multimodal benchmark
MMMU-Pro Leaderboard
A more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings. Achieves significantly lower model performance (16.8-26.9%) compared to original MMMU, providing more rigorous evaluation that closely mimics real-world scenarios.
Updated Aug 17, 2026
Models68
Model coverage68
MetricScore
EvidenceB
MMMU-Pro Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelGO | Score83.6% | Percentile100.0% | Participants68 | EvidenceC | Evaluated |
| Rank02 | ModelOP | Score83.2% | Percentile98.5% | Participants68 | EvidenceC | Evaluated |
| Rank03 | ModelOP | Score83.0% | Percentile97.0% | Participants68 | EvidenceC | Evaluated |
| Rank04 | ModelBY | Score82.7% | Percentile95.5% | Participants68 | EvidenceC | Evaluated |
| Rank05 | ModelAC | Score82.3% | Percentile94.0% | Participants68 | EvidenceC | Evaluated |
| Rank06 | ModelBY | Score82.2% | Percentile92.5% | Participants68 | EvidenceC | Evaluated |
| Rank07 | ModelMA | Score81.6% | Percentile91.0% | Participants68 | EvidenceC | Evaluated |
| Rank08 | ModelGO | Score81.2% | Percentile89.5% | Participants68 | EvidenceC | Evaluated |
| Rank09 | ModelOP | Score81.2% | Percentile88.1% | Participants68 | EvidenceC | Evaluated |
| Rank10 | ModelGO | Score81.0% | Percentile86.6% | Participants68 | EvidenceC | Evaluated |
| Rank11 | ModelOP | Score80.7% | Percentile85.1% | Participants68 | EvidenceC | Evaluated |
| Rank12 | ModelGO | Score80.5% | Percentile83.6% | Participants68 | EvidenceC | Evaluated |
| Rank13 | ModelME | Score80.4% | Percentile82.1% | Participants68 | EvidenceC | Evaluated |
| Rank14 | ModelMA | Score80.1% | Percentile80.6% | Participants68 | EvidenceC | Evaluated |
| Rank15 | ModelOP | Score79.5% | Percentile79.1% | Participants68 | EvidenceC | Evaluated |
| Rank16 | ModelAC | Score79.0% | Percentile77.6% | Participants68 | EvidenceC | Evaluated |
| Rank17 | ModelAC | Score78.8% | Percentile76.1% | Participants68 | EvidenceC | Evaluated |
| Rank18 | ModelMA | Score78.5% | Percentile74.6% | Participants68 | EvidenceC | Evaluated |
| Rank19 | ModelOP | Score78.4% | Percentile73.1% | Participants68 | EvidenceC | Evaluated |
| Rank20 | ModelOP | Score78.4% | Percentile71.6% | Participants68 | EvidenceC | Evaluated |
| Rank21 | ModelMI | Score78.1% | Percentile70.2% | Participants68 | EvidenceC | Evaluated |
| Rank22 | ModelXI | Score77.9% | Percentile68.7% | Participants68 | EvidenceC | Evaluated |
| Rank23 | ModelAN | Score77.3% | Percentile67.2% | Participants68 | EvidenceC | Evaluated |
| Rank24 | ModelGO | Score76.9% | Percentile65.7% | Participants68 | EvidenceC | Evaluated |
| Rank25 | ModelAC | Score76.9% | Percentile64.2% | Participants68 | EvidenceC | Evaluated |
| Rank26 | ModelGO | Score76.8% | Percentile62.7% | Participants68 | EvidenceC | Evaluated |
| Rank27 | ModelOP | Score76.6% | Percentile61.2% | Participants68 | EvidenceC | Evaluated |
| Rank28 | ModelOP | Score76.4% | Percentile59.7% | Participants68 | EvidenceC | Evaluated |
| Rank29 | ModelOP | Score76.0% | Percentile58.2% | Participants68 | EvidenceC | Evaluated |
| Rank30 | ModelAC | Score75.8% | Percentile56.7% | Participants68 | EvidenceC | Evaluated |
| Rank31 | ModelAN | Score75.6% | Percentile55.2% | Participants68 | EvidenceC | Evaluated |
| Rank32 | ModelAC | Score75.3% | Percentile53.7% | Participants68 | EvidenceC | Evaluated |
| Rank33 | ModelAC | Score75.1% | Percentile52.2% | Participants68 | EvidenceC | Evaluated |
| Rank34 | ModelAC | Score75.0% | Percentile50.8% | Participants68 | EvidenceC | Evaluated |
| Rank35 | ModelTM | Score74.0% | Percentile49.3% | Participants68 | EvidenceC | Evaluated |
| Rank36 | ModelME | Score74.0% | Percentile47.8% | Participants68 | EvidenceC | Evaluated |
| Rank37 | ModelGO | Score73.8% | Percentile46.3% | Participants68 | EvidenceC | Evaluated |
| Rank38 | ModelAC | Score69.3% | Percentile44.8% | Participants68 | EvidenceC | Evaluated |
| Rank39 | ModelGO | Score69.1% | Percentile43.3% | Participants68 | EvidenceC | Evaluated |
| Rank40 | ModelAC | Score68.1% | Percentile41.8% | Participants68 | EvidenceC | Evaluated |
| Rank41 | ModelAC | Score68.1% | Percentile40.3% | Participants68 | EvidenceC | Evaluated |
| Rank42 | ModelOP | Score66.1% | Percentile38.8% | Participants68 | EvidenceC | Evaluated |
| Rank43 | ModelAC | Score65.3% | Percentile37.3% | Participants68 | EvidenceC | Evaluated |
| Rank44 | ModelAM Nova 2 ProAmazon | Score63.5% | Percentile35.8% | Participants68 | EvidenceC | Evaluated |
| Rank45 | ModelCO | Score63.0% | Percentile34.3% | Participants68 | EvidenceC | Evaluated |
| Rank46 | ModelAC | Score63.0% | Percentile32.8% | Participants68 | EvidenceC | Evaluated |
| Rank47 | ModelAMNova 2 LiteAmazon | Score61.8% | Percentile31.3% | Participants68 | EvidenceC | Evaluated |
| Rank48 | ModelAMNova 2 OmniAmazon | Score61.4% | Percentile29.9% | Participants68 | EvidenceC | Evaluated |
| Rank49 | ModelAC | Score60.4% | Percentile28.4% | Participants68 | EvidenceC | Evaluated |
| Rank50 | ModelAC | Score60.4% | Percentile26.9% | Participants68 | EvidenceC | Evaluated |
| Rank51 | ModelMA | Score60.0% | Percentile25.4% | Participants68 | EvidenceC | Evaluated |
| Rank52 | ModelOP | Score59.9% | Percentile23.9% | Participants68 | EvidenceC | Evaluated |
| Rank53 | ModelME | Score59.6% | Percentile22.4% | Participants68 | EvidenceC | Evaluated |
| Rank54 | ModelAC | Score57.0% | Percentile20.9% | Participants68 | EvidenceC | Evaluated |
| Rank55 | ModelAC | Score55.9% | Percentile19.4% | Participants68 | EvidenceC | Evaluated |
| Rank56 | ModelGO | Score54.3% | Percentile17.9% | Participants68 | EvidenceC | Evaluated |
| Rank57 | ModelAC | Score53.2% | Percentile16.4% | Participants68 | EvidenceC | Evaluated |
| Rank58 | ModelGO | Score52.6% | Percentile14.9% | Participants68 | EvidenceC | Evaluated |
| Rank59 | ModelAC | Score51.1% | Percentile13.4% | Participants68 | EvidenceC | Evaluated |
| Rank60 | ModelAC | Score49.5% | Percentile11.9% | Participants68 | EvidenceC | Evaluated |
| Rank61 | ModelAC | Score46.2% | Percentile10.4% | Participants68 | EvidenceC | Evaluated |
| Rank62 | ModelME | Score45.2% | Percentile9.0% | Participants68 | EvidenceC | Evaluated |
| Rank63 | ModelGO | Score44.2% | Percentile7.5% | Participants68 | EvidenceC | Evaluated |
| Rank64 | ModelMI | Score38.5% | Percentile6.0% | Participants68 | EvidenceC | Evaluated |
| Rank65 | ModelAC | Score38.3% | Percentile4.5% | Participants68 | EvidenceC | Evaluated |
| Rank66 | ModelAC | Score36.6% | Percentile3.0% | Participants68 | EvidenceC | Evaluated |
| Rank67 | ModelME | Score33.0% | Percentile1.5% | Participants68 | EvidenceC | Evaluated |
| Rank68 | ModelLA | Score30.5% | Percentile0.0% | Participants68 | EvidenceC | Evaluated |
MMMU-Pro Highlights
The leading models and scores on this benchmark.
MMMU-Pro Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for MMMU-Pro
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
Ranking basisThis mmmu-pro AI model leaderboard uses descending score in the benchmark's original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
- 01GOGemini 3.5 FlashGoogle
- Score
- 83.6%
- Price
- $1.5 input / $9.0 output per 1M tokens
- Speed
- Up to 1.2 tok/s via Google
Strengths
- Ranks #1 of 68 compared models
- 100th percentile on this benchmark
- C evidence result
Considerations
- This result measures MMMU-Pro, not total model capability
- 02OPGPT-5.5OpenAI
- Score
- 83.2%
- Price
- $5.0 input / $30 output per 1M tokens
Strengths
- Ranks #2 of 68 compared models
- 99th percentile on this benchmark
- C evidence result
Considerations
- This result measures MMMU-Pro, not total model capability
- 03OPGPT-5.6 SolOpenAI
- Score
- 83.0%
- Price
- $5.0 input / $30 output per 1M tokens
- Speed
- Up to 27 tok/s via OpenAI
Strengths
- Ranks #3 of 68 compared models
- 97th percentile on this benchmark
- C evidence result
Considerations
- This result measures MMMU-Pro, not total model capability
- 04BYSeed 2.1 ProByteDance
- Score
- 82.7%
Strengths
- Ranks #4 of 68 compared models
- 96th percentile on this benchmark
- C evidence result
Considerations
- This result measures MMMU-Pro, not total model capability
- 05ACQwen3.8 MaxAlibaba Cloud / Qwen Team
- Score
- 82.3%
- Price
- $2.0 input / $6.0 output per 1M tokens
Strengths
- Ranks #5 of 68 compared models
- 94th percentile on this benchmark
- C evidence result
Considerations
- This result measures MMMU-Pro, not total model capability
Selection summary
Best AI Models for MMMU-Pro
Gemini 3.5 Flash currently leads MMMU-Pro with 83.6%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price and runtime.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.
What is MMMU-Pro?
What MMMU-Pro measures and how its scores work.
A more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings. Achieves significantly lower model performance (16.8-26.9%) compared to original MMMU, providing more rigorous evaluation that closely mimics real-world scenarios.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- MMMU-Pro
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- mmmu-pro|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about MMMU-Pro.
Which model scores highest on MMMU-Pro?
Gemini 3.5 Flash is currently ranked first with 83.6%.
What does MMMU-Pro measure?
A more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings. Achieves significantly lower model performance (16.8-26.9%) compared to original MMMU, providing more rigorous evaluation that closely mimics real-world scenarios.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
68 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.