Mistral open-weights Mistral Large 4 (Le Chonk), its 1T flagship

Mistral has released the weights for Mistral Large 4, codenamed Le Chonk, and says it rivals leading open models such as GLM-5.3 on the Artificial Analysis Coding Agent Index. According to ZDNET, the company also claims substantial gains in reasoning efficiency that lower inference costs.

TestingCatalog reports on X that Mistral Large 4, codenamed Le Chonk, is now available on Hugging Face. The post cites claims of substantial gains in reasoning efficiency and lower inference costs, and highlights the model's ranking among open-weight models.

Mistral has published Mistral Large 4 on Hugging Face, according to a post on the LocalLLaMA subreddit that links the official blog. The poster says the model is a 1T-parameter release and points to llama.cpp support.

According to Mistral's description relayed on Reddit's LocalLLaMA, Mistral Large 4 (codename Le Chonk) is its largest and most capable model yet: a 1-trillion-parameter natively multimodal MoE with 49B active parameters per token (52B including embeddings and output layers), trained from scratch on 3,800 NVIDIA Grace chips. Community GGUF and NVFP4 quantizations are already up on Hugging Face.

A post on Reddit's LocalLLaMA says Mistral's flagship Mistral Large 4, codenamed Le Chonk, is now open weight, following its release as a 1-trillion-parameter natively multimodal MoE. The post does not specify the license or exactly what weights were released.

ZDNet reports that Mistral has released the weights for Large 4, with its coding model Le Chonk ranking among top open models alongside Codex GPT-6.1 Terra on Artificial Analysis. The release gives developers a new open-weight option for coding work.

Mistral has released a new open-weight model called Le Chonk, claiming it matches or outperforms the best models from DeepSeek and Z.ai on coding and cybersecurity tasks, Wired reports. The comparison comes from Mistral itself, and the report cites no independent verification.

French startup Mistral has released Large 4, a trillion-parameter model it says reaches frontier performance among open models for agentic coding and cyber defense. Mistral calls it the best open model built in the West and is pitching it to businesses worried about losing access to closed AI.

Mistral has released the weights of Mistral Large 4, its 1-trillion-parameter model, on Hugging Face about three weeks after first previewing it, The Next Web reports. The Paris-based company says three extra weeks of reinforcement learning guided by customer feedback make it the strongest open model for agentic coding and cyber defence.

Mistral AI has released the weights of Mistral Large 4, codenamed Le Chonk, a 1T-parameter multimodal model it calls the best open-weight model for coding. The company says the model was trained in Europe.