List your company

Agentic Action Prediction on SCIWorld (in domain)

61Success Rateours
45.449.4553.557.55Jun 25, 2026

Evaluation results

MethodLinks
oursModel=Qwen3-8B2026.0661
GRPOModel=Qwen3-8B2026.0657
oursModel=Qwen3-4B2026.0649
GRPOModel=Qwen3-4B2026.0646