Mistral Saba: benchmark results
Provider: Mistral. Released 2025-02-17. Access: API.
Aggregate Rating 1543 ± 22, rank #1803 of 10916 rated models, from 189 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Persian Fluency Bench - Instruction Following | 67.7 | Judge Score (%, judge google/gemini-2.5-flash) | 100 |
| Persian Fluency Bench - Conciseness | 36.5 | Judge Score (%, judge google/gemini-2.5-flash) | 93.3 |
| Persian Fluency Bench - Context Retention | 84.9 | Judge Score (%, judge google/gemini-2.5-flash) | 93.3 |
| Persian Fluency Bench - Safety | 89.1 | Judge Score (%, judge google/gemini-2.5-flash) | 93.3 |
| Persian Fluency Bench | 39.1 | Judge Score (%, judge google/gemini-2.5-flash) | 90 |
| Arabic Broad Leaderboard - General Knowledge | 8.97 | Average Score (0-10) | 77.1 |
| TextClass Benchmark | 1620.99 | Meta-Elo (imported) | 71 |
| Evals for Every Language - Galician (gl) | 71.83 | Average Score (%) | 68.2 |
| Evals for Every Language - Portuguese (pt) | 75.41 | Average Score (%) | 67.7 |
| Evals for Every Language - Armenian (hy) | 70.53 | Average Score (%) | 67.2 |
| Evals for Every Language - Norwegian Bokmål (nb) | 76.92 | Average Score (%) | 63.1 |
| Evals for Every Language - Japanese (ja) | 71.98 | Average Score (%) | 62.7 |
Interactive version: theaggregate.ai/model?slug=mistral-saba · How It Works · Data refreshed daily, snapshot 2026-11-08.