Skip to content

Benchmark

VITA-Bench

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

Models measured
29
Scores recorded
42
Sources
3
IQ weight
1.05

Score distribution

27.1%
Mean
26.3%
Median
13.2%
Std dev
12.3%
10th percentile
45.6%
90th percentile
0.8%
Min
0.8%49.7%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.05 for chat models, measured over 29 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Qwen3.5 397B A17B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim49.7%
2Qwen3.5 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.49.7%
3Qwen3.7 Max· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim47.9%
4Qwen3.7 MaxQwen3.7-Max release · subdomain average · Claude 4.5 Sonnet judge· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.47.9%
5Qwen3.7 Plus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim45.6%
6Qwen3.7 PlusQwen3.7-Plus release · subdomain average · Claude 4.5 Sonnet judge· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.45.6%
7Qwen3.6 Plus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim44.3%
8Qwen3.6 PlusQwen official release table · thinking mode unless benchmark requires otherwise · average subdomain score · Claude Sonnet 4 judge· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.44.3%
9Qwen3.5 27B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim41.9%
10Qwen3 Max· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim40.9%
11Qwen3.6 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim35.6%
12Qwen3.6 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.35.6%
13Qwen3.5 122B-A10B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim33.6%
14Gemini 3 Flashthinking high· VitaBench32.5%
15Qwen3.5 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim31.9%
16Gemini 3 Pro Previewthinking high· VitaBench31.5%
17Gemini 3 Pro Previewnon-thinking low· VitaBench30%
18Qwen3.5 9B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim29.8%
19LongCat-Flash-Thinkingthinking· VitaBench29.3%
20Claude Opus 4.5thinking· VitaBench28.5%
21OpenAI o3thinking high· VitaBench26.3%
22GPT 5.2thinking xhigh· VitaBench24.3%
23DeepSeek V3.2thinking· VitaBench24%
24Claude Sonnet 4.5thinking· VitaBench23.5%
25Claude Opus 4.5non-thinking· VitaBench23.3%
26LongCat Flash Chatnon-thinking· VitaBench22.8%
27Qwen3.5 4B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim22%
28o4 Minithinking high· VitaBench19.5%
29Doubao Seed 1.8thinking high, thinking· VitaBench18.8%
30DeepSeek V3.2non-thinking· VitaBench18.5%
31GLM 4.7thinking· VitaBench18.3%
32Claude Sonnet 4.5non-thinking· VitaBench17%
33GLM 4.7non-thinking· VitaBench15.5%
34Qwen 3 235B A22B Thinking 2507thinking· VitaBench14.5%
35Qwen3 Maxnon-thinking· VitaBench14.3%
36Doubao Seed 1.8non-thinking· VitaBench13.8%
37Kimi K2 Thinkingthinking· VitaBench12.8%
38Qwen3 235B A22B Instruct 2507non-thinking· VitaBench12.3%
39Kimi K2 0905non-thinking· VitaBench11.5%
40Qwen3 32Bthinking· VitaBench5.3%
41Qwen3 32Bnon-thinking· VitaBench4%
42GPT 5.2non-thinking none· VitaBench0.8%

Data credits: Data by LLM Stats · 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.