ERQA

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

In index
Multimodal
Best in multimodal

Models18

Top result77.8%Qwen3.8 Max

Top-3 spread5.8 pts

Year—

Result and price

13 of 18 results

Results

18 results

#ModelResult

01Qwen3.8 Max77.8%

02Qwen3.8-Flash-Next72.3%

03Seed 2.1 Pro72.0%

04Seed 2.1 Turbo71.3%

05Qwen3.8-Omni-Flash71.0%

06Gemini 3 Pro70.5%

07Qwen3.7 Plus69.8%

08Gemini 3.1 Pro69.4%

09Qwen3.5 397B67.5%

10Qwen3.8-27B65.5%

11GPT-5.465.4%

12Muse Spark64.7%

13Qwen3.6-27B62.5%

14GPT-5.259.8%

15Grok 4.2054.1%

16Claude Opus 4.651.6%

17LFM3-VL-2B48.2%

18Claude Opus 4.546.8%