Skip to content

Benchmark

MMMU

Vision benchmarks measure how well models understand images. Scores are percentages. Higher is better.

Models measured
76
Scores recorded
84
Sources
4
IQ weight
1.31

Score distribution

66.7%
Mean
69.4%
Median
14.3%
Std dev
50.7%
10th percentile
82.9%
90th percentile
0%
Min
0%86%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.31 for chat models, measured over 71 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Qwen3.6 Plus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim86%
2Qwen3.6 PlusQwen official release table · thinking mode unless benchmark requires otherwise· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86%
3GPT 5.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim85.4%
4GPT-5.1-Instant· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim85.4%
5Qwen3.5 397B A17BQwen3.5 release/model card · published setting· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.85%
6GPT 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim84.2%
7Qwen3.5 122B-A10B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim83.9%
8Doubao Seed 1.8· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim83.4%
9OpenAI o3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim82.9%
10Qwen3.6 27B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim82.9%
11Qwen3.6 27BQwen official release table · thinking mode unless benchmark requires otherwise· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.82.9%
12Qwen3.5 27B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim82.3%
13Gemini 2.5 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim82%
14Qwen3.6 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim81.7%
15Qwen3.6 FlashQwen3.6-35B-A3B official model card · hosted API name Qwen3.6-Flash· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.81.7%
16o4 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim81.6%
17Qwen3.5 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim81.4%
18Claude Opus 4.5validation · 64K thinking · interleaved scratchpads · 200K context · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.7%
19Qwen3.5 Omni PlusQwen3.5-Omni release · published setting· Reported by Qwen, Qwen Team· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.1%
20Gemini 2.5 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim79.7%
21Seed 2.0 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim79.7%
22Grok 3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim78%
23Claude Sonnet 4.5validation split· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.77.8%
24o1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim77.6%
25Qwen3.5 Omni FlashQwen3.5-Omni release · published setting· Reported by Qwen, Qwen Team· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.76.9%
26Claude Opus 4validation split · extended thinking up to 64K tokens· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.76.5%
27Gemini 2.0 Flashthinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim75.4%
28GPT-4.5 Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim75.2%
29Command A Plus 05 2026· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim75.1%
30Claude Sonnet 3.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim75%
31GPT 4.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim74.8%
32Claude Sonnet 4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim74.4%
33Claude Sonnet 4validation split · extended thinking up to 64K tokens· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.74.4%
34Claude Opus 4validation split · no extended thinking· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.7%
35Llama 4 Maverick· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim73.4%
36Claude Haiku 4.5validation split· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.2%
37Gemini 2.5 Flash-Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim72.9%
38GPT-4.1 mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim72.7%
39Claude Sonnet 4validation split · no extended thinking· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.72.6%
40GPT 4o· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
72.2%
41Gemini 2.0 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim70.7%
42QVQ 72B Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim70.3%
43Qwen2.5 VL 72B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim70.2%
44Llama 4 Scout· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim69.4%
45Llama 4 Scoutinstruction-tuned table· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.69.4%
46Claude 3.5 Sonnet· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim68.3%
47Gemini 2.0 Flash Lite· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim68%
48Grok 2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim66.1%
49Gemini 1.5 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim65.9%
50Pixtral Large· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim64%

Showing 1–50 of 84 results.

Data credits: WildVision Arena, MIT (licence) · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats