Mistral owned the news cycle in this window: trillion-parameter Large 4 previews and the open-weight Le Chonk push framed as Europe’s sovereign-AI answer to US and Chinese stacks. Under that noise: NVIDIA’s Nemotron olympiad fine-tunes, Liquid’s edge decision models, and a serious LMCache RCE in the inference stack.
Today
- Mistral Large 4 preview — API preview and open-weight positioning around ~1T total / ~49B active parameters; read one clear breakdown instead of a dozen rewrites. shattered.io
- With Mistral Large 4, Europe’s AI is (somewhat) competitive again — Competitive framing vs US hyperscaler models and what “sovereign” buyers care about. Techzine Global
- Meet Le Chonk: Mistral’s new open-weight model — “Le Chonk” as on-prem / data-sovereignty play, not just another leaderboard entry. IT Pro
- Mistral unveils Le Chonk to compete with Chinese open-weight models — Geopolitical angle on who captures value from downloadable weights. The Hindu
- One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO — NVIDIA documents competitive math/coding olympiad results from one Nemotron family — signal for “reasoning” product narratives. Hugging Face Blog
- Multimodal open d1 decision models for the edge — Liquid AI on small open decision models for edge deployment (latency and control, not datacenter scale). Hugging Face Blog
- Unpatched critical LMCache flaw (RCE) — Unauthenticated remote code execution in LMCache matters for anyone caching KV across distributed inference. The Hacker News
- China’s open AI models are winning global users — Who captures economic value when weights are free but compute and distribution are not. South China Morning Post
- New open-weight AI models could disrupt the industry — Macro take on how downloadable weights change vendor lock-in and margin. Morning Brew
- Lawmaker asks U.S. AI firms about possible Chinese theft of model weights — Policy pressure on weight security and export controls. TradingView (syndicated)
- Five keys to controlling AI token costs — Practical ops: caching, routing, batching, and model tiering for production spend. InfoWorld
- ML Engineer, AI Engineer, or LLM Engineer in 2026? — Role boundaries for teams shipping agents vs training vs inference. KDnuggets
- Researchers built a coding AI that steals secrets while it helps you — Supply-chain / malicious assistant research worth tracking for internal copilot rollouts. Startup Fortune
- Jan AI on Mac: local LLM offline, no account — Consumer-friendly on-device stack reminder for privacy-sensitive workflows. Pasquale Pillitteri
- Firefox on enterprise security, AI model choice, and open source — Browser vendor view on picking models and keeping data out of opaque SaaS. diginomica
Watch: Open-weight models
- Mistral Unveils Large 4, 1T-parameter open-weight model — Consolidated launch coverage. TechnologyOrg
- Mistral steps up open-weight competition with Le Chonk — EU vendor positioning vs Chinese OSS leaders. Silicon UK
Watch: Claude / Anthropic · OpenAI
- Mistral AI takes on OpenAI, Anthropic with 1T-parameter model — Competitive set for frontier API buyers comparing US closed models vs Mistral previews. Analytics India Magazine
Built with utilities/ai_report_build.py (24-hour lookback). Edited for publication. About the AI Report.