LLM Stats (ScreenSpot): leaderboard
GUI grounding benchmark testing the ability to locate specific screen elements (buttons, fields, icons) from natural language descriptions across mobile and desktop.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 32B Instruct | 95.8 | #283 |
| 2 | Qwen 3 VL 32B (Thinking) | 95.7 | #294 (Qwen 3 VL 32B) |
| 3 | Qwen 3 VL 235B A22B Instruct | 95.4 | #273 |
| 4 | Qwen 3 VL 235B A22B (Thinking) | 95.4 | #238 (Qwen 3 VL 235B A22B) |
| 5 | Qwen 3 VL 30B A3B Instruct | 94.7 | #384 |
| 6 | Qwen 3 VL 30B A3B (Thinking) | 94.7 | #351 (Qwen 3 VL 30B A3B) |
| 7 | Qwen 3 VL 8B Instruct | 94.4 | #420 |
| 8 | Qwen 3 VL 4B Instruct | 94 | #515 |
| 9 | Qwen 3 VL 8B (Thinking) | 93.6 | |
| 10 | Qwen 3 VL 4B (Thinking) | 92.9 | #480 (Qwen 3 VL 4B) |
| 11 | Qwen 2.5 VL 32B Instruct | 88.5 | #453 |
| 12 | Qwen 2.5 VL 72B Instruct | 87.1 | #377 |
| 13 | Qwen 2.5 VL 7B Instruct | 84.7 | #654 |
| 14 | Nova 2.0 Lite | 83.3 | #417 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-screenspot · How It Works · Data refreshed daily, snapshot 2026-10-20.