Benchmark
wildvision-arena-elo
Vision benchmarks measure how well models understand images. Higher scores are better.
All vision benchmarksScores from WildVision Arena data.
- Models measured
- 20
- Scores recorded
- 20
- Sources
- 1
- IQ weight
- 1.25
Score distribution
- 1,058.6
- Mean
- 1,055.7
- Median
- 56.7
- Std dev
- 1,003.1
- 10th percentile
- 1,117.9
- 90th percentile
- 977.8
- Min
This benchmark has a weight of 1.25 for chat models, measured over 20 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | GPT 4o· WildVision Arena data | 1,189.6 |
| 2 | Claude 3.5 Sonnet· WildVision Arena data | 1,159.7 |
| 3 | gemini-1.5-pro-latest· WildVision Arena data | 1,117.9 |
| 4 | GPT-4 Turbo· WildVision Arena data | 1,110.7 |
| 5 | GPT-4o mini· WildVision Arena data | 1,107.5 |
| 6 | Pixtral 12B2409· WildVision Arena data | 1,102.6 |
| 7 | GPT-4-vision-preview· WildVision Arena data | 1,083.9 |
| 8 | Gemini 1.5 Flashlatest· WildVision Arena data | 1,065.3 |
| 9 | Qwen2 VL 7b Instruct· WildVision Arena data | 1,060.3 |
| 10 | Claude 3 Opus· WildVision Arena data | 1,055.7 |
| 11 | Reka Flash· WildVision Arena data | 1,043.8 |
| 12 | Qwen-VL Max· WildVision Arena data | 1,037.6 |
| 13 | LLaVA-OneVision Qwen2 72B OV· WildVision Arena data | 1,023.2 |
| 14 | Gemini Pro Vision· WildVision Arena data | 1,013.7 |
| 15 | Reka Core· WildVision Arena data | 1,010.9 |
| 16 | Phi-3 Vision 128K Instruct· WildVision Arena data | 1,010.9 |
| 17 | LLaVA 1.6 34B· WildVision Arena data | 1,009.4 |
| 18 | Claude 3 Sonnet· WildVision Arena data | 1,003.1 |
| 19 | Qwen-VL Plus· WildVision Arena data | 987.5 |
| 20 | Claude 3 Haiku· WildVision Arena data | 977.8 |
Data credits: WildVision Arena, MIT (licence)