Wikibench

LVBench

multimodal

LVBench is an extreme long video understanding benchmark designed to evaluate multimodal models on videos up to two hours in duration. It contains 6 major categories and 21 subcategories, with videos averaging five times longer than existing datasets. The benchmark addresses applications requiring comprehension of extremely long videos.

Leaderboard

Showing 20 of 20 results

  1. Kimi K2.5
    75.9%
    i
  2. Qwen3.5-122B-A10B
    74.4%
    i
  3. Qwen3.5-27B
    73.6%
    i
  4. Qwen3.5-35B-A3B
    71.4%
    i
  5. Qwen3.6-35B-A3B
    71.4%
    i
  6. Qwen3 VL 235B A22B Instruct
    67.7%
    i
  7. Qwen3 VL 32B Instruct
    63.8%
    i
  8. Qwen3 VL 235B A22B Thinking
    63.6%
    i
  9. Qwen3 VL 32B Thinking
    62.6%
    i
  10. Qwen3 VL 30B A3B Instruct
    62.5%
    i
  11. Qwen3 VL 30B A3B Thinking
    59.2%
    i
  12. Qwen3 VL 8B Instruct
    58.0%
    i
  13. Qwen3 VL 4B Instruct
    56.2%
    i
  14. Qwen3 VL 8B Thinking
    55.8%
    i
  15. Qwen3 VL 4B Thinking
    53.5%
    i
  16. Qwen2.5 VL 32B Instruct
    49.0%
    i
  17. Qwen2.5 VL 72B Instruct
    47.3%
    i
  18. Qwen2.5 VL 7B Instruct
    45.3%
    i
  19. Nova Pro
    41.6%
    i
  20. Nova Lite
    40.4%
    i