SPACEXAI
4/5 workers samplingPOSSIBLE_MODEL_CHANGE · gpt-5
public · multimodal

MMMU

College-level multimodal understanding across disciplines.

Items
1730
Revision
2025.09
Contamination
watched
Scorer
scorer.mmmu.v3
Dataset revision
[email protected]
Scorer
choice+open · scorer.mmmu.v3
License
CC BY-NC 4.0
Kind
Public reproduction suite
ModelScore$/taskMedian sToolsEvidence
Gemini 2.5 Pro88.4$0.0258.80.2LIVE_MEASURED
GPT-584.8$0.06313.00.2LIVE_MEASURED
Gemini 2.5 Flash82.6$0.00113.40.2LIVE_MEASURED
Claude 4.1 Opus79.4$0.26716.70.2LIVE_MEASURED
GPT-5 Mini78.2$0.00315.20.2LIVE_MEASURED
Grok 4 Heavy77.0$0.13222.00.2LIVE_MEASURED
Grok 476.2$0.03810.20.2LIVE_MEASURED
Claude 4.1 Sonnet76.1$0.04711.40.2LIVE_MEASURED
Qwen3-Max75.2$0.00569.60.2LIVE_MEASURED
o4-pro73.6$0.63836.40.2LIVE_MEASURED
Mistral Large 373.4$0.0159.90.2LIVE_MEASURED
Llama 4 405B71.6$0.00439.10.2REPRODUCED
Llama 4 70B66.8$0.000941.90.2REPRODUCED
Grok 3 Mini66.4$0.00164.00.2LIVE_MEASURED
DeepSeek V3.140.2$0.00427.30.2REPRODUCED