Mistral Large 4 ships as open weights for coding and cyber
Original title: [AINews] Building AIE CODE: The top Coding Agent Conference
The 1.05T model has 52B active parameters and a 1M context; Mistral's ranking claims are narrower than independent scores.
Mistral published updated open weights for Mistral Large 4, nicknamed Le Chonk, aimed at agentic coding and cybersecurity and customizable through Mistral Forge. Artificial Analysis lists 1.05T total and 52B active parameters, a 1M context, and text plus image input, with pricing of $0.70/$2.10 per 1M input/output tokens. Mistral says it ranks first among open models on Cognition's FrontierCode; Artificial Analysis scores it 26 on Intelligence Index v5, still 6 points behind DeepSeek V4.1 Flash 1017 and GLM-5.3. Day-0 support includes vLLM, Unsloth GGUF quants, TRL fine-tuning, and SkyRL.
Unconfirmed: A researcher claims four zero-days found in a messaging app in two hours; the note is cut off and unverified.
- The MEOW license needs a separate commercial license for model-as-a-service use.
- A researcher reports a 64% preference win versus GLM 5.3 and Kimi K3.
- Hallucination rate on AA-Omniscience fell from 41% to 28%; accuracy is 26%.
- Unsloth's 1-bit GGUF is 223GB versus 1.1TB in FP8.
- The same newsletter also promotes the invite-only AIE CODE summit.
- Vendors
- Mistral
- Products
- Mistral ForgevLLMUnsloth
- Organizations
- Artificial AnalysisCrowdStrikeCloudflareZscaler
From the original article
14 min read at latent.spaceMistral’s open-weight release, frontier monitoring, and the day’s AI Twitter and Reddit recaps.
By swyx .ai · latent.space
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.