MiniMax M3 (High): benchmark results

Provider: MiniMax. Released 2026-06-01. Access: Open.

Unified ELO 1707 ± 24, rank #349 of 2055 rated models, from 15 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
SceneActBench - Articulated58.4Task score (0-100): 100 S2O articulated objects: rig a close60
SceneActBench - Reconstruction9.6Task score (0-100): 100 3D-FRONT rooms from about 11 calibra60
SceneActBench - Camera25.6Task score (0-100): 100 furnished 3D-FRONT scenes: recover t40
MazeBench0Gems collected (gems)37.5
GameLogicBench - Repo13.04Solve rate (%; the 23 Repo tasks (a mechanic inside a real o29.2
CWE-Bench22.7Pass@1 (%)21.4
GameLogicBench19.44Solve rate (%; all 72 tasks; one sample per model-scaffold c16.7
GameLogicBench - Combo17.86Solve rate (%; the 28 Combo tasks (several interacting mecha16.7
AI Coding Daily (OpenCode) - React-TS Code Quality15React-TS Code Quality (max 20) points, LLM-judged rubric sco15.2
SceneActBench - Layout58.1Task score (0-100): 100 3D-FRONT rooms from one view: place 10
AI Coding Daily (OpenCode) - Laravel Code Quality15.3Laravel Code Quality (max 20) points, LLM-judged rubric scor8.7
GameLogicBench - Atom28.57Solve rate (%; the 21 Atom tasks (one mechanic in a minimal 8.3

Interactive version: theaggregate.ai/model?slug=minimax-m3-high · How It Works · Data refreshed daily, snapshot 2026-09-29.