Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Multimodal & Grounded benchmark
Every model the catalog carries a published ScreenSpot Pro value for, ranked by that value.
A GUI-grounding benchmark for 1,581 instructions in full-screen, high-resolution professional interfaces. It tests where a target is, not whether an agent can finish the surrounding workflow. 22 models in the ModelChoice catalog have a published value.
Each point is a model with a published ScreenSpot Pro value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
22 models with a published ScreenSpot Pro value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Static interface element localization |
|---|---|---|---|
| 1 | OpenAI | 92.7 | |
| 2 | Anthropic | 87.9 | |
| 3 | OpenAI | 85.4 |
What the benchmark measures and how its value is published, from the source’s own record.
ScreenSpot Pro is a multimodal & grounded benchmark published by ScreenSpot-Pro: GUI Grounding for Professional High-Resolution Computer Use. A GUI-grounding benchmark for 1,581 instructions in full-screen, high-resolution professional interfaces. It tests where a target is, not whether an agent can finish the surrounding workflow. Its evaluation set is described by the source as 1,581 grounding instructions. The source rates its difficulty as professional GUI grounding.
Each model's result is published by ScreenSpot-Pro: GUI Grounding for Professional High-Resolution Computer Use as static interface element localization. ModelChoice prints that value to 1 decimal place, exactly as published, and never converts or rescales it. This page ranks the 22 models that carry a published value, highest value first; a model the source never scored is left out rather than ranked at zero. The source does not state whether a higher value is the better result, so neither does this page.
All leaderboards · Multimodal & Grounded capability leaderboard
| 4 | Alibaba | 84.5 |
| 5 | 84.4 |
| 6 | Meta | 84.1 |
| 7 | TMInkling 2 Small | Thinking Machines Lab | 83.8 |
| 8 | Anthropic | 83.1 |
| 9 | Alibaba | 79.0 |
| 10 | TMInkling 2 Mini | Thinking Machines Lab | 76.5 |
| 11 | Meta | 75.4 |
| 12 | 72.7 |
| 13 | HCHolo2-235B-A22B | H Company | 70.6 |
| 14 | Alibaba | 68.2 |
| 15 | HCHolo2-30B-A3B | H Company | 66.1 |
| 16 | Alibaba | 65.6 |
| 17 | Alibaba | 62.0 |
| 18 | HCHolo2-8B | H Company | 58.9 |
| 19 | NVIDIA | 57.8 |
| 20 | HCHolo2-4B | H Company | 57.2 |
| 21 | LiquidAI | 55.5 |
| 22 | Anthropic | 45.7 |
Evidence key: Observed
Rows are ordered by the value ScreenSpot-Pro: GUI Grounding for Professional High-Resolution Computer Use published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.