BlogLLM Benchmark Scores as Published, Side by Side
MMMU: Results as Published
Every result the catalog holds for MMMU: 41 models from 9 developers, read from 15 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area Multimodal UnderstandingMaintainer MMMU team (Xiang Yue and co-authors; Ohio State University, Carnegie Mellon University and partners)Introduced 2023Data version 2026103001, as of 2026-10-30
MMMU (Massive Multi-discipline Multimodal Understanding) tests college-level reasoning over images and text together. It holds 11,500 questions drawn from exams, quizzes and textbooks across six disciplines (art and design, business, science, health and medicine, humanities and social science, technology and engineering) with 30 image types such as charts, diagrams, chemical structures and sheet music. Accuracy on the questions is the score.
MMMU (validation)
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.2OpenAI, released 2025-12-11 | 86.7% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15 | - |
| GPT-5.1OpenAI, released 2025-11-12 | 85.4% | OpenAI, Introducing GPT-5.1 for developerschecked 2026-10-15; printed on 1 more page | Effort: high |
| GPT-5OpenAI, released 2025-08-07 | 84.2% | OpenAI, Introducing GPT-5checked 2026-10-15; printed on 3 more pages | Effort: high |
| 74.4% | OpenAI, Introducing GPT-5checked 2026-10-15 | Effort: none | |
| GPT-5 miniOpenAI, released 2025-08-07 | 79% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15 | - |
| o3OpenAI, released 2025-04-16 | 82.9% | OpenAI, Introducing GPT-5checked 2026-10-15; printed on 2 more pages | Effort: high |
| GPT-4.1OpenAI, released 2025-04-14 | 74.8% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15 | - |
| o1OpenAI, released 2024-12-17 | 78.2% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | - |
| GPT-4oOpenAI, released 2024-05-13 | 72.2% | OpenAI, Introducing GPT-5checked 2026-10-15; printed on 1 more page | - |
| Claude Opus 4.5Anthropic, released 2025-11-24 | 80.7% | Anthropic, Claude Opus 4.5checked 2026-10-15; printed on 2 more pages | Effort: high |
| Claude Haiku 4.5Anthropic, released 2025-10-15 | 73.2% | Anthropic, Claude Haiku 4.5checked 2026-10-15 | Effort: thinking budget 128K |
| Claude Sonnet 4.5Anthropic, released 2025-09-29 | 77.8% | Anthropic, Claude Sonnet 4.5checked 2026-10-15; printed on 4 more pages | - |
| Claude Opus 4.1Anthropic, released 2025-08-05 | 77.1% | Anthropic, Claude Opus 4.1checked 2026-10-15; printed on 2 more pages | Effort: extended thinking, up to 64K |
| Claude Opus 4Anthropic, released 2025-05-22 | 76.5% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15; printed on 1 more page | Effort: extended thinking, up to 64K |
| Claude Sonnet 4Anthropic, released 2025-05-22 | 74.4% | Anthropic, Claude Sonnet 4.5checked 2026-10-15; printed on 3 more pages | - |
| Claude 3.7 SonnetAnthropic, released 2025-02-24 | 75% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15; printed on 2 more pages | Effort: extended thinking, 64K |
| 71.8% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | Effort: none | |
| Claude 3.5 SonnetAnthropic, released 2024-10-22 | 70.4% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | - |
| Gemma 4 26B A4BGoogle, released 2026-04-02 | 78.4% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15 | - |
| Gemma 4 31BGoogle, released 2026-04-02 | 80.4% | Alibaba, Qwen3.6-27B model cardchecked 2026-10-15; printed on 1 more page | - |
| Gemini 3 ProGoogle, released 2025-11-18 | 87.2% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15 | - |
| Gemini 2.5 ProGoogle, released 2025-03-25 | 79.6% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15; printed on 3 more pages | - |
| Gemma 3 12BGoogle, released 2025-03-12 | 50% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Harness: VLMEvalKit |
| Grok 3xAI, released 2025-02-17 | 76% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | Effort: extended thinking |
| 78% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | Effort: extended thinking; Attempt budget: not stated (second value of the printed pair) | |
| Llama 4 MaverickMeta, released 2025-04-05 | 73.4% | Meta, Meta Llama 4 model cardchecked 2026-10-15; printed on 1 more page | - |
| Llama 4 ScoutMeta, released 2025-04-05 | 69.4% | Meta, Meta Llama 4 model cardchecked 2026-10-15 | - |
| Qwen3.6-27BAlibaba, released 2026-04-22 | 82.9% | Alibaba, Qwen3.6-27B model cardchecked 2026-10-15 | - |
| Qwen3.6-35B-A3BAlibaba, released 2026-04-16 | 81.7% | Alibaba, Qwen3.6-35B-A3B model cardchecked 2026-10-15; printed on 1 more page | - |
| Qwen3.5-122B-A10BAlibaba, released 2026-02-24 | 83.9% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15 | - |
| Qwen3.5-27BAlibaba, released 2026-02-24 | 82.3% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15; printed on 2 more pages | - |
| Qwen3.5-35B-A3BAlibaba, released 2026-02-24 | 81.4% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15; printed on 1 more page | - |
| Qwen3.5-397B-A17BAlibaba, released 2026-02-16 | 85% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15; printed on 1 more page | - |
| Qwen3-VL-32B-InstructAlibaba, released 2025-10-21 | 70.6% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Harness: VLMEvalKit; Context: 32768 |
| 68.6% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Harness: VLMEvalKit; Context: 4096 | |
| Qwen3-VL-32B-ThinkingAlibaba, released 2025-10-21 | 72.2% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Effort: thinking; Harness: VLMEvalKit; Context: 40960 |
| 72% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Effort: thinking; Harness: VLMEvalKit; Context: 4096 | |
| Qwen3-VL-8B-InstructAlibaba, released 2025-10-15 | 64.6% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Harness: VLMEvalKit; Context: 32768 |
| 60.7% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Harness: VLMEvalKit; Context: 4096 | |
| Qwen3-VL-8B-ThinkingAlibaba, released 2025-10-15 | 65.3% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Effort: thinking; Harness: VLMEvalKit; Context: 40960 |
| 59.3% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Effort: thinking; Harness: VLMEvalKit; Context: 4096 | |
| Qwen3-VL-235B-A22BAlibaba, released 2025-09-23 | 80.6% | Alibaba, Qwen3.5-122B-A10B model cardchecked 2026-10-15; printed on 1 more page | - |
| Kimi K2.5Moonshot AI, released 2026-01-27 | 84.3% | Alibaba, Qwen3.5-397B-A17B model cardchecked 2026-10-15 | - |
| Kimi-VL-A3B-InstructMoonshot AI, released 2025-04-10 | 52% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Harness: VLMEvalKit |
| Kimi-VL-A3B-ThinkingMoonshot AI, released 2025-04-10 | 60.2% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Effort: thinking; Harness: VLMEvalKit |
| Mistral Medium 3Mistral AI, released 2025-05-07 | 66.1% | Mistral AI, Medium is the new largechecked 2026-10-15 | - |
| Phi-4-reasoning-vision-15BMicrosoft, released 2026-03-04 | 54.3% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Harness: VLMEvalKit |
| 52% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Effort: force nothink; Harness: VLMEvalKit | |
| 55% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Effort: force thinking; Harness: VLMEvalKit | |
| Phi-4-multimodal-instructMicrosoft, released 2025-02-26 | 42.3% | Microsoft, Microsoft Phi-4-reasoning-vision-15B model cardchecked 2026-10-15 | Harness: VLMEvalKit |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.