Learn about this first
Mistral AI has released the weights for its flagship model, Mistral Large 4 (nicknamed Le Chonk), which was previously in public preview. The model is a natively multimodal MoE with about 1T total parameters, 49B–52B active per token, and a 1M-token context window. It is available in FP8 and NVFP4 versions. The weights use the MIT-derived Mistral Ecosystem Open-Weight License: use, modification, fine-tuning, and commercial deployment are free, but enterprises offering the model as a service must obtain a separate commercial license from Mistral. API pricing has also been adjusted to $0.70 per million input tokens, $0.07 per million cached input tokens, and $2.10 per million output tokens.
Mistral says the model focuses on agentic coding and cybersecurity. It ranks first among open models on Cognition’s FrontierCode, and its cyber defense capabilities have been tested by vendors including CrowdStrike, Cloudflare, Rubrik, and Zscaler. Artificial Analysis reports that, with 1.05T total parameters, it is the strongest open-weight model developed outside China; The Decoder notes that it leads open-source models in coding but still trails Chinese models overall. vLLM announced Day 0 support the same day.
Unsloth released a Dynamic 1-bit GGUF quantized version of the model, saying it can run locally on Mac Studio while retaining 91% top-1 accuracy. The smallest version uses about 233GB of disk space and requires 256–270GB of memory, making it about 80% smaller than the 1.1TB FP8 version.
AI-generated from reports · updated 12 days agoLatest developments
Hands-On Testing of Mistral Large 4 for Django CVE Vulnerability Remediation
Event progress6
Vulnerability Discovery and Remediation Testing of Mistral Large 4 on a Django CVE
Mistral AI: Hands-On Testing of Mistral Large 4 for Django CVE Vulnerability RemediationMistral AI releases an update to its Large 4 model
Mistral AI: Three-week training camp for Mistral’s Le Chonk modelArtificial Analysis publishes evaluation results for Mistral Large 4
Artificial Analysis: Mistral Large 4 open-weight release, with 1.05T total parameters, becomes the strongest open-weight model developed outside ChinaUnsloth releases GGUF for running Mistral Le Chonk locally
Unsloth: Unsloth releases a 1-bit GGUF quantized version of Mistral Le Chonk, enabling local use of a 1T-parameter modelMistral AI releases Le Chonk weights
Hugging Face Official Team Blog: Mistral Large 4 (Le Chonk) comes to Hugging Face: an open-weight, multimodal MoE model with 1T parametersvLLM announces support for Mistral Large 4
vLLM Official Blog: vLLM releases Day 0 support for Mistral Large 4