EvoGUI-Bench - Successor Discrimination: leaderboard
Metric: Pairwise successor accuracy (%; share of 1,000 items in which the model picks the logged next screen over a cross-trajectory, same-domain or later-step distractor; 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 September 2027. 28 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 83.6 |
| 2 | Doubao-Seed-2.0-Pro-260215 | 80.9 |
| 3 | Kimi K2.6 | 79.4 |
| 4 | Qwen 3.5 122B A10B | 77.8 |
| 5 | GLM-5V Turbo | 76.9 |
| 6 | Kimi K2.5 | 76.3 |
| 7 | MiMo-V2.5 | 76.1 |
| 8 | GPT-5.4 | 75.4 |
| 9 | Qwen 3.5 35B A3B | 74.7 |
| 10 | Qwen 3.6 35B A3B | 74.7 |
| 11 | Qwen 3.5 27B | 74.1 |
| 12 | GPT-5.4 Nano | 73.5 |
| 13 | Qwen 3.6 27B | 72.6 |
| 14 | Grok 4.20 | 72.3 |
| 15 | Qwen 3.5 2B | 71.3 |
Interactive version: theaggregate.ai/benchmark?slug=evogui-bench-successor-discrimination · How It Works · Data refreshed daily, snapshot 2026-09-29.