MMMU (validation)
Official MMMU validation-set leaderboard.
As of 25 Oct 2026, GPT-5.1 leads MMMU (validation) on BenchLeader with 85.4%, ahead of Gemini 2.5 Deep Think at 84.0%, across 50 model configurations with a published result.
- Published by
- MMMU
- Category
- Multimodal
- Index weight
- 0.5
- Models
- 50
- Data as of
- 25 Oct 2026
- Newest result
- 24 Nov 2025
The publisher has posted no new results on this board since 24 Nov 2025. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.
MMMU (Yue et al.).
What the test looks like
MMMU is a college-level multimodal exam: about 900 validation questions with diagrams, charts, photographs and tables across six broad disciplines and thirty subjects.
How it is scored
Accuracy on the validation set, as reported on the official leaderboard.
What to keep in mind
Results are submitted by model developers and reflect their own runs and prompts. Some questions are answerable from text alone.
- 1GPT-5.185.4%
- 2Gemini 2.5 Deep Think (thinking)84.0%
- 3o382.9%
- 4o4-mini81.6%
- 5Claude Opus 4.580.7%
- 6Gemini 2.5 Flash79.7%
- 7Gemini 2.5 Pro79.6%
- 8Qwen3 VL 235B A22B Instruct78.7%
- 9o178.2%
- 10Grok 378.0%
- 11Claude Sonnet 4.577.8%
- 12Claude Opus 4.177.1%
- 13Claude Opus 476.5%
- 14Claude 3.7 Sonnet75.0%
- 15Seed 1.6 Thinking (thinking)74.8%
| # | ||||
|---|---|---|---|---|
| 1 |  GPT-5.1 | 85.4% | 56.8 | 2025-11-13 |
| 2 |  Gemini 2.5 Deep Thinkthinking | 84.0% | – | 2025-05-20 |
| 3 | o3 | 82.9% | 59.5 | 2025-04-16 |
| 4 | o4-mini | 81.6% | 56.9 | 2025-04-16 |
| 5 |  Claude Opus 4.5 | 80.7% | 57.9 | 2025-11-24 |
| 6 | Gemini 2.5 Flash | 79.7% | 52.7 | 2025-05-20 |
| 7 | Gemini 2.5 Pro | 79.6% | 53.9 | 2025-05-06 |
| 8 |  Qwen3 VL 235B A22B Instructopen ↗ | 78.7% | 48.7 | 2025-09-22 |
| 9 | o1 | 78.2% | 53.1 | 2024-09-12 |
| 10 |  Grok 3 | 78.0% | 51.5 | 2025-02-19 |
| 11 | Claude Sonnet 4.5 | 77.8% | 53.7 | 2025-09-29 |
| 12 | Claude Opus 4.1 | 77.1% | 51.5 | 2025-08-05 |
| 13 | Claude Opus 4 | 76.5% | 52.6 | 2025-05-23 |
| 14 | Claude 3.7 Sonnet | 75.0% | 50.0 | 2025-02-24 |
| 15 |  Seed 1.6 Thinkingthinking | 74.8% | – | 2025-06-25 |
| 16 | Claude Sonnet 4 | 74.4% | 50.5 | 2025-05-23 |
| 17 | GPT-4.5 | 74.4% | 49.7 | 2025-02-28 |
| 18 |  Llama 4 Maverickopen ↗ | 73.4% | 42.8 | 2025-04-05 |
| 19 | Gemini 2.0 Pro | 72.7% | 47.0 | 2025-02-05 |
| 20 | Gemini 2.0 Flash | 71.7% | 43.2 | 2025-02-05 |
| 21 | GPT-4o | 70.7% | 43.9 | 2024-11-20 |
| 22 |  InternVL2_5-78B | 70.1% | – | 2024-12-05 |
| 23 | Llama 4 Scoutopen ↗ | 69.4% | 39.7 | 2025-04-05 |
| 24 | Claude 3.5 Sonnet | 68.3% | 46.3 | 2024-06-20 |
| 25 | Gemini 2.0 Flash-Lite | 68.0% | 47.2 | 2025-02-05 |
| 26 |  Nova Premier | 68.0% | 45.3 | 2025-04-30 |
| 27 | Gemini 1.5 Pro | 65.8% | 46.1 | 2024-08-01 |
| 28 | Gemma 3 27Bopen ↗ | 64.9% | 38.9 | 2025-03-12 |
| 29 | Qwen2-VL-72B | 64.5% | – | 2024-08-29 |
| 30 | Qwen2.5 VL 72B Instructopen ↗ | 64.5% | – | 2025-02-19 |
| 31 | Nova Pro | 62.0% | 39.8 | 2025-04-30 |
| 32 | Llama 3.2 90B | 60.3% | – | 2024-09-25 |
| 33 | Gemma 3 12Bopen ↗ | 59.6% | 36.9 | 2025-03-12 |
| 34 | Claude 3 Opus | 59.4% | 39.2 | 2024-03-05 |
| 35 | GPT-4o mini | 59.4% | 35.8 | 2024-07-18 |
| 36 | Qwen2.5-VL 7B Instructopen ↗ | 58.6% | – | 2025-02-19 |
| 37 |  LLaVA-OneVision 72B | 56.8% | – | 2024-08-06 |
| 38 |  Reka Core | 56.3% | 44.5 | 2024-04-23 |
| 39 | Gemini 1.5 Flash | 56.1% | 44.3 | 2024-05-23 |
| 40 | InternVL2-40B | 55.2% | – | 2024-07-04 |
| 41 | Reka Flash | 53.3% | 42.4 | 2024-04-23 |
| 42 | Claude 3 Sonnet | 53.1% | 38.0 | 2024-03-05 |
| 43 | Qwen-VLmax | 51.4% | – | 2024-01-27 |
| 44 | Claude 3 Haiku | 50.2% | 37.0 | 2024-03-05 |
| 45 | Minicpm V 2.6 | 49.8% | – | 2024-08-06 |
| 46 | Gemma 3 4Bopen ↗ | 48.8% | 34.5 | 2025-03-12 |
| 47 | Gemini 1.0 Pro | 47.9% | – | 2023-12-06 |
| 48 | Qwen-VL Plus | 45.2% | – | 2023-12-15 |
| 49 |  Aya Vision 32B | 45.1% | – | 2025-05-13 |
| 50 |  Phi 3.5 Vision | 43.0% | – | 2024-05-21 |
More multimodal benchmarks · All LLM benchmarks
Cite as: BenchLeader, “MMMU (validation) leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/mmmu_val.
MMMU (validation): questions
- What does MMMU (validation) measure?
- MMMU is a college-level multimodal exam: about 900 validation questions with diagrams, charts, photographs and tables across six broad disciplines and thirty subjects. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads MMMU (validation)?
- GPT-5.1 leads MMMU (validation) with 85.4% as of 25 Oct 2026, ahead of Gemini 2.5 Deep Think at 84.0%.
- How many models have MMMU (validation) results?
- 50 model configurations have a MMMU (validation) result on BenchLeader, all taken from MMMU.
- Who runs MMMU (validation) and how often is it updated?
- MMMU (validation) is published by MMMU. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does MMMU (validation) count toward the BenchLeader Index?
- Yes. MMMU (validation) contributes to the multimodal category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.