Qwen, open weights
Qwen2-VL
Qwen2.5-VL
Qwen3-VL
Qwen3.5
Qwen3.6
OpenAI API
GPT-5
GPT-5.4
GPT-5.6
GPT-6
Pareto frontier
10%
20%
30%
40%
50%
60%
200
400
600
800
1,000
Output tokens per caption (mean over 109 images)
Accuracy (judged by Phi-4)
Most attractive quadrant
fewer tokens and higher accuracy than the median model
GPT-6 Astra
GPT-5.6 Sol
GPT-5.6 Terra
GPT-6 Luna
GPT-6 Sol
Qwen2.5-VL-7B
Qwen2-VL-72B
GPT-5
GPT-5.4
GPT-5 mini
GPT-5.6 Luna
Qwen3.5-397B-A17B
Qwen3.5-27B
Qwen3-VL-235B-A22B
Qwen3.6-35B-A3B
Qwen3.5-35B-A3B
Qwen3.6-27B
Qwen3.5-122B-A10B
Qwen3-VL-32B
GPT-5.4 mini
Qwen3.5-9B
Qwen3.5-4B
Qwen3-VL-4B
Qwen3-VL-8B
Qwen3-VL-30B-A3B
Qwen3.5-2B
GPT-5 nano
GPT-5.4 nano
Qwen3-VL-2B
Qwen3.5-0.8B
Qwen2.5-VL-72B
Qwen2.5-VL-32B
Qwen2-VL-7B
Qwen2.5-VL-3B
Qwen2-VL-2B