List your company

Text-based Game Playing on TW-Cooking 1000 (test)

83.5Pass@1REUSERL-SegCost (no buffer)
3.73224.44145.1565.859May 29, 2026

Evaluation results

MethodLinks
REUSERL-SegCost (no buffer)Buffer=None2026.0583.5
REUSERL-SegCostBuffer=Global success buffer2026.0581.73
Vanilla GRPO2026.0574.97
Pure Round-Length2026.0564.03
Vanilla2026.056.8