Mistral previews Large 4, a 1-trillion-parameter model, and says the weights come out by the end of the month

Mistral launched a public preview of Mistral Large 4 on 6 October. It describes the model as “a 1 trillion-parameter natively multimodal model with 49 billion active parameters”. The preview API costs $1.36 per million input tokens and $4.18 per million output tokens, according to the company's price table.

The weights are not out yet. “We will release the weights by the end of the month,” Mistral writes. “Until then, we are red-teaming the model in real-world settings with cybersecurity leaders, vetted partners, and state authorities, who will access the same model with reduced moderation and expanded cyber capabilities.” The post names no licence for the weights.

The benchmark figures are Mistral's own. It reports 61.7% on DeepSWE v1.1 and says the model solves 93% of the challenges in Cybench. It says its combined Coding Agent Index score of 49.8% places it ahead of DeepSeek V4 Pro 0813 and Qwen3.8 Max, two Chinese open-weight models. Reuters reported the chief executive saying the model beats Chinese ones in some areas; the Reuters article was not opened.

It is the second large open-weight model announced in two days with the weights promised for later this month, after Reflection's Beam on 5 October.

• Confirmed Public preview launched 6 October: 1 trillion parameters, 49 billion active, at $1.36 in and $4.18 out per million tokens. Mistral
• Claimed Weights to be released by the end of the month; until then, cyber firms, partners and state authorities red-team a version with reduced moderation. Mistral
• Claimed A Coding Agent Index score of 49.8% puts it ahead of DeepSeek V4 Pro 0813 and Qwen3.8 Max, by Mistral's own measure. Mistral