Mistral releases Large 4 weights and cuts its API price

Mistral has published the weights of Mistral Large 4, its 1-trillion-parameter multimodal model, after a public preview earlier this month. The API price is now $0.70 per million input tokens, $0.07 cached and $2.10 output, and the weights are on Hugging Face for self-hosting with vLLM or SGLang. The licence lets you use, modify and sell products built on it, but companies that offer the model itself as a hosted service need a separate commercial licence. Mistral's own benchmark claims include 73.3% on a finance spreadsheet test (Finch) and lead scores among open-weight models for coding; these are vendor figures, not independent tests. On Hacker News (330 comments), the largest sub-thread, 161 replies under one comment, argues over whether that licence counts as truly open. Of 37 top-level comments, at least 9 are positive, mostly about keeping data in Europe or the price, and at least 6 doubt it beats the best Chinese or US models. No consensus. One commenter notes the smallest compressed version still needs about 256GB of GPU memory, so self-hosting is a server job. Sorted take: for most teams this is an API option to test, not a model to run in the office.