Mistral releases Mistral Large 4
Mistral announced Mistral Large 4, nicknamed 'Le Chonk', on October 6, 2026, with a preview available through its API and open weights promised by the end of the month. The model scores 38 on Artificial Analysis, far above Mistral Large 3's 9, and Mistral bills it as one of the strongest models for cybersecurity. Community testing puts it roughly six months behind the frontier, with 50% off pricing in OpenCode for two weeks.
Mistral announced Mistral Large 4, nicknamed 'Le Chonk', on October 6, 2026, with a preview available through its API and open weights promised by the end of the month. The model scores 38 on Artificial Analysis, far above Mistral Large 3's 9, and Mistral bills it as one of the strongest models for cybersecurity. Community testing puts it roughly six months behind the frontier, with 50% off pricing in OpenCode for two weeks.
Mistral AI released Mistral Large 4 (ML4), nicknamed Le Chonk: a one-trillion-parameter, natively multimodal model with 49B active parameters, available now via API in public preview with open weights due October 27 after safety testing. Mistral claims it is the best open-weights model from Europe or the US, optimized for cyber defense, coding, finance, and manufacturing. The launch follows Mistral's $3.3 billion raise at a $24 billion valuation and positions the lab as a sovereign alternative amid US restrictions on frontier model access.
Mistral AI released Mistral Large 4 (ML4), nicknamed Le Chonk: a one-trillion-parameter, natively multimodal model with 49B active parameters, available now via API in public preview with open weights due October 27 after safety testing. Mistral claims it is the best open-weights model from Europe or the US, optimized for cyber defense, coding, finance, and manufacturing. The launch follows Mistral's $3.3 billion raise at a $24 billion valuation and positions the lab as a sovereign alternative amid US restrictions on frontier model access.
Mistral showcases Mistral Large 4 engineering-drawing inspection demo
Mistral showcases Mistral Large 4 engineering-drawing inspection demo
A post on X showcases Mistral Large 4 solving an out-of-distribution malware reverse-engineering task: given an unknown binary, the model analyzes it end-to-end, identifies it as Cobalt Strike, and extracts indicators of compromise and the malware configuration. The demo highlights frontier-model applicability to security investigation work beyond training-distribution tasks.
A post on X showcases Mistral Large 4 solving an out-of-distribution malware reverse-engineering task: given an unknown binary, the model analyzes it end-to-end, identifies it as Cobalt Strike, and extracts indicators of compromise and the malware configuration. The demo highlights frontier-model applicability to security investigation work beyond training-distribution tasks.
On the AutomationBench benchmark, Mistral Large 4 completed 657 business workflows spanning simulated workplace apps including Gmail, Google Sheets, Slack, and Salesforce. It finished ahead of Kimi K3, MiMo-V2.6-Pro, and DeepSeek V4 Pro.
On the AutomationBench benchmark, Mistral Large 4 completed 657 business workflows spanning simulated workplace apps including Gmail, Google Sheets, Slack, and Salesforce. It finished ahead of Kimi K3, MiMo-V2.6-Pro, and DeepSeek V4 Pro.