MiMo-V2.5-Pro: benchmark results
Xiaomi MiMo-V2.5 Pro open MoE model (42B active). Provider: Xiaomi. Released 2026-04-23. Access: Open.
Aggregate Rating 1722 ± 5, rank #158 of 1656 rated models, from 624 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| FinCUABuildBench (Direct Prompting) | 12.3 | Qualified task rate (%; share of the 576 preregistered const | 100 |
| LLM Stats (CMMLU) | 90.2 | Score (%) | 100 |
| LLM Stats (TAU3-Bench) | 72.9 | Score (%) | 100 |
| RealClawBench - Code Fixing | 61.4 | Verifier pass rate (%) on the 63 code-fixing tasks of 281 re | 100 |
| RideWay - Success Rate | 87.4 | Success rate (%; share of trials passing the three-judge LLM | 100 |
| AA IFBench | 79.86 | Accuracy (%) | 98 |
| AA-Omniscience Hallucination Rate | 24.7 | Hallucination Rate (%) | 96.2 |
| AA Omniscience - Software Engineering (SWE) - Kotlin | 46 | Accuracy (%) | 96.1 |
| CLAW-Eval | 80.6 | Score (%) | 96 |
| TRACE Bench - Character Consistency | 94.95 | Checklist completion rate (%) | 96 |
| TRACE Bench - Language Quality | 98.73 | Per-turn LLM-judge language quality (%) | 96 |
| AA Omniscience - Software Engineering (SWE) - R | 40 | Accuracy (%) | 95.6 |
Interactive version: theaggregate.ai/model?slug=mimo-v2-5-pro · How It Works · Data refreshed daily, snapshot 2026-10-20.