Mistral Small 4
Mistral·
LLMssmallcloud + localWorkstation
Overview
119B MoE model (6B active) unifying reasoning, vision, and coding with configurable reasoning effort.
Capabilities and innovations
119B MoE (6B active)Configurable reasoning effortMultimodal (vision)Agentic coding256k contextApache 2.0Unified Magistral + Pixtral + Devstral128-Expert MoE Architecture40% Latency Reduction vs Small 33x Throughput vs Small 3
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| GPQA Diamond | 71.2% | unsourced |
| Humanity’s Last Exam (no tools) | 9.5% | unsourced |
| MMLU-Pro | 78% | unsourced |
| Terminal-Bench 2.x | 17.4% | unsourced |
- SWE-bench Verified: benchmark did not exist at release
- SWE-bench Pro: not reported by the vendor
- MMLU: not reported by the vendor
- MMMU-Pro: not reported by the vendor
Architecture and hardware
- Parameters
- 119B total, 6B active per token (MoE)
- Estimated VRAM at Q4
- ~69 GB, Workstation class
Links
More from Mistral
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.