MiMo-V2.5-Pro (Thinking): benchmark results
Xiaomi's open-weight MiMo-V2.5-Pro (April 2026), a 1.02-trillion-parameter, 42-billion-active MoE model with hybrid attention and multi-token prediction, supporting up to a 1-million-token context. Provider: Xiaomi. Released 2026-04-23. Access: Open.
Aggregate Rating 1691 ± 20, rank #384 of 2170 rated models, from 75 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - jaster (2-shot) - JCommonsenseQA | 100 | Exact match (%) | 94.8 |
| Nejumi 4 - Toxicity - Fairness | 99.44 | Criteria met (%) | 94.1 |
| Nejumi 4 - jaster (2-shot) - JSQuAD | 81.56 | Character F1 (x100) | 94.1 |
| Nejumi 4 - MT-Bench (Japanese) - Coding | 100 | Judge rating (1-10, x10) | 91 |
| Nejumi 4 - jaster (0-shot) - JSQuAD | 81.23 | Character F1 (x100) | 90.3 |
| Nejumi 4 - jaster (2-shot) - JaMP | 80 | Exact match (%) | 88.5 |
| Nejumi 4 - MT-Bench (Japanese) - Roleplay | 98.5 | Judge rating (1-10, x10) | 84.7 |
| Nejumi 4 - jaster (2-shot) - JNLI | 90 | Exact match (%) | 84.4 |
| Nejumi 4 - BFCL - Live AST | 73.15 | Accuracy (%) | 83.3 |
| Wolfram LLM Benchmarking Project | 60.1 | Correct Functionality (%) | 82.7 |
| Nejumi 4 - jaster (0-shot) - JaNLI | 100 | Exact match (%) | 82.6 |
| Nejumi 4 - Toxicity - Social Norms | 97.62 | Criteria met (%) | 78.8 |
Interactive version: theaggregate.ai/model?slug=mimo-v2-5-pro-thinking · How It Works · Data refreshed daily, snapshot 2026-11-05.