MiMo-V2.5: benchmark results
Xiaomi's open multimodal MiMo-V2.5 MoE (310B total, 15B active) with a 1M-token context. Provider: Xiaomi. Released 2026-04-23. Access: Open.
Aggregate Rating 1700 ± 7, rank #196 of 1656 rated models, from 464 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| PinchBench - 147 Tasks - OpenClaw 2026.5.20 (e510042) | 91.87 | Best Official Run Score (%) | 100 |
| TRACE Bench - Language Quality | 99.03 | Per-turn LLM-judge language quality (%) | 100 |
| VideoGAIA - Culture | 53.45 | Pass@1 accuracy (%; the 58 culture tasks; unified ReAct agen | 94.7 |
| UAV-DualCog - Landmark Visibility Counting | 58.5 | Visibility-count accuracy (%; 1,024 flight videos in which t | 92.9 |
| TRACE Bench - Short-Term Memory | 91 | Memory-probe completion rate (%) | 92 |
| AA-Omniscience Hallucination Rate | 31.93 | Hallucination Rate (%) | 91.2 |
| PanoSpace-Bench - Camera Rotation | 42.8 | Accuracy (%; 250 five-option multiple-choice questions on on | 90.9 |
| PanoSpace-Bench - Object Reorientation | 42 | Accuracy (%; 250 five-option multiple-choice questions on on | 90.9 |
| PanoSpace-Bench - Relative Direction | 42 | Accuracy (%; 250 five-option multiple-choice questions on on | 90.9 |
| UAV-DualCog - Flight Behavior Recognition (Composite) | 35.9 | Composite-level behavior accuracy (%; 1,024 first-person UAV | 90.4 |
| UGI - Writing | 56.43 | Writing Score | 89.8 |
| PinchBench - 147 Tasks - OpenClaw 2026.5.4 (325df3e) | 91.42 | Best Official Run Score (%) | 89.4 |
Interactive version: theaggregate.ai/model?slug=mimo-v2-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.