Mistral Large 4: Open Weights, Price & Access

Mistral Large 4 is Mistral’s 1T-parameter open-weight model. The weights are free to download, and the API is $0.70 input and $2.10 output per 1M tokens.

Quick answer: Mistral AI released the open weights of Mistral Large 4, nicknamed Le Chonk, on October 29, 2026, three weeks after a public preview. The released checkpoint had extra reinforcement learning for coding and cyber defense, and the API price fell to $0.70 per million input tokens, $0.07 cached and $2.10 output. You can use it in Mistral Studio, the Mistral API, OpenRouter or Vibe Code, or download it from Hugging Face. We checked the figures against the Mistral Large 4 open-weight announcement and the Mistral docs on November 2, 2026.

What Mistral Large 4 is

• Size: a Mixture-of-Experts model with 1.05T total parameters, 52B active per token and a 1.6B vision encoder.
• Context: 1M tokens.
• Inputs: natively multimodal, so it reads images as well as text.
• Languages: Mistral says it’s fluent in 160+ languages.
• Reasoning modes: low for simple tasks, medium for everyday work and high (the default) for the hardest jobs.
• Training: trained from scratch on European infrastructure, which Mistral pitches for sovereign AI use.

Mistral Large 4 Pricing (2026)

Access	Price	Notes
API input	$0.70 per 1M tokens	Was $1.36 during the preview
API cached input	$0.07 per 1M tokens	For repeated prompt prefixes
API output	$2.10 per 1M tokens	Was $4.18 during the preview
Open weights	Free download	Mistral Ecosystem Open-Weight License

The license is a modified MIT license. You can use, modify, fine-tune and sell products built on the model. The one condition: if you or an affiliate run a model-as-a-service business, you need a separate commercial license from Mistral first. If token prices are new to you, our guide to tokens in AI explains what a million tokens buys.

Mistral Large 4 benchmarks

Mistral reports these coding results for the released checkpoint:

• FrontierCode 2.0 (Cognition): 19.7%, the top score in Mistral’s chart.
• Artificial Analysis Coding Agent Index: 48%, effectively tied with GLM-5.3 while using 47% fewer tokens per task.
• Terminal-Bench 4.0: 41%, up 13 points from the preview, with 30% fewer output tokens.

In blind pairwise tests by Surge, coders preferred it over GLM-5.3, Kimi K3 and DeepSeek V4.1 Flash, ranking it behind only Claude Opus 5.5. For how it stacks up against closed models, see our best AI model comparison.

Can you run Mistral Large 4 locally?

Technically yes, but not on a laptop. The weights come in FP8 and an NVFP4 build for NVIDIA Blackwell GPUs, and Mistral recommends serving with vLLM or SGLang on your own infrastructure. It’s also available as an NVIDIA NIM container. For models that do fit on a desktop, check our best local AI models list.

Limitations

• Huge hardware needs: a 1T-parameter model needs a multi-GPU server to self-host.
• License condition: model-as-a-service companies need a commercial license.
• Mostly self-reported results: the technical report is still “coming soon”.
• Cheaper rivals exist: some open models cost less per task on independent tests.

Mistral Large 4 is now one of the strongest open-weight coding models, and the API price cut makes it easy to test.

Use the API for most projects. Self-hosting only makes sense if you have the GPUs and a sovereignty reason.