List your company

Mathematical Problem Solving on OlympiadBench (test)

42.7Pass@1 AccuracyPRM-CoT (Process-Aware)
14.089621.517328.94536.3727Dec 2, 2025Jan 5, 2026Feb 9, 2026Mar 16, 2026Apr 20, 2026May 25, 2026Jun 29, 2026

Evaluation results

MethodLinks
PRM-CoT (Process-Aware)Base Policy=Qwen2.5-Math-7B2025.1242.738.4252.6756.67——————
RLVR (Ground Truth)Base Policy=Qwen2.5-Math-7B2025.124036.0848.6754——————
GRPOBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.0639.658.356.560.545.951.563.866.669.171.5
DIVER-EDModel=Qwen2.5-7B2026.0639.42—————71.11———
PRM (Process-Aware)Base Policy=Qwen2.5-Math-7B2025.1238.735.3348.6752——————
DIVER-TDModel=Qwen2.5-7B2026.0638.36—————72.19———
REINFORCEBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.0638.255.853.757.643.84960.963.866.368.7
W-RFBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.0638.259.45761.74551.365.6697274.6
BBGBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.0637.161.358.864.145.152.468.37275.778.5
GRPOModel=Qwen2.5-7B2026.0637.08—————71.7———
SFT (Baseline)Base Policy=Qwen2.5-Math-7B2025.1234.731.8343.3348.33——————
DQOModel=Qwen2.5-7B2026.0631.86—————66.2———
DIVER-EDModel=Qwen2.5-3B2026.0629.31—————66.67———
Base ModelBase Model=Qwen2.5-Math-7B, Temperature=0.6, top-p=0.95, Max tokens=4096, Scoring=exact any-boxed answer scoring2026.0628.55855.261.438.547.566.57175.178.5
DIVER-TDModel=Qwen2.5-3B2026.0628.22—————65.93———
GRPOModel=Qwen2.5-3B2026.0627.81—————64.44———
DQOModel=Qwen2.5-3B2026.0626.48—————63.85———
BaseModel=Qwen2.5-7B2026.0619.1—————66.52———
BaseModel=Qwen2.5-3B2026.0615.19—————60.3———