Skip to content

RealWorldQA

A grounded visual QA benchmark focused on answering practical questions about real-world images and scenes.

16 models

Higher is better.

  1. 1Qwen3.8-Flash-NextAlibaba88.5
  2. 2Qwen3.8 MaxAlibaba88.0
  3. 3Qwen3.8-Omni-FlashAlibaba87.7
  4. 4Qwen3.7 PlusAlibaba86.9
  5. 5Qwen3.8-27BAlibaba85.9
  6. 6Qwen3.6-35B-A3BAlibaba85.3
  7. 7Qwen3.6-27BAlibaba84.1
  8. 8Qwen3.5 397BAlibaba83.9
  9. 9Gemini 3 ProGoogle83.3
  10. 10GPT-5.2OpenAI83.3
  11. 11Kimi K2.5Moonshot AI81.0
  12. 12Ternary Bonsai 2 27BPrism ML80.1
  13. 13Claude Opus 4.5Anthropic77.0
  14. 14LFM2.5-VL-3BLiquid AI73.1
  15. 15North Micro Vision InstructCohere62.2
  16. 16LFM2.5-VL-450MLiquid AI58.4
Higher is better. Bar colour: blue proprietary, orange open weights, green other.

About this benchmark

Metric
Image-grounded QA
Tasks
Real-world visual question answering
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.