Mistral-Large-3-675B-Instruct-2512: benchmark results
Mistral's open Large 3 sparse MoE (675B total, 41B active), December 2025. Provider: Mistral. Released 2025-12-02. Access: Open.
Aggregate Rating 1673 ± 18, rank #262 of 1656 rated models, from 32 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PM-Bench (Auto Heartbeat 30m) | 74.7 | Set-F1 (%; F1 of the prospective actions the agent selects a | 100 |
| PM-Bench (Hierarchical Multi-Agent) | 58.1 | Set-F1 (%; F1 of the prospective actions the agent selects a | 100 |
| UGI Leaderboard | 58.55 | UGI Score | 98.2 |
| PM-Bench | 75.3 | Set-F1 (%; F1 of the prospective actions the agent selects a | 85.7 |
| PM-Bench (Auto Heartbeat 60m) | 69.3 | Set-F1 (%; F1 of the prospective actions the agent selects a | 85.7 |
| RewardBench 2 Safety | 90.33 | Score (%) | 85.3 |
| RewardBench 2 Focus | 90.51 | Score (%) | 84.9 |
| UGI - Natural Intelligence | 38.79 | NatInt Score | 84.7 |
| Twinkle Eval - TMMLU+ | 73.31 | Accuracy (%) | 82.1 |
| RewardBench 2 Math | 72.27 | Score (%) | 80.6 |
| MEDIC Benchmark | 74.25 | MEDIC Public Table Average | 80 |
| Twinkle Eval - TW-Legal Benchmark v1 | 48.87 | Accuracy (%) | 78.6 |
Interactive version: theaggregate.ai/model?slug=mistral-large-3-675b-instruct-2512 · How It Works · Data refreshed daily, snapshot 2026-10-20.