Benchmark leaderboard - 26 entries
MMMU leaderboard
MMMU evaluates multimodal models on massive multi-discipline tasks demanding college-level subject knowledge and deliberate reasoning, using 11.5K multimodal questions from college exams, quizzes, and textbooks across six disciplines, 30 subjects and 183 subfields. As of October 17, 2026, GPT-5.1 ranks first on the MMMU board with 85.4*.
Last verified from mmmu-benchmark.github.io. Read the caveats
MMMU
26 entries, in the board's order
| 1 | GPT-5.1 MMMU (Val) Overall, zero-shot; rank 1 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://openai.com/index/gpt-5-1-for-developers/ Official leaderboard | 85.4* |
| 2 | DreamPRM-1.5 (GPT-5-mini w/ thinking) MMMU (Val) Overall, zero-shot; rank 2 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://github.com/coder-qicao/DreamPRM-1.5 Official leaderboard | 84.6* |
| 3 | GPT-5 w/ thinking MMMU (Val) Overall, zero-shot; rank 3 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://openai.com/gpt-5/ Official leaderboard | 84.2* |
| 4 | Gemini 2.5 Pro Deep-Think MMMU (Val) Overall, zero-shot; rank 4 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://deepmind.google/technologies/gemini/pro/ Official leaderboard | 84.0* |
| 5 | o3 MMMU (Val) Overall, zero-shot; rank 5 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://openai.com/index/introducing-o3-and-o4-mini/ Official leaderboard | 82.9* |
| 6 | o4-mini MMMU (Val) Overall, zero-shot; rank 6 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://openai.com/index/introducing-o3-and-o4-mini/ Official leaderboard | 81.6* |
| 7 | Claude Opus 4.5 MMMU (Val) Overall, zero-shot; rank 7 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://www.anthropic.com/news/claude-opus-4-5 Official leaderboard | 80.7* |
| 8 | dots.vlm1 MMMU (Val) Overall, zero-shot; rank 8 among model entries (Human Expert rows excluded); size: 37B; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://github.com/rednote-hilab/dots.vlm1 Official leaderboard | 80.1* |
| 9 | Gemini 2.5 Flash 05-20 MMMU (Val) Overall, zero-shot; rank 9 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://deepmind.google/models/gemini/flash/ Official leaderboard | 79.7* |
| 10 | Gemini 2.5 Pro 05-06 MMMU (Val) Overall, zero-shot; rank 10 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://deepmind.google/technologies/gemini/pro/ Official leaderboard | 79.6* |
| 11 | Qwen3-VL 235B-A22B MMMU (Val) Overall, zero-shot; rank 11 among model entries (Human Expert rows excluded); size: 22B; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://qwen.ai/blog?id=99f0335c4ad9ff6153e517418d48535ab6d8afef&from=research.latest-advancements-list Official leaderboard | 78.7* |
| 12 | o1 MMMU (Val) Overall, zero-shot; rank 12 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://openai.com/index/introducing-openai-o1-preview/ Official leaderboard | 78.2* |
| 13 | Grok 3 Beta MMMU (Val) Overall, zero-shot; rank 13 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://x.ai/news/grok-3 Official leaderboard | 78.0* |
| 14 | Seed 1.5-VL Thinking MMMU (Val) Overall, zero-shot; rank 14 among model entries (Human Expert rows excluded); size: 20B; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://seed.bytedance.com/en/tech/seed1_5_vl Official leaderboard | 77.9* |
| 15 | Claude Sonnet 4.5 MMMU (Val) Overall, zero-shot; rank 15 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://www.anthropic.com/news/claude-sonnet-4-5 Official leaderboard | 77.8* |
| 16 | Intern-S1 MMMU (Val) Overall, zero-shot; rank 16 among model entries (Human Expert rows excluded); size: 22B; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://github.com/InternLM/Intern-S1/ Official leaderboard | 77.7* |
| 17 | Claude Opus 4.1 MMMU (Val) Overall, zero-shot; rank 17 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://www.anthropic.com/news/claude-opus-4-1 Official leaderboard | 77.1* |
| 18 | Claude Opus 4 MMMU (Val) Overall, zero-shot; rank 18 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://www.anthropic.com/news/claude-4 Official leaderboard | 76.5* |
| 19 | Llama 4 Behemoth MMMU (Val) Overall, zero-shot; rank 19 among model entries (Human Expert rows excluded); size: 288B; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://ai.meta.com/blog/llama-4-multimodal-intelligence/ Official leaderboard | 76.1* |
| 20 | Skywork-R1V3-38B MMMU (Val) Overall, zero-shot; rank 20 among model entries (Human Expert rows excluded); size: 38B; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://huggingface.co/Skywork/Skywork-R1V3-38B Official leaderboard | 76.0* |
| 21 | GLM-4.5V w/ Thinking MMMU (Val) Overall, zero-shot; rank 21 among model entries (Human Expert rows excluded); size: 12B; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://docs.z.ai/guides/vlm/glm-4.5v Official leaderboard | 75.4* |
| 22 | Claude 3.7 Sonnet MMMU (Val) Overall, zero-shot; rank 22 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://www.anthropic.com/claude/sonnet Official leaderboard | 75.0* |
| 23 | Seed 1.6-Thinking MMMU (Val) Overall, zero-shot; rank 23 among model entries (Human Expert rows excluded); size: 20B; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://seed.bytedance.com/en/seed1_6 Official leaderboard | 74.8* |
| 24 | GPT-5 w/o thinking MMMU (Val) Overall, zero-shot; tied rank 24 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://openai.com/gpt-5/ Official leaderboard | 74.4* |
| 25 | GPT-4.5 MMMU (Val) Overall, zero-shot; tied rank 24 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026 Official leaderboard | 74.4* |
| 26 | Claude Sonnet 4 MMMU (Val) Overall, zero-shot; tied rank 24 among model entries (Human Expert rows excluded); size: -; * = results provided by the authors; leaderboard last updated 07/01/2026; model link on leaderboard: https://www.anthropic.com/news/claude-4 Official leaderboard | 74.4* |
About this board
- Ranking column choice: the official leaderboard (https://mmmu-benchmark.github.io/#leaderboard, 'Last updated: 07/01/2026') sorts by MMMU-Pro Overall by default; these entries rank by MMMU (Val) Overall, the classic MMMU score. 26 entries returned because three models tie at 74.4* for rank 24.
- Newer frontier models appear on the official leaderboard with MMMU-Pro scores only and no MMMU (Val) score, so they are not in this list: Chance Vision 1.5 (MMMU-Pro 86.9*), GPT-5.4 Thinking w/ tools (82.1*), GPT-5.4 Thinking w/o tools (81.2*), Gemini 3.0 Pro (81.0*), Gemini 3.1 Pro Thinking (High) (80.5*), Muse Spark Thinking (80.4*), GPT-5.2 Thinking w/o Python (80.4*), GPT-5.2 Thinking w/o tools (79.5*), GPT-5.1 Thinking (79.0*), Claude Opus 4.6 w/ tools (77.3*), Gemma 4 31B (76.9*), Claude Sonnet 4.6 w/ tools (75.6*). A separate MMMU-Pro leaderboard page could cover them.
- Human Expert rows on the MMMU (Val) column (High 88.6, Medium 82.6, Low 76.2; dated 2024-01-31) were left out of the model ranking.
- MMMU (Test) Overall: no current frontier model has a test-set score on the leaderboard (the top test score shown is KiQ-v0 at 65.9), so the test split was not used for ranking.
- GPT-4.5 leaderboard row links to a broken in-site URL (https://mmmu-benchmark.github.io/GPT%E2%80%914.5), so no vendor link is included for it.
FAQ
What does MMMU measure?
MMMU evaluates multimodal models on massive multi-discipline tasks demanding college-level subject knowledge and deliberate reasoning, using 11.5K multimodal questions from college exams, quizzes, and textbooks across six disciplines, 30 subjects and 183 subfields.
Which model leads MMMU?
As of October 17, 2026, GPT-5.1 ranks first on the MMMU board with 85.4*. That is the board's own order, not a Developers Digest ranking.
Where do these MMMU scores come from?
All 26 entries come from the official leaderboard at mmmu-benchmark.github.io. Every entry links to its source, which was read on October 17, 2026, and lists the setting it was run under.
Related reading
- Kimi K3 vs K2.7 Code: Is the Upgrade Worth It for Coding?Kimi K3 vs K2.7 Code: 1M vs 256K context, $3/$15 vs $0.95/$4 per million tokens, and when the bigger model earns its price on coding work.
- Inkling: Thinking Machines Lab Drops a 975B Open-Weights ModelA new American open-weights frontier model with multimodal capabilities, 1M token context, and competitive benchmarks. Here's what the HN community thinks.
Sources
Last verified . Each row links to the page its score was copied from.
Newsletter
Get new benchmark results as they land
One email per week. New model releases, the leaderboard moves that matter and what they mean for what you build.