Skip to content

ERQA

A grounded visual reasoning benchmark focused on evidence-based question answering over real images.

16 models

Higher is better.

  1. 1Qwen3.8 MaxAlibaba77.8
  2. 2Qwen3.8-Flash-NextAlibaba72.3
  3. 3Seed 2.1 ProByteDance72.0
  4. 4Qwen3.8-Omni-FlashAlibaba71.0
  5. 5Gemini 3 ProGoogle70.5
  6. 6Qwen3.7 PlusAlibaba69.8
  7. 7Gemini 3.1 ProGoogle69.4
  8. 8Qwen3.5 397BAlibaba67.5
  9. 9Qwen3.8-27BAlibaba65.5
  10. 10GPT-5.4OpenAI65.4
  11. 11Muse SparkMeta64.7
  12. 12Qwen3.6-27BAlibaba62.5
  13. 13GPT-5.2OpenAI59.8
  14. 14Grok 4.20xAI54.1
  15. 15Claude Opus 4.6Anthropic51.6
  16. 16Claude Opus 4.5Anthropic46.8
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Grounded image reasoning
Tasks
Evidence-based visual QA
Authors
Qwen (2026)

Scores are as reported by model developers or evaluators and may use different settings (for example reasoning effort or tools). Compare with care.