MMMU-Pro (official)
Official MMMU-Pro leaderboard. Shown for reference.
As of 25 Oct 2026, Gemini 3 Pro leads MMMU-Pro (official) on BenchLeader with 81.0%, ahead of Gemini 3.1 Pro at 80.5%, across 28 model configurations with a published result.
- Published by
- MMMU
- Category
- Multimodal
- Index weight
- Reference only
- Models
- 28
- Data as of
- 25 Oct 2026
- Newest result
- 8 Apr 2026
The publisher has posted no new results on this board since 8 Apr 2026. 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-Pro is a harder version of the college-level multimodal exam MMMU: questions with images from science, engineering, medicine and the arts, with ten answer options and a vision-only variant where the question itself is inside the image.
How it is scored
Accuracy as submitted to the official leaderboard.
What to keep in mind
The official MMMU-Pro leaderboard: the harder ten-option version of MMMU with a vision-only setting. Self-submitted runs; shown for reference next to Artificial Analysis' independent run. See also the MMMU-Pro board.
- 1Gemini 3 Pro81.0%
- 2Gemini 3.1 Pro (high)80.5%
- 3Muse Spark (thinking)80.4%
- 4GPT-5.1 (thinking)79.0%
- 5Gemma 4 31B76.9%
- 6o376.4%
- 7GPT-5.176.0%
- 8Claude Opus 4.573.9%
- 9Gemma 4 26B A4B73.8%
- 10Claude Sonnet 4.568.9%
- 11Qwen3 VL 235B A22B Instruct68.1%
- 12Gemini 2.5 Pro68.0%
- 13Seed 1.6 Thinking (thinking)66.4%
- 14Gemma 4 E4B52.6%
- 15GPT-4o51.9%
| # | ||||
|---|---|---|---|---|
| 1 |  Gemini 3 Pro | 81.0% | 60.4 | 2025-11-18 |
| 2 | Gemini 3.1 Prohigh | 80.5% | 60.3 | 2026-02-19 |
| 3 |  Muse Sparkthinking | 80.4% | – | 2026-04-08 |
| 4 |  GPT-5.1thinking | 79.0% | 54.1 | 2025-11-13 |
| 5 | Gemma 4 31Bopen ↗ | 76.9% | 48.4 | 2026-04-02 |
| 6 | o3 | 76.4% | 59.5 | 2025-04-16 |
| 7 | GPT-5.1 | 76.0% | 56.8 | 2025-11-13 |
| 8 |  Claude Opus 4.5 | 73.9% | 57.9 | 2025-11-24 |
| 9 | Gemma 4 26B A4Bopen ↗ | 73.8% | 56.2 | 2026-04-02 |
| 10 | Claude Sonnet 4.5 | 68.9% | 53.7 | 2025-09-29 |
| 11 |  Qwen3 VL 235B A22B Instructopen ↗ | 68.1% | 48.7 | 2025-09-22 |
| 12 | Gemini 2.5 Pro | 68.0% | 53.9 | 2025-05-06 |
| 13 |  Seed 1.6 Thinkingthinking | 66.4% | – | 2025-06-25 |
| 14 | Gemma 4 E4Bopen ↗ | 52.6% | – | 2026-04-02 |
| 15 | GPT-4o | 51.9% | 43.9 | 2024-05-13 |
| 16 | Claude 3.5 Sonnet | 51.5% | 46.3 | 2024-06-20 |
| 17 |  InternVL2_5-78B | 48.6% | – | 2024-12-05 |
| 18 | Gemini 1.5 Pro | 46.9% | 46.1 | 2024-08-01 |
| 19 | Qwen2-VL-72B | 46.2% | – | 2024-08-29 |
| 20 | Qwen2.5 VL 72B Instructopen ↗ | 46.2% | – | 2025-02-19 |
| 21 | Gemma 4 E2Bopen ↗ | 44.2% | – | 2026-04-02 |
| 22 | Llama 3.2 90B | 39.5% | – | 2024-09-25 |
| 23 | Qwen2.5-VL 7B Instructopen ↗ | 38.3% | – | 2025-02-19 |
| 24 | GPT-4o mini | 37.6% | 35.8 | 2024-07-18 |
| 25 | InternVL2-40B | 34.2% | – | 2024-07-04 |
| 26 |  LLaVA-OneVision 72B | 31.0% | – | 2024-08-06 |
| 27 | Minicpm V 2.6 | 27.2% | – | 2024-08-06 |
| 28 |  Phi 3.5 Vision | 19.7% | – | 2024-05-21 |
More multimodal benchmarks · All LLM benchmarks
Cite as: BenchLeader, “MMMU-Pro (official) leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/mmmu_pro_official.
MMMU-Pro (official): questions
- What does MMMU-Pro (official) measure?
- MMMU-Pro is a harder version of the college-level multimodal exam MMMU: questions with images from science, engineering, medicine and the arts, with ten answer options and a vision-only variant where the question itself is inside the image. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads MMMU-Pro (official)?
- Gemini 3 Pro leads MMMU-Pro (official) with 81.0% as of 25 Oct 2026, ahead of Gemini 3.1 Pro at 80.5%.
- How many models have MMMU-Pro (official) results?
- 28 model configurations have a MMMU-Pro (official) result on BenchLeader, all taken from MMMU.
- Who runs MMMU-Pro (official) and how often is it updated?
- MMMU-Pro (official) is published by MMMU. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does MMMU-Pro (official) count toward the BenchLeader Index?
- No. MMMU-Pro (official) is shown for reference but left out of the composite index.