On October 29, a French lab put a trillion parameters up for download. Mistral AI released the full weights of Mistral Large 4, nicknamed “Le Chonk,” on Hugging Face, where the checkpoint is listed as Mistral-Large-4-1T-A52B. In the corridors of European tech policy, that matters more than the joke in the name.
Made in Europe, on purpose
Mistral says it trained the model from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own European datacenters. The model is natively multimodal and fluent in more than 160 languages, including every official EU language. For organizations that need strict data privacy or must satisfy a regulator, a top-tier model they can host themselves is no longer a hypothetical. Everything else on offer is a closed API, and terms there can change without notice.
What “open” does and does not cover
The weights carry the Mistral Ecosystem Open-Weight License, a modified MIT license. Use, modify, fine-tune, commercialize: all permitted. One condition applies. A business running a Model-as-a-Service offering needs a separate commercial license from Mistral first. The announcement does not say exactly where ordinary software wrapping ends and MaaS begins, so anyone building a hosted product would be wise to ask.
Coding, security and the numbers
Mistral calls it the best open-weight model for coding. On the Artificial Analysis Coding Agent Index it scored 48%, effectively tied with GLM-5.3, the leading open-weight rival, while using 47% fewer tokens per task on average. On the Artificial Analysis Cyber Index, which tests finding, reproducing and patching vulnerabilities, it scored 51%. It resisted 97.1% of attacks on the B3 Agent Security Benchmark. Mistral also reports a 96.6% average refusal rate on malicious cyber requests, while saying legitimate defensive work goes through without needless refusals.
Running it
This is a mixture-of-experts system of roughly 1 trillion parameters, with about 52 billion active per token. Nobody will run it on a laptop. Mistral’s own view, and the likelier outcome, is that cloud inference services, research institutions and smaller vendors will use it as a base for specialized models. It is available through the Mistral API, OpenRouter, NVIDIA-hosted endpoints with an optimized NIM container, and Microsoft Foundry. API pricing is $0.70 per million input tokens, $2.10 per million output tokens and $0.07 per million cached input tokens.
The gap in the record
VentureBeat, reviewing the Artificial Analysis benchmarks, flagged a factual accuracy figure of 26%, below several rivals, and noted that Mistral’s release materials do not mention it. Teams that depend on the model for fact-based generation should plan on checking its output. The weights are live now.
Sources
- Mistral Releases 1 Trillion Parameter 'Le Chonk' / Hundreds of Textbooks in Gemini Notebook / The Day Gemma 4 Writes Poetry in Space / OpenAI Safety Report and Defense AI Rogue Incident [Creative AI Digest No.228 2026.10.31]｜CHIHIRO | AI visual artist
- Mistral’s 1-Trillion-Parameter Le Chonk Checkpoint Lands on Hugging Face
- Mistral Opens 1T 'Le Chonk' Large 4 Under Modified MIT License | AI Weekly
Photo: NOIRLab/NSF/AURA/T. Slovinský (CC BY 4.0)
