS Machine Learning BlogFirst-partyDeploy NVIDIA Nemotron Models on Amazon EKS and SageMaker HyperPod EKS
AWS explains how to deploy NVIDIA Nemotron 3.5 Super VL on Amazon EKS, EKS Auto Mode, and SageMaker HyperPod EKS using NVIDIA Dynamo and vLLM. The guide covers the model’s 120B total parameters and 12B active parameters per token, deployment architecture, and validation on B200 and H200 instances. It also shows how to use aiperf concurrency sweeps to compare aggregated and disaggregated deployments and plan capacity against latency targets.
AWS Machine Learning BlogFirst-partyConfigure AWS Batch quota sharing for SageMaker Training jobs across teams
AWS explains how to configure quota management for SageMaker Training jobs using AWS Batch and the SageMaker Python SDK v3. The walkthrough sets up three quota shares with LEND_AND_BORROW, LEND, and RESERVE strategies, then demonstrates reclaiming lent capacity through cross-share preemption. It also shows how updating a submitted job’s priority can trigger in-share preemption so higher-priority work runs sooner.
Tomorrow at 08:00, a briefing like this arrives in your inbox.
The same kind of briefing as the one above: a few stories worth reading. Free. Mail starts after you confirm, and every letter has a one-click unsubscribe. Read a past issue · 2026-10-29