MiniMax-M2.5: benchmark results
MiniMax's open M2.5 MoE (230B total, 10B active) for cost-efficient coding and agentic work. Provider: MiniMax. Released 2026-02-12. Access: Open.
Aggregate Rating 1654 ± 4, rank #303 of 1656 rated models, from 813 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| TurkBench - Complex Reasoning | 97.12 | Accuracy (%) | 100 |
| TurkBench - Summarization | 84.09 | LLM-as-a-judge score (0-100) | 100 |
| TurkBench - Reading Comprehension | 95.41 | LLM-as-a-judge score (0-100) | 97.3 |
| MERA Code - CodeLinterEval | 97.18 | pass@1 (%) | 96.9 |
| SWE-bench Verified | 75.8 | Resolved (%) | 96.7 |
| CringeBench | 0.54 | Cringe Score (0-10, lower is better) | 95.5 |
| Tau2-Bench Telecom | 95.32 | Success Rate (%) | 94.6 |
| AA TAU-2 Bench | 95.32 | Accuracy (%) | 94.5 |
| Swiss-Bench SBP-002 - EFK | 10.4 | Correct rate (%) on the 48 Swiss Federal Audit Office (EFK) | 94.4 |
| TurkBench - Turkish Vocabulary | 96 | Accuracy (%) | 93.2 |
| RewardBench 2 Precise IF | 66.72 | Score (%) | 93.1 |
| MedRoundsQA - Multi-Turn Diagnosis | 36.4 | Accuracy (%) | 92.9 |
Interactive version: theaggregate.ai/model?slug=minimax-m2-5 · How It Works · Data refreshed daily, snapshot 2026-10-20.