✳ the wire · analysis
Mistral Large 4's weights are out, and Mistral says they beat its own preview
An email when every new Mistral model answers on a public API, about 15 minutes after members, plus every frontier launch · no card
Mistral published the weights of Mistral Large 4 on Hugging Face on October 29, two days inside the end-of-October date it gave when the model went to public preview on October 6. The repository, mistralai/Mistral-Large-4-1T-A52B, holds the full checkpoint in 54 shards and is not gated. A smaller NVFP4 build for Nvidia Blackwell GPUs sits beside it.
THE LICENSE
The Mistral Ecosystem Open-Weight License is a modified MIT license. It allows commercial and non-commercial use with one condition: a business that sells the model as a hosted API service needs a separate commercial license from Mistral.
THE MODEL
Per Mistral's model card it is a natively multimodal mixture-of-experts with about 1 trillion total parameters and 49 billion active per token (52 billion counting embeddings and output layers), a 1M-token context, text and image input, and three reasoning efforts: low, medium and high, the default. The checkpoint carries a built-in draft head for speculative decoding.
WHAT MISTRAL SAYS CHANGED
Mistral says the published weights are an improved version of the preview, after three more weeks of reinforcement learning, and its API serves them under the same mistral-large-4 id. By Mistral's own figures Terminal-Bench 4.0 rose 13 points to 41%, and the model scores 48% on Artificial Analysis's Coding Agent Index, level with GLM-5.3 while using 47% fewer tokens per task. These are Mistral's figures.
THE PRICE
Mistral's docs list $0.70 per million input tokens, $0.07 cached and $2.10 output, which the blog says applies from today. OpenRouter still listed mistralai/mistral-large-4-0 at $1.36 / $4.18 when we read it around 15:00 UTC.
THE INDEPENDENT READ
Artificial Analysis's page for Mistral Large 4 (high) reads 26 on its Intelligence Index v5.0. Mistral calls that the highest among western open-weight models and says the leading Chinese open models score higher on the same index.
WHAT WE DO NOT KNOW
Whether boards beyond Artificial Analysis have scored the improved checkpoint yet, and when the preview label comes off Mistral's API.
Source: Mistral: 'Mistral Large 4 (Le Chonk) is now open weight' (mistral.ai/news/mistral-large-4-oss), the Hugging Face model card for mistralai/Mistral-Large-4-1T-A52B and docs.mistral.ai (all read directly) ↗ · Mistral Large 4 tracker · the bench index