MiniMax-M1: benchmark results
MiniMax's open-weight 456B MoE reasoning model (45.9B active) using hybrid lightning attention to support a 1M-token context (June 2025). Provider: MiniMax. Released 2025-06-16. Access: Open.
Aggregate Rating 1579 ± 17, rank #514 of 1656 rated models, from 80 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| BenchTable - Tech | 80.1 | Weighted Score (%) | 89.1 |
| Context Arena MRCR (2-needle) | 30.4 | AUC@1M (%) | 77.5 |
| BenchTable - STEM | 61.5 | Weighted Score (%) | 65.7 |
| SuperCLUE General (July 2025) - Code Generation | 75.84 | Score | 65.2 |
| SnakeBench | 24.1 | TrueSkill Rating | 60.1 |
| BenchTable | 52.3 | Total Score (%) | 56.7 |
| ReLE - Law - Lawyer Qualification Exam | 71 | Accuracy (%) | 55.3 |
| Chatbot Arena (Text - English) | 1385 | Arena Score | 53.3 |
| LLM2014 Logic 2025-06 | 43.83 | Median Score | 52.8 |
| Chatbot Arena (Text - Mathematical) | 1380 | Arena Score | 52.3 |
| Chatbot Arena (Text - Coding) | 1416 | Arena Score | 52.2 |
| SuperCLUE General (July 2025) - Science Reasoning | 34.96 | Score | 52.2 |
Interactive version: theaggregate.ai/model?slug=minimax-m1 · How It Works · Data refreshed daily, snapshot 2026-10-20.