Arena Hardarena-hard · general
58.3%
ReportedSep 2026
VENICE ATLAS · MODEL DOSSIER · MISTRAL SMALL 4
Model overview
Mistral Small 4 unifies instruction following, reasoning, coding, and vision in a single 119B MoE model with 256K context and configurable reasoning effort.
See the tests behind each score. They use the benchmark’s original scoring scale.
Related models