MUSE
Models answer questions about artworks used in language education. The tasks cover visual details, emotions, cultural context, and relationships within the image.
- Made by
- Luyao Zhu and colleagues
- Released
- September 2026
- Kind
- Other professions
- Level
- School
- Scored
- Checked automatically
- Answers
- Multiple choice
- Measures
- Capability
- Subject
- Arts, Languages and literature
- Language
- English
- Interaction
- Single turn
- Items from
- Real work and use
- Refresh
- One fixed set
- Run by
- A host
- In views
- Vision
- Strongest model tested
- GPT-5.6 Sol #14 on the capability index
- Last model added
- July 2026
- Scores refreshed
- November 20th, 2026
GPT-5.6 Sol leads with 56.7% among models released by July 2026 GPT-5.6 Sol leads with 56.7% among models released by July 2026
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Table 2 ranks models by average performance. The headline is the mean of its twelve task scores. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in July 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 15 more
- 56.7 GPT-5.6 Sol
- 48.5
Qwen3-VL-32B-Instruct
- 43.0
Qwen2.5-VL-72B
- 42.1
Qwen3-VL-8B-Instruct
- 41.6
InternVL3-38B
- 39.7
Qwen2.5-VL-32B
- 39.5
InternVL3-14B
- 38.4 GPT-4o (snapshot unspecified)
- 35.6
Qwen2.5-VL-7B
- 35.4
Gemma 3 27B
- 35.0
Gemma 3 12B
- 33.9
InternVL3-9B
- 33.1
DeepSeek-VL2
- 33.0 MiniCPM-V-2_6
- 32.9 MiniCPM-o-2_6
- 32.4
InternVL3-8B
- 32.4 GLM-4V-9B
- 32.1
MiniCPM-Llama3-V 2.5
- 29.4
DeepSeek-VL2-Small
- 29.2
Qwen2.5-VL-3B
- 28.9
Gemma 3 4B
- 28.6 llama3-llava-next-8b
- 28.3
InternVL3-2B
- 27.0
InternVL3-1B
- 26.2
Yi-VL-6B
- 25.8 CogVLM2-19B-Chat
- 25.4
Yi-VL-34B
- 22.6
LLaVA-v1.6-34B
- 22.3
LLaVA-NeXT-72B
- 22.2
DeepSeek-VL2-Tiny
- 1 GPT-5.6 Sol 56.7%
- 2
Qwen3-VL-32B-Instruct 48.5%
- 3
Qwen2.5-VL-72B 43%
- 4
Qwen3-VL-8B-Instruct 42.1%
- 5
InternVL3-38B 41.6%
- 6
Qwen2.5-VL-32B 39.7%
- 7
InternVL3-14B 39.5%
- 8 GPT-4o (snapshot unspecified) 38.4%
- 9
Qwen2.5-VL-7B 35.6%
- 10
Gemma 3 27B 35.4%
- 11
Gemma 3 12B 35%
- 12
InternVL3-9B 33.9%
- 13
DeepSeek-VL2 33.1%
- 14 MiniCPM-V-2_6 33%
- 15 MiniCPM-o-2_6 32.9%
- 16
InternVL3-8B 32.4%
- 17 GLM-4V-9B 32.4%
- 18
MiniCPM-Llama3-V 2.5 32.1%
- 19
DeepSeek-VL2-Small 29.4%
- 20
Qwen2.5-VL-3B 29.2%
- 21
Gemma 3 4B 28.9%
- 22 llama3-llava-next-8b 28.6%
- 23
InternVL3-2B 28.3%
- 24
InternVL3-1B 27%
- 25
Yi-VL-6B 26.2%
- 26 CogVLM2-19B-Chat 25.8%
- 27
Yi-VL-34B 25.4%
- 28
LLaVA-v1.6-34B 22.6%
- 29
LLaVA-NeXT-72B 22.3%
- 30
DeepSeek-VL2-Tiny 22.2%
Source: The MUSE paper
- 56.7 GPT-5.6 Sol
- 48.5
Qwen3-VL-32B-Instruct
- 43.0
Qwen2.5-VL-72B
- 42.1
Qwen3-VL-8B-Instruct
- 41.6
InternVL3-38B
- 39.7
Qwen2.5-VL-32B
- 39.5
InternVL3-14B
- 35.6
Qwen2.5-VL-7B
- 35.4
Gemma 3 27B
- 35.0
Gemma 3 12B
- 33.9
InternVL3-9B
- 33.1
DeepSeek-VL2
- 33.0 MiniCPM-V-2_6
- 32.9 MiniCPM-o-2_6
- 32.4
InternVL3-8B
- 32.4 GLM-4V-9B
- 32.1
MiniCPM-Llama3-V 2.5
- 29.4
DeepSeek-VL2-Small
- 29.2
Qwen2.5-VL-3B
- 28.9
Gemma 3 4B
- 28.6 llama3-llava-next-8b
- 28.3
InternVL3-2B
- 27.0
InternVL3-1B
- 26.2
Yi-VL-6B
- 25.8 CogVLM2-19B-Chat
- 25.4
Yi-VL-34B
- 22.6
LLaVA-v1.6-34B
- 22.3
LLaVA-NeXT-72B
- 22.2
DeepSeek-VL2-Tiny
- 1 GPT-5.6 Sol 56.7%
- 2
Qwen3-VL-32B-Instruct 48.5%
- 3
Qwen2.5-VL-72B 43%
- 4
Qwen3-VL-8B-Instruct 42.1%
- 5
InternVL3-38B 41.6%
- 6
Qwen2.5-VL-32B 39.7%
- 7
InternVL3-14B 39.5%
- 8
Qwen2.5-VL-7B 35.6%
- 9
Gemma 3 27B 35.4%
- 10
Gemma 3 12B 35%
- 11
InternVL3-9B 33.9%
- 12
DeepSeek-VL2 33.1%
- 13 MiniCPM-V-2_6 33%
- 14 MiniCPM-o-2_6 32.9%
- 15
InternVL3-8B 32.4%
- 16 GLM-4V-9B 32.4%
- 17
MiniCPM-Llama3-V 2.5 32.1%
- 18
DeepSeek-VL2-Small 29.4%
- 19
Qwen2.5-VL-3B 29.2%
- 20
Gemma 3 4B 28.9%
- 21 llama3-llava-next-8b 28.6%
- 22
InternVL3-2B 28.3%
- 23
InternVL3-1B 27%
- 24
Yi-VL-6B 26.2%
- 25 CogVLM2-19B-Chat 25.8%
- 26
Yi-VL-34B 25.4%
- 27
LLaVA-v1.6-34B 22.6%
- 28
LLaVA-NeXT-72B 22.3%
- 29
DeepSeek-VL2-Tiny 22.2%
Source: The MUSE paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Sol | 56.7% | Jul 2026 | Official leaderboard | |
| 2 | | 48.5% | Oct 2025 | Official leaderboard | |
| 3 | | 43% | Jan 2025 | Official leaderboard | |
| 4 | | 42.1% | Oct 2025 | Official leaderboard | |
| 5 | | 41.6% | Apr 2025 | Official leaderboard | |
| 6 | | 39.7% | Mar 2025 | Official leaderboard | |
| 7 | | 39.5% | Apr 2025 | Official leaderboard | |
| 8 | GPT-4o (snapshot unspecified) | 38.4% | Official leaderboard | ||
| 9 | | 35.6% | Jan 2025 | Official leaderboard | |
| 10 | | 35.4% | Mar 2025 | Official leaderboard | |
| 11 | | 35% | Mar 2025 | Official leaderboard | |
| 12 | | 33.9% | Apr 2025 | Official leaderboard | |
| 13 | | 33.1% | Dec 2024 | Official leaderboard | |
| 14 | MiniCPM-V-2_6 | 33% | Aug 2024 | Official leaderboard | |
| 15 | MiniCPM-o-2_6 | 32.9% | Jan 2025 | Official leaderboard | |
| 16 | | 32.4% | Apr 2025 | Official leaderboard | |
| 17 | GLM-4V-9B | 32.4% | Jun 2024 | Official leaderboard | |
| 18 | | 32.1% | May 2024 | Official leaderboard | |
| 19 | | 29.4% | Dec 2024 | Official leaderboard | |
| 20 | | 29.2% | Jan 2025 | Official leaderboard | |
| 21 | | 28.9% | Mar 2025 | Official leaderboard | |
| 22 | llama3-llava-next-8b | 28.6% | Apr 2024 | Official leaderboard | |
| 23 | | 28.3% | Apr 2025 | Official leaderboard | |
| 24 | | 27% | Apr 2025 | Official leaderboard | |
| 25 | | 26.2% | Dec 2023 | Official leaderboard | |
| 26 | CogVLM2-19B-Chat | 25.8% | May 2024 | Official leaderboard | |
| 27 | | 25.4% | Dec 2023 | Official leaderboard | |
| 28 | | 22.6% | Jan 2024 | Official leaderboard | |
| 29 | | 22.3% | May 2024 | Official leaderboard | |
| 30 | | 22.2% | Dec 2024 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.6 Sol | 56.7% | Jul 2026 | Official leaderboard | |
| 2 | | 48.5% | Oct 2025 | Official leaderboard | |
| 3 | | 43% | Jan 2025 | Official leaderboard | |
| 4 | | 42.1% | Oct 2025 | Official leaderboard | |
| 5 | | 41.6% | Apr 2025 | Official leaderboard | |
| 6 | | 39.7% | Mar 2025 | Official leaderboard | |
| 7 | | 39.5% | Apr 2025 | Official leaderboard | |
| 8 | | 35.6% | Jan 2025 | Official leaderboard | |
| 9 | | 35.4% | Mar 2025 | Official leaderboard | |
| 10 | | 35% | Mar 2025 | Official leaderboard | |
| 11 | | 33.9% | Apr 2025 | Official leaderboard | |
| 12 | | 33.1% | Dec 2024 | Official leaderboard | |
| 13 | MiniCPM-V-2_6 | 33% | Aug 2024 | Official leaderboard | |
| 14 | MiniCPM-o-2_6 | 32.9% | Jan 2025 | Official leaderboard | |
| 15 | | 32.4% | Apr 2025 | Official leaderboard | |
| 16 | GLM-4V-9B | 32.4% | Jun 2024 | Official leaderboard | |
| 17 | | 32.1% | May 2024 | Official leaderboard | |
| 18 | | 29.4% | Dec 2024 | Official leaderboard | |
| 19 | | 29.2% | Jan 2025 | Official leaderboard | |
| 20 | | 28.9% | Mar 2025 | Official leaderboard | |
| 21 | llama3-llava-next-8b | 28.6% | Apr 2024 | Official leaderboard | |
| 22 | | 28.3% | Apr 2025 | Official leaderboard | |
| 23 | | 27% | Apr 2025 | Official leaderboard | |
| 24 | | 26.2% | Dec 2023 | Official leaderboard | |
| 25 | CogVLM2-19B-Chat | 25.8% | May 2024 | Official leaderboard | |
| 26 | | 25.4% | Dec 2023 | Official leaderboard | |
| 27 | | 22.6% | Jan 2024 | Official leaderboard | |
| 28 | | 22.3% | May 2024 | Official leaderboard | |
| 29 | | 22.2% | Dec 2024 | Official leaderboard |
Rankings by part
We read 12 parts from the benchmark’s paper in 5 groups: visual perception, semantic understanding, affective interpretation, compositional reasoning and cultural. Each ranks models on one part of the test. GPT-5.6 Sol leads 7 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Object Classification
29 models we track have a score here.
- 1 GPT-5.6 Sol 76%
- 2
Qwen3-VL-32B-Instruct 54%
- 3
Qwen2.5-VL-72B 52%
- 4
InternVL3-38B ↑1 49%
- 5
InternVL3-14B ↑2 48%
- 6
Qwen3-VL-8B-Instruct ↓2 46.5%
- 7
Qwen2.5-VL-32B ↓1 44.5%
- 8 llama3-llava-next-8b ↑13 38%
- 9 MiniCPM-o-2_6 ↑5 34.5%
- 10
InternVL3-8B ↑5 33%
Object Count
29 models we track have a score here.
- 1 GPT-5.6 Sol 71.5%
- 2
Qwen3-VL-32B-Instruct 59%
- 3
Qwen2.5-VL-72B 51%
- 4
DeepSeek-VL2 ↑8 50%
- 5
Qwen2.5-VL-32B ↑1 50%
- 6
Qwen3-VL-8B-Instruct ↓2 48.5%
- 7 MiniCPM-o-2_6 ↑7 47%
- 8
InternVL3-38B ↓3 47%
- 9 MiniCPM-V-2_6 ↑4 46%
- 10
Qwen2.5-VL-3B ↑9 45%
Activity Localization
29 models we track have a score here.
- 1
Qwen3-VL-32B-Instruct ↑1 72.5%
- 2 GPT-5.6 Sol ↓1 69.5%
- 3
InternVL3-38B ↑2 61.5%
- 4
Qwen3-VL-8B-Instruct 58%
- 5
Qwen2.5-VL-72B ↓2 56%
- 6
InternVL3-14B ↑1 55%
- 7
LLaVA-NeXT-72B ↑21 54.5%
- 8
Qwen2.5-VL-7B 50%
- 9 MiniCPM-o-2_6 ↑5 47.5%
- 10
Gemma 3 27B ↓1 47%
Activity Description
29 models we track have a score here.
- 1
Qwen3-VL-32B-Instruct ↑1 50%
- 2
Qwen2.5-VL-72B ↑1 47%
- 3
Yi-VL-6B ↑21 45%
- 4
InternVL3-14B ↑3 44.5%
- 5
Qwen3-VL-8B-Instruct ↓1 42.5%
- 6 MiniCPM-V-2_6 ↑7 41%
- 7
InternVL3-38B ↓2 38.5%
- 8
Qwen2.5-VL-3B ↑11 37%
- 9
LLaVA-v1.6-34B ↑18 36.5%
- 10
MiniCPM-Llama3-V 2.5 ↑7 36.5%
Scene Classification
29 models we track have a score here.
- 1
Qwen3-VL-32B-Instruct ↑1 87%
- 2 CogVLM2-19B-Chat ↑23 86.5%
- 3 GPT-5.6 Sol ↓2 86.5%
- 4
Qwen2.5-VL-72B ↓1 86%
- 5
InternVL3-38B 85%
- 6 MiniCPM-V-2_6 ↑7 84.5%
- 7
Qwen3-VL-8B-Instruct ↓3 84.5%
- 8
DeepSeek-VL2 ↑4 83%
- 9
Qwen2.5-VL-7B ↓1 83%
- 10
Qwen2.5-VL-32B ↓4 83%
Emotion Detection
29 models we track have a score here.
- 1 GPT-5.6 Sol 39.5%
- 2
Qwen3-VL-32B-Instruct 29.5%
- 3
Qwen3-VL-8B-Instruct ↑1 25%
- 4
Qwen2.5-VL-72B ↓1 23.5%
- 5
Qwen2.5-VL-32B ↑1 21.5%
- 6 MiniCPM-o-2_6 ↑8 19.5%
- 7
Gemma 3 27B ↑2 18%
- 8
InternVL3-38B ↓3 18%
- 9
Gemma 3 12B ↑1 17.5%
- 10
DeepSeek-VL2 ↑2 17%
Visual Clue Identification
29 models we track have a score here.
- 1 GPT-5.6 Sol 50.9%
- 2
Qwen3-VL-32B-Instruct 44.3%
- 3
Qwen3-VL-8B-Instruct ↑1 44.1%
- 4
Gemma 3 27B ↑5 42.2%
- 5
Gemma 3 12B ↑5 41.6%
- 6
DeepSeek-VL2-Small ↑12 41%
- 7
Qwen2.5-VL-72B ↓4 40.3%
- 8
DeepSeek-VL2 ↑4 40.1%
- 9 MiniCPM-V-2_6 ↑4 40.1%
- 10
Qwen2.5-VL-32B ↓4 39.6%
Emotion Cause Inference
29 models we track have a score here.
- 1 GPT-5.6 Sol 49.2%
- 2
Qwen3-VL-32B-Instruct 40.2%
- 3
Qwen3-VL-8B-Instruct ↑1 38.5%
- 4
Gemma 3 27B ↑5 35.7%
- 5
Qwen2.5-VL-32B ↑1 34.3%
- 6
Gemma 3 12B ↑4 33.7%
- 7
Qwen2.5-VL-72B ↓4 33.6%
- 8
DeepSeek-VL2 ↑4 32.1%
- 9
DeepSeek-VL2-Small ↑9 31.6%
- 10
InternVL3-38B ↓5 30.9%
Relative Position
29 models we track have a score here.
- 1
InternVL3-14B ↑6 10%
- 2
Qwen2.5-VL-72B ↑1 8.50%
- 3
InternVL3-38B ↑2 8.50%
- 4
Qwen2.5-VL-32B ↑2 7.50%
- 5
InternVL3-1B ↑18 7.50%
- 6
Qwen2.5-VL-3B ↑13 6.00%
- 7
Qwen3-VL-32B-Instruct ↓5 5.00%
- 8
Gemma 3 27B ↑1 4.50%
- 9 GPT-5.6 Sol ↓8 4.00%
- 10
LLaVA-NeXT-72B ↑18 3.00%
Remote Interaction
29 models we track have a score here.
- 1 GPT-5.6 Sol 86.5%
- 2
Qwen3-VL-32B-Instruct 52%
- 3
Qwen2.5-VL-72B 46%
- 4
Qwen2.5-VL-32B ↑2 44%
- 5
InternVL3-38B 43%
- 6
Qwen3-VL-8B-Instruct ↓2 40.5%
- 7
InternVL3-14B 39%
- 8
Qwen2.5-VL-7B 32%
- 9 GLM-4V-9B ↑7 27.5%
- 10
Gemma 3 12B 26.5%
Jigsaw Puzzle
29 models we track have a score here.
- 1 GLM-4V-9B ↑15 42%
- 2 GPT-5.6 Sol ↓1 35.5%
- 3
InternVL3-38B ↑2 31.5%
- 4
InternVL3-9B ↑7 31.5%
- 5
MiniCPM-Llama3-V 2.5 ↑12 28.5%
- 6
Qwen3-VL-32B-Instruct ↓4 28.5%
- 7
InternVL3-14B 28%
- 8
LLaVA-NeXT-72B ↑20 27%
- 9
InternVL3-1B ↑14 25.5%
- 10
DeepSeek-VL2 ↑2 25%
Cultural Identification
29 models we track have a score here.
- 1 GPT-5.6 Sol 76.5%
- 2
Qwen3-VL-32B-Instruct 60%
- 3
Qwen3-VL-8B-Instruct ↑1 58.5%
- 4
Gemma 3 12B ↑6 55%
- 5
Qwen2.5-VL-7B ↑3 54.5%
- 6 GLM-4V-9B ↑10 51.5%
- 7
Qwen2.5-VL-32B ↓1 51.5%
- 8
InternVL3-38B ↓3 51%
- 9
Qwen2.5-VL-72B ↓6 50.5%
- 10
DeepSeek-VL2 ↑2 50%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The MUSE paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MUSE leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Law, finance and medicine
- MedXpertQA-MM In use Models answer expert-level medical exam questions that combine images with clinical records and test results. Scoring is accuracy.
Qwen 3.8 Max 80.4%
- HealthBench Professional In use 525 physician-authored tasks from real clinician chats (care consult, documentation, research) with physician-adjudicated rubrics. Claude Sonnet 5.5 69.2%
- HealthBench In use 5,000 realistic health conversations graded against physician-written rubrics. A length-adjusted variant penalizes verbosity. GPT-5 67.2%
- FinSearchComp In use Agents search the web to fetch live financial data, look up history, and run multi-step investigations, across 635 expert-written questions. Scored by answer accuracy.
Ling-3.1-flash 81.7%
- KorMedMCQA In use Models answer 7,469 multiple-choice questions from Korean licensing exams for doctors, nurses, pharmacists and dentists, held from 2012 to 2024. o1-preview 92.7%
- Harvey LAB-AA v1.1 In use Artificial Analysis's run of Harvey's Legal Agent Benchmark on Harvey's newer private set of 120 tasks across 24 practice areas. A deliverable with a material hallucination scores zero, and the headline is the share of judges who find every rubric criterion passed. GPT-6.2 Astra 15%
Data audit
9 of 10 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark
Catalog record 1 warning 1 pass
- Warning Task access We don’t know whether the tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (30 entries).
- Pass Models scored 29 models we track, plus 1 other entry on its leaderboard.
- Pass Newest score The newest is from July 2026, for GPT-5.6 Sol.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.6 Sol (#14).
- Pass Headroom The best is 56.7%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 29 rows for models we track and 1 other entry at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.