List your company

Arithmetic Reasoning on Countdown (val)

76.27Validation Reward (mean@4)ACPO
4.125222.855141.58560.3149Jun 21, 2026

Evaluation results

MethodLinks
ACPOBackbone=Qwen2.5-3B-Instruct, Policy Regime=OffP.2026.0676.27
ACPOBackbone=Qwen2.5-3B-Instruct, Policy Regime=N-OnP.2026.0675.74
DAPOBackbone=Qwen2.5-3B-Instruct, Policy Regime=OffP.2026.0674.38
CISPOBackbone=Qwen2.5-3B-Instruct, Policy Regime=N-OnP.2026.0674.25
Low-EntropyBackbone=Qwen2.5-3B-Instruct, Policy Regime=OffP.2026.0674.04
High-EntropyBackbone=Qwen2.5-3B-Instruct, Policy Regime=OffP.2026.0673.76
DAPOBackbone=Qwen2.5-3B-Instruct, Policy Regime=N-OnP.2026.0673.27
High-EntropyBackbone=Qwen2.5-3B-Instruct, Policy Regime=N-OnP.2026.0670.58
AR-LoptiBackbone=Qwen2.5-3B-Instruct, Policy Regime=N-OnP.2026.0665.82
Low-EntropyBackbone=Qwen2.5-3B-Instruct, Policy Regime=N-OnP.2026.0661.81
AR-LoptiBackbone=Qwen2.5-3B-Instruct, Policy Regime=OffP.2026.0657.41
CISPOBackbone=Qwen2.5-3B-Instruct, Policy Regime=OffP.2026.0655.12
Base ModelBackbone=Qwen2.5-3B-Instruct, Policy Regime=N-OnP.2026.066.9