Atlas

Benchmarks

← All benchmarks

WorldVQA Accuracy

Multimodal · 2026-01-28

Primary WorldVQA accuracy across the first eight categories (3,000 questions), excluding People. Correct answers count toward accuracy; incorrect and unattempted answers do not.

Top models (higher is better)

ModelScore
Gemini 3 Pro Preview47.4
Kimi K2.546.3
Gemini 2.5 Pro36.9
Opus 4.536.8
Seed1.5-VL34.9
GPT-5.228.0
GPT-5.124.5
Qwen3-VL-235B-A22B-Instruct23.5
GPT-4o22.2
Grok 4.1 Fast21.1
Sonnet 4.520.0
GLM-4.6V19.0
Grok 4 Fast18.9
Qwen3 VL 32B Instruct17.7
GLM 4.6V Flash14.8
Loading Atlas data…