Benchmark
ScreenSpot Pro - CAD
Vision benchmarks measure how well models understand images. Scores are percentages.
- Models measured
- 8
- Scores recorded
- 10
- Sources
- 1
- IQ weight
- 0.95
Score distribution
- 24.6%
- Mean
- 14.2%
- Median
- 23.8%
- Std dev
- 1.5%
- 10th percentile
- 54.8%
- 90th percentile
- 0.77%
- Min
This benchmark has a weight of 0.95 for chat models, measured over 8 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 | Holo2 30B A3BAgentic· ScreenSpot data | 72.8% |
| 2 | Holo2 30B A3B· ScreenSpot data | 54.8% |
| 3 | Qwen2.5 VL 72B Instruct· ScreenSpot data | 44.4% |
| 4 | Qwen 2.5 VL 32B Instruct· ScreenSpot data | 31% |
| 5 | GPT 5resized, minimal· ScreenSpot data | 14.2% |
| 6 | Qwen2.5-VL 7B Instruct· ScreenSpot data | 13% |
| 7 | qwen2.5-vl-3b-instruct· ScreenSpot data | 7.3% |
| 8 | GPT 5resized, high· ScreenSpot data | 5.8% |
| 9 | GPT 4o· ScreenSpot data | 1.5% |
| 10 | Qwen2-VL (72B) Instruct· ScreenSpot data | 0.77% |
Data credits: ScreenSpot grounding leaderboard (GUI-Agent), Apache-2.0 (licence)