TokenPost·

Mistral Releases Large 4 Weights After Three More Weeks of Reinforcement Learning

Your call?

The model scored 48% on a coding-agent index and 51% on a separate cyber index. Its license restricts some model-as-a-service offerings.

Share

More news