List your company

Mathematical Reasoning In-Distribution various (test)

66.4AIME 24 ScoreTreePO
8.1623.2838.453.52Dec 6, 2025Jan 5, 2026Feb 5, 2026Mar 8, 2026Apr 8, 2026May 9, 2026Jun 9, 2026

Evaluation results

MethodLinks
TreePOModel=Qwen3-14B2026.0666.4——94.845.467.87495
PCLModel=Qwen3-14B2026.0666.2——9545.167.673.995.1
TRACEModel=Qwen3-14B2026.0666.1——94.947.371.574.994.7
GRPOModel=Qwen3-14B2026.0665.6——94.245.268.473.594.3
TreePOModel=Qwen3-8B2026.0664.3——92.640.963.170.491.3
PCLModel=Qwen3-8B2026.0664——92.840.76370.491.5
TRACEModel=Qwen3-8B2026.0663.9——93.441.265.871.191.2
GRPOModel=Qwen3-8B2026.0663.6——92.340.262.87091
ReActModel=Qwen3-14B2026.0661.5——91.84062.36989.4
ReActModel=Qwen3-8B2026.0652.3——89.837.858.665.187
Oat-ZeroApproach=RLVR2025.1233.411.961.27834.643.443.7—
DaGRPOOff-policy=true2025.1229.520.9638637.555.848.8—
SimpleRL-ZeroApproach=RLVR2025.12276.854.9762534.737.4—
SFT -> GRPOPipeline=Hybrid2025.1225.721.662.284.638.246.846.5—
DaGRPO (w/o Off-policy)Off-policy=false2025.1223.216.163.18536.751.245.9—
LUFFYApproach=Hybrid2025.122320.863.485.435.654.347.1—
GRPOMode=On-policy RL2025.1222.714.858.581.84046.744.1—
SFTMode=Fine-tuned2025.1222.222.352.882.640.843.744.1—
PRIME-ZeroApproach=RLVR2025.121712.85481.43940.340.7—
OpenReasoner-ZeroApproach=RLVR2025.1216.51552.182.433.147.141—
Qwen2.5-Math-InstructVersion=Instruct2025.1212.510.248.580.432.74137.6—
Qwen2.5-MathVersion=Base2025.1210.43.329.847.69.116.119.4—