ScreenSpot-V2 — Same as '(Acc)' variant — ScreenSpot-V2 reports one accuracy metric, '(Acc)' is a redundant label.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen3 VL Thinking (8B) | Alibaba | 93.6 | 2 | 2026-06-05 |
| 2 | Kimi VL A3B Instruct | Moonshot | 92.8 | 2 | 2026-06-12 |
| 3 | Step3 VL 10B | StepFun | 92.6 | 2 | 2026-06-05 |
| 4 | GLM-4.6V-Flash (9B) | Z.ai | 92.1 | 2 | 2026-06-05 |
| 5 | Kimi VL A3B Thinking | Moonshot | 91.4 | 1 | 2026-06-05 |
| 6 | MiMo VL RL 2508 (7B) | Xiaomi | 90.8 | 2 | 2026-06-05 |
| 7 | InternVL-3.5 (8B) | OpenGVLab | 84 | 2 | 2026-06-05 |
| 8 | GPT-4o | OpenAI | 18.1 | 2 | 2026-06-05 |
| 9 | GPT-4o mini | OpenAI | 6.9 | 1 | 2026-06-05 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.