MiniMax-M2.7: benchmark results
MiniMax's open self-evolving M2.7 agentic MoE (230B total, 10B active). Provider: MiniMax. Released 2026-03-18. Access: Open.
Aggregate Rating 1675 ± 4, rank #258 of 1656 rated models, from 1050 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ClawSentry - Reviewer Replaceability | 27.78 | Task Success Rate (%) | 100 |
| CodeRouterBench - LLM Router (Online Agent) | 45.27 | Average task score (%) of the backend the model routes each | 100 |
| EnvSimBench - Simple | 50 | Config Match (%): share of the 50 Simple state-changing samp | 100 |
| FinReportBench | 22.6 | Provisional hierarchical score (0-100): mean over the 244 bi | 100 |
| IndicDB (DIN-SQL with Evidence) - Marathi | 63.67 | Execution accuracy (%) of generated SQL on the Marathi varia | 100 |
| IndicDB (DIN-SQL with Evidence) - Tamil | 68.07 | Execution accuracy (%) of generated SQL on the Tamil variant | 100 |
| IndicDB (Zero-Shot with Evidence) - Bengali | 67.11 | Execution accuracy (%) of generated SQL on the Bengali varia | 100 |
| IndicDB (Zero-Shot with Evidence) - English | 76.91 | Execution accuracy (%) of generated SQL on the English varia | 100 |
| LLM Stats (Multi-SWE-Bench) | 52.7 | Score (%) | 100 |
| OverEager-Bench (Claude Code) | 11.8 | Overeager rate (%; share of 500 benign coding scenarios in w | 100 |
| OverEager-Bench (Gemini CLI) | 10 | Overeager rate (%; share of 500 benign coding scenarios in w | 100 |
| OverEager-Bench (OpenHands) | 0.2 | Overeager rate (%; share of 500 benign coding scenarios in w | 100 |
Interactive version: theaggregate.ai/model?slug=minimax-m2-7 · How It Works · Data refreshed daily, snapshot 2026-10-20.