MiniMax-M2: benchmark results
MiniMax's open M2 MoE (230B total, 10B active) for coding and agentic work. Provider: MiniMax. Released 2025-10-23. Access: Open.
Aggregate Rating 1647 ± 7, rank #317 of 1656 rated models, from 426 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EvolMem - Learning | 51.35 | Learning (%) | 100 |
| EvolMem - Summarization | 71.74 | Summarization (%) | 100 |
| HeaRTS - Physiological Classification | 48 | Mean task score (x100) over HeaRTS's physiological classific | 100 |
| TurkBench - Bias | 91.4 | Accuracy (%) | 100 |
| TurkBench - Summarization | 83.16 | LLM-as-a-judge score (0-100) | 97.3 |
| HeaRTS - Trajectory Analysis | 55 | Mean task score (x100) over HeaRTS's trajectory analysis tas | 96.7 |
| OpenCompass Math - College - English (CompassBench 2510) | 60 | Score (%) | 96.2 |
| AA LiveCodeBench | 82.65 | Pass@1 (%) | 94.2 |
| Enkrypt AI - Bias Risk | 48.06 | Risk Score | 93.1 |
| Enkrypt AI - Risk Score | 17.64 | Risk Score | 92.6 |
| OpenCompass LLM - Math - English (CompassBench 2510) | 70.6 | Score (%) | 92.3 |
| WebCoderBench - Functional Alignment | 93.62 | Score (%) | 92.3 |
Interactive version: theaggregate.ai/model?slug=minimax-m2 · How It Works · Data refreshed daily, snapshot 2026-10-20.