Day 7 receipt
Mistral Large 4
Day 2 of 7Mistral's 1-trillion-parameter multimodal model (52B active), previewed on 6 Oct, with weights released on 29 Oct.
Launch dayWhat they claimed, quoted word for word · 3
“ML4 is one of the world's strongest AI models for cybersecurity.”
The page ties this to a top-five rank on the Artificial Analysis Cyber Index. No user has yet reported running security work on it.
“We will release the weights by the end of the month.”
Weights were published on Hugging Face on 29 October, under a modified MIT licence that requires a commercial licence for model-as-a-service businesses.
“significantly outperforming any open-weight model developed in the US or Europe”
Mistral's own comparison. One user with a personal benchmark says it is fine but not up to the best Chinese or US open-weight models; no controlled test yet.
Day 7What real users report · 3
Serving the model through vLLM behind a gateway, the reporter says Claude Code fails on every skill load because the chat template rejects a tool message that follows a user message. They say two other open models returned no such errors. A Mistral reply calls this expected given different training and API.
Open as of 31 Oct · 1 reply · Open the source
A user who tried it on a code documentation task says it ignored their skill files and made some hallucinations, and wonders whether the cause was their harness or the model.
Single comment in the weights thread · Open the source
A user who added it to a personal benchmark says it is fine, but certainly not up to the best Chinese or US open-weight models. They tested the model before the weights release.
Single comment in the weights thread · Open the source