Mistral Small 4: benchmark results
Mistral AI's Small 4, a 119B sparse MoE (6.5B active) model (March 2026). Provider: Mistral. Released 2026-03-16. Access: Open.
Aggregate Rating 1579 ± 8, rank #517 of 1656 rated models, from 555 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SEA-HELM (Indonesian) - Sentiment Analysis | 81.69 | Normalized Score | 98.2 |
| SEA-HELM (Vietnamese) - Sentiment Analysis | 64.67 | Normalized Score | 98.2 |
| Nejumi 4 - jaster (2-shot) - MAWPS | 99 | Exact match (%) | 96.9 |
| Nejumi 4 - jaster (0-shot) - JSICK | 85 | Exact match (%) | 96.2 |
| YapBench | 46.2 | YapIndex (lower is better) | 95.1 |
| SpeechMap Compliance | 90.9 | % Requests Completed | 92.8 |
| SEA-HELM (Vietnamese) - NLU | 70.87 | Normalized Score | 91.2 |
| StereoTales | 117 | Emissions (self-reported) | 90.9 |
| Horangi 4 - Korean Hate Speech | 67 | Accuracy (%) | 88 |
| AA Omniscience - Software Engineering (SWE) - Dart | 30 | Accuracy (%) | 85.2 |
| Polish Knowledge Benchmark - Industry | 83.71 | Accuracy (%) | 83.1 |
| Polish Knowledge Benchmark - Media & Journalism | 77.37 | Accuracy (%) | 80.9 |
Interactive version: theaggregate.ai/model?slug=mistral-small-4 · How It Works · Data refreshed daily, snapshot 2026-10-20.