Wikibench

Hallusion Bench

reasoning

A comprehensive benchmark designed to evaluate image-context reasoning in large visual-language models (LVLMs) by challenging models with 346 images and 1,129 carefully crafted questions to assess language hallucination and visual illusion

Leaderboard

Showing 16 of 16 results

  1. Qwen3.5-27B
    70.0%
    i
  2. Qwen3.6-35B-A3B
    69.8%
    i
  3. Qwen3.5-35B-A3B
    67.9%
    i
  4. Qwen3.5-122B-A10B
    67.6%
    i
  5. Qwen3 VL 32B Thinking
    67.4%
    i
  6. Qwen3 VL 235B A22B Thinking
    66.7%
    i
  7. Qwen3 VL 30B A3B Thinking
    66.0%
    i
  8. Qwen3 VL 8B Thinking
    65.4%
    i
  9. Qwen3 VL 4B Thinking
    64.1%
    i
  10. Qwen3 VL 32B Instruct
    63.8%
    i
  11. Qwen3 VL 235B A22B Instruct
    63.2%
    i
  12. Qwen3 VL 30B A3B Instruct
    61.5%
    i
  13. Qwen3 VL 8B Instruct
    61.1%
    i
  14. Qwen3 VL 4B Instruct
    57.6%
    i
  15. Qwen2.5 VL 72B Instruct
    55.2%
    i
  16. Qwen2.5 VL 7B Instruct
    52.9%
    i