Multimodal
MMMU
College-level questions that require reading charts, diagrams, tables and photographs alongside the text.
MMMU is scored as a percentage of questions answered correctly. 39 of the 159 models in this index report a score for it. The highest published figure here is 84.2%, from GPT-5.
Reported scores
Top 20 on MMMU
Ordered by the figure each maker published. Models that have not reported this benchmark are not listed — an absent score is not a low score.
| # | Model | Published score |
|---|---|---|
| 1 | GPT-5OpenAI | |
| 2 | o3OpenAI | |
| 3 | Claude Opus 4.5Anthropic | |
| 4 | Gemini 2.5 ProGoogle DeepMind | |
| 5 | o4-miniOpenAI | |
| 6 | Gemini 3 ProGoogle DeepMind | |
| 7 | Gemini 2.5 FlashGoogle DeepMind | |
| 8 | Claude Sonnet 4.5Anthropic | |
| 9 | o1OpenAI | |
| 10 | GPT-4.1OpenAI | |
| 11 | Claude Sonnet 4Anthropic | |
| 12 | Step-3StepFun | |
| 13 | Llama 4 MaverickMeta AI | |
| 14 | GPT-4.1 miniOpenAI | |
| 15 | Claude 3.7 SonnetAnthropic | |
| 16 | Gemini 2.0 FlashGoogle DeepMind | |
| 17 | Claude 3.5 SonnetAnthropic | |
| 18 | Qwen2.5-VL-72BAlibaba Qwen | |
| 19 | Amazon Nova PremierAmazon | |
| 20 | Llama 4 ScoutMeta AI |
Where these numbers come from
Every score on this page is a published figure, taken from the model's own card, system card, technical report or release post, or from a public leaderboard. CorX Labs did not run these evaluations. Most are self-reported by the lab that built the model, which means they were produced under that lab's own choice of prompt, scaffold and number of attempts — so treat them as a starting point for a shortlist, not as a settled ranking.
A score someone other than the model's maker measured is marked Independent and names its measurer. Those are the stronger numbers on this page — an outside harness has no reason to flatter anyone — and there are not many of them.
Where a figure has not been published, the cell reads Not reported rather than an estimate. Nothing here is inferred, interpolated or guessed. Each model records the month its row was last checked. Full method and caveats.
How to read this score
MMMU is the standard test of whether a model can genuinely read an image rather than caption it — the questions require pulling values off a chart, reading a circuit diagram, or interpreting a medical scan alongside the text.
Where it is weak
Human experts score around 88%, so there is real headroom, but many questions can be narrowed down from the text alone. Reported scores also vary with how the image was encoded and at what resolution.