List your company

Real World Visual Reasoning on MEGA

54.7AccuracyGemini 2.0 Flash
30.88437.06743.2549.433Feb 12, 2026

Evaluation results

MethodLinks
Gemini 2.0 FlashZero-shot=true2026.0254.7
Qwen2.5-VL-32B + AT-RL (Ours)Zero-shot=true2026.0253
OpenAI GPT-4oZero-shot=true2026.0252.7
Claude 3.5 SonnetZero-shot=true2026.0252.5
Qwen2.5-VL-32B + VPPOZero-shot=true2026.0252.3
Qwen2.5-VL-72B InstructZero-shot=true2026.0249.6
Qwen2.5-VL-32B InstructZero-shot=true2026.0248.9
Llama 4 ScoutZero-shot=true2026.0231.8