List your company

Arithmetic Reasoning on Countdown

92Pass@1Countdown (A) + GRPO
36.8851.1965.579.81Apr 13, 2026

Evaluation results

MethodLinks
Countdown (A) + GRPOBackbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=LLaMA-70B teacher answers (A), Optimization=GRPO2026.049298
GRPO (Countdown)Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Optimization=GRPO2026.049198
Countdown (A') + GRPOBackbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=Self-revised answers (A'), Optimization=GRPO2026.049198
BaseBackbone=Qwen2.5-7B, Generation budget=4096 tokens2026.044190
Countdown (A')Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=Self-revised answers (A')2026.044088
Countdown (A)Backbone=Qwen2.5-7B, Generation budget=4096 tokens, Training Data=LLaMA-70B teacher answers (A)2026.043991