Ministral 3 8B: benchmark results
Mistral's Apache-2.0 8B dense edge model from the Mistral 3 family, with image understanding (December 2025). Provider: Mistral. Released 2025-12-02. Access: Open.
Aggregate Rating 1529 ± 8, rank #687 of 1656 rated models, from 471 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| AGC-Bench - mops | 1.99 | Dataset z-score | 97.6 |
| AGC-Bench - schnovel | 1.63 | Dataset z-score | 97.6 |
| AGC-Bench - thenextchapter | 1.72 | Dataset z-score | 96.3 |
| AGC-Bench - tinystories | 1.56 | Dataset z-score | 95.1 |
| AGC-Bench - pun_eval | 1.22 | Dataset z-score | 93.9 |
| SpeechMap Compliance | 91.6 | % Requests Completed | 93.3 |
| AGC-Bench - unfun_corpus | 0.72 | Dataset z-score | 92.3 |
| MMOOC - Missing Knowledge and Background (Multiple Choice) | 44.5 | Out-of-context score (%; mean of the refusal rate, the share | 88.2 |
| AGC-Bench - crowd_vote | 1.17 | Dataset z-score | 87.8 |
| AgentFloor | 55.2 | TCR (self-reported) | 87.5 |
| MMOOC - Missing Knowledge and Background (Open-Ended VQA) | 58 | Out-of-context score (%; mean of the refusal rate, the share | 85.3 |
| Evals for Every Language - Nuer (nus) | 35.22 | Average Score (%) | 85.2 |
Interactive version: theaggregate.ai/model?slug=ministral-3-8b · How It Works · Data refreshed daily, snapshot 2026-10-20.