MiniMax-M2.1: benchmark results
MiniMax's open M2.1, a 230B sparse text MoE with 10B active parameters (December 2025). Provider: MiniMax. Released 2025-12-22. Access: Open.
Aggregate Rating 1661 ± 6, rank #289 of 1656 rated models, from 483 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenEvals - MMLU-Pro | 88 | Accuracy (%) | 100 |
| SuperCLUE-WebCoding - Overall | 75.66 | Score | 100 |
| AA MMLU-Pro | 87.49 | Accuracy (%) | 98 |
| Medmarks - MMLU-Pro Health | 82.97 | Score (%) | 97.7 |
| SAGE (Service Agent) - Conversational Quality | 75.29 | Conversational Quality Score (0-100): weighted judge ratings | 96.2 |
| AA LiveCodeBench | 80.95 | Pass@1 (%) | 93 |
| SAGE (Service Agent) - Airline Refund | 66.37 | Overall Assessment Score (0-100), 0.8 x logical compliance + | 92.3 |
| SAGE (Service Agent) - Logistics Delivery | 68.54 | Overall Assessment Score (0-100), 0.8 x logical compliance + | 92.3 |
| WebTestBench - Content | 7.7 | Defect-detection F1 (%) on content test items, on 100 genera | 88.9 |
| Medmarks - MedConceptsQA Hard | 76.72 | Score (%) | 86.4 |
| MedConclusion - Non-Contradiction | 81.89 | Non-contradiction rate (judge score 0-100): the model writes | 85.7 |
| OctoBench - Claude Code - Checkitem Success | 82.75 | Checkitem Success Rate (%) | 85.7 |
Interactive version: theaggregate.ai/model?slug=minimax-m2-1 · How It Works · Data refreshed daily, snapshot 2026-10-20.