Mistral Large 4 "Le Chonk" public preview: 1.05T-total / 49B-active multimodal MoE, 1M context, open weights by month-end

Mistral opened a public-preview API for Mistral Large 4 ("Le Chonk") on 2026-10-06, with open weights promised by the end of October. Per the official docs it is a granular Mixture-of-Experts with 1.05T total and 49B active parameters, a 1.6B vision encoder, native multimodality and a 1M-token context window. It was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters on data spanning 160+ languages, and post-trained with large-scale asynchronous RL over composable environments.

Vendor-reported results (coding numbers privately evaluated by Artificial Analysis): DeepSWE v1.1 61.7%, SWE-Atlas-QnA 59.4%, Terminal-Bench 4 28.3%, Coding Agent Index 49.8% (ahead of DeepSeek V4 Pro 0813 and Qwen3.8 Max); Cybench 93% and 82% on the AA Cyber Index reproduce-and-patch test, the highest of any model; AutomationBench 59.9%; AA-Briefcase 1,393 Elo; Dense 200 visual grounding 42% vs GPT-6 Astra 41%; Lakera B3 attack resistance 93.3%. Independent verification awaits the weight release.

The model id is mistral-large-4 via Mistral Studio/Console, with chat completions, function calling, structured outputs, batch and agents/conversations built-in tools. Pricing is $1.36 input / $0.14 cached / $4.18 output per 1M tokens, with the docs also listing a half-price tier.