List your company

VLM Agent Success Rate Estimation on Sokoban 6x6

80.5Success RateSTAPO
8.94827.52446.164.676Jul 6, 2026

Evaluation results

MethodLinks
STAPOBackbone=Qwen2.5-VL-3B-Instruct2026.0780.5
GiGPOBackbone=Qwen2.5-VL-3B-Instruct2026.0776.9
GRPOBackbone=Qwen2.5-VL-3B-Instruct2026.0767.1
BaseBackbone=Qwen2.5-VL-3B-Instruct2026.0711.7