EvoGUI-Bench - Inverse Action and Value: leaderboard
The inverse probe of EvoGUI-Bench (Tsinghua University, 2026): from two consecutive web screenshots, predict the click, type or select action that caused the change and its typed or selected value. Joint accuracy in percent; higher is better.
Metric: Joint action and value accuracy (%; share of 1,000 consecutive screenshot pairs for which the model predicts both the click, type or select action and its typed or selected value; on EvoGUI-Bench, 3,000 zero-shot visual question answering probes mined from Mind2Web and WebLINX web trajectories across 120 domains, screenshots at max_image_side 512, temperature 0; higher is better). Source: arxiv.org. Saturation forecast: Around 2030. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 65.4 | #144 |
| 2 | Qwen 3.5 397B A17B | 56.6 | #246 |
| 3 | Qwen 3.6 35B A3B | 56.6 | #462 |
| 4 | Qwen 3.5 122B A10B | 56.4 | #331 |
| 5 | Doubao-Seed-2.0-Pro-260215 | 55.7 | #152 |
| 6 | Qwen 3.6 27B | 55.1 | #341 |
| 7 | Qwen 3 VL 235B A22B Instruct | 55.1 | #522 |
| 8 | Qwen 3.5 27B | 54.8 | #359 |
| 9 | Qwen 3.5 35B A3B | 54.8 | #482 |
| 10 | MiMo-V2.5 | 54.5 | #354 |
| 11 | GLM-5V Turbo | 54 | #344 |
| 12 | Kimi K2.5 | 52.8 | #256 |
| 13 | Kimi K2.6 | 52.6 | #178 |
| 14 | GPT-5 Mini | 52.5 | #348 |
| 15 | GLM-4.5V | 52.3 | #744 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=evogui-bench-inverse-action-and-value · How It Works · Data refreshed daily, snapshot 2026-11-13.