Mistral released the weights of Mistral Large 4. It is a 1T-parameter multimodal model. It leads open-weight benchmarks for coding. It also leads in defensive cybersecurity tasks. The model was trained in Europe. It offers three reasoning modes for cost-performance balance. It refuses 96.6% of malicious cyber requests. It scores 19.7% on FrontierCode 2.0. It scores 48% on the Artificial Analysis Coding Agent Index. It scores 41% on Terminal-Bench 4.0. It scores 51% on the Artificial Analysis Cyber Index. It scores 73.3% on Finch for finance workflows. It scores 84.5% on BrowseComp for web browsing.
What changed
Mistral Large 4 was in public preview with closed weights.
What it unlocks
Self-hosting and customizing a 1T-parameter open-weight model for coding and cybersecurity tasks.
- 1T parameters
- 19.7% on FrontierCode 2.0
- 48% on Artificial Analysis Coding Agent Index
- 41% on Terminal-Bench 4.0
What you need to act on it
- access to European infrastructure for sovereign deployment
Sources