Wikibench

MVBench

reasoning

A comprehensive multi-modal video understanding benchmark covering 20 challenging video tasks that require temporal understanding beyond single-frame analysis. Tasks span from perception to cognition, including action recognition, temporal reasoning, spatial reasoning, object interaction, scene transition, and counterfactual inference. Uses a novel static-to-dynamic method to systematically generate video tasks from existing annotations.

Leaderboard

Showing 17 of 17 results

  1. Qwen3.5-122B-A10B
    76.6%
    i
  2. Qwen3.6-27B
    75.5%
    i
  3. Qwen3.5-35B-A3B
    74.8%
    i
  4. Qwen3.5-27B
    74.6%
    i
  5. Qwen3.6-35B-A3B
    74.6%
    i
  6. Qwen2-VL-72B-Instruct
    73.6%
    i
  7. Qwen3 VL 32B Thinking
    73.2%
    i
  8. Qwen3 VL 32B Instruct
    72.8%
    i
  9. Qwen3 VL 30B A3B Instruct
    72.3%
    i
  10. Qwen3 VL 30B A3B Thinking
    72.0%
    i
  11. Qwen2.5 VL 72B Instruct
    70.4%
    i
  12. Qwen2.5-Omni-7B
    70.3%
    i
  13. Qwen2.5 VL 7B Instruct
    69.6%
    i
  14. Qwen3 VL 4B Thinking
    69.3%
    i
  15. Qwen3 VL 8B Thinking
    69.0%
    i
  16. Qwen3 VL 4B Instruct
    68.9%
    i
  17. Qwen3 VL 8B Instruct
    68.7%
    i