Vero: An Open RL Recipe for General Visual Reasoning

Benchmark Model Rank Results
natural-language-visual-grounding-on-screenspotVero-Qwen3I-8B#1Accuracy (%): 93.6
visual-question-answering-vqa-on-countbenchqaVero-Qwen3.5-9B#1Accuracy: 95.5
chart-question-answering-on-chartqaVero-Qwen3I-8B#41:1 Accuracy: 91.6
visual-question-answering-on-v-benchVero-Qwen3I-8B#4Accuracy: 89.5
visual-question-answering-vqa-on-blinkVero-Qwen3.5-9B#5Accuracy: 70.9
computer-use-agents-on-screenspot-proVero-Qwen3I-8B#16Accuracy: 61.4
multimodal-reasoning-on-mathvisionVero-Qwen3T-8B#23Accuracy: 63.5
visual-question-answering-vqa-on-realworldqaVero-Qwen3I-8B#33Accuracy: 73.3