Wikibench

ScreenSpot Pro

multimodal

ScreenSpot-Pro is a novel GUI grounding benchmark designed to rigorously evaluate the grounding capabilities of multimodal large language models (MLLMs) in professional high-resolution computing environments. The benchmark comprises 1,581 instructions across 23 applications spanning 5 industries and 3 operating systems, featuring authentic high-resolution images from professional domains with expert annotations. Unlike previous benchmarks that focus on cropped screenshots in consumer applications, ScreenSpot-Pro addresses the complexity and diversity of real-world professional software scenarios, revealing significant performance gaps in current MLLM GUI perception capabilities.

Leaderboard

Showing 20 of 22 results

  1. Claude Opus 4.8
    87.9%
    i
  2. GPT-5.2
    86.3%
    i
  3. Muse Spark
    84.1%
    i
  4. Gemini 3 Pro
    72.7%
    i
  5. Qwen3.5-122B-A10B
    70.4%
    i
  6. Qwen3.5-27B
    70.3%
    i
  7. Gemini 3 Flash
    69.1%
    i
  8. Qwen3.5-35B-A3B
    68.6%
    i
  9. Qwen3.6 Plus
    68.2%
    i
  10. Qwen3 VL 235B A22B Instruct
    62.0%
    i
  11. Qwen3 VL 235B A22B Thinking
    61.8%
    i
  12. Qwen3 VL 30B A3B Instruct
    60.5%
    i
  13. Qwen3 VL 4B Instruct
    59.5%
    i
  14. Qwen3 VL 32B Instruct
    57.9%
    i
  15. Qwen3 VL 30B A3B Thinking
    57.3%
    i
  16. Qwen3 VL 32B Thinking
    57.1%
    i
  17. Qwen3 VL 8B Instruct
    54.6%
    i
  18. Qwen3 VL 4B Thinking
    49.2%
    i
  19. Qwen3 VL 8B Thinking
    46.6%
    i
  20. Qwen2.5 VL 72B Instruct
    43.6%
    i