List your company

Reasoning on AIME 2024, GSM8k, MATH 500, and GPQA

60AIME 2024 ScoreCritique-GRPO
-2.35236813.83526630.022946.210534Jun 3, 2025Jul 13, 2025Aug 22, 2025Oct 1, 2025Nov 10, 2025Dec 20, 2025Jan 30, 2026

Evaluation results

MethodLinks
Critique-GRPOVariation=Self-Critique2025.0660—92.6—68.1352.666.29560.647.9870.036046.7
Qwen3-8B + R1-GRPOTraining=RL-GRPO2025.0650—91—63.7552.665.682.557.940.47053.3343.3
Qwen3-8BThinking=Enabled2025.0640—82—53.2341.244.167.546.935.8668.253023.3
Qwen3-8B + SFTTraining=Supervised Fine-Tuning2025.0640—83.2—56.2543.846.482.548.938.3866.8133.340
DirectRefusalSize=8B2026.010.74580.84310.93630.59410.7798————————
SafePathSize=8B2026.010.74580.84890.93850.61240.7864————————
InitialSize=8B2026.010.74170.85280.94180.50690.7608————————
THINKSAFESize=4B2026.010.73330.88060.93530.53790.7718————————
THINKSAFESize=8B2026.010.72920.880.9310.59670.785————————
STAR-1Size=8B2026.010.7250.90290.93730.57830.7859————————
SafePathSize=4B2026.010.72080.84450.93330.53540.7585————————
SafeChainSize=8B2026.010.70.92980.93530.58210.7868————————
SafeKeySize=8B2026.010.70.92440.9330.59910.7891————————
DirectRefusalSize=4B2026.010.68330.82580.9320.53030.7429————————
InitialSize=4B2026.010.6750.84690.93430.52270.7447————————
SafeKeySize=4B2026.010.67080.91790.92870.51830.7589————————
STAR-1Size=4B2026.010.6250.90970.93050.51960.7462————————
SafeChainSize=4B2026.010.62080.89590.93030.51010.7393————————
STAR-1Size=1.7B2026.010.46250.84380.8830.41160.6502————————
InitialSize=1.7B2026.010.44580.84310.88850.41730.6487————————
THINKSAFESize=1.7B2026.010.44170.8380.89050.40530.6439————————
DirectRefusalSize=1.7B2026.010.43750.82780.8810.41290.6398————————
SafePathSize=1.7B2026.010.43330.84330.88320.42420.646————————
SafeKeySize=1.7B2026.010.38330.84310.88120.40030.627————————
SafeChainSize=1.7B2026.010.34580.85290.85720.38130.6093————————
InitialSize=0.6B2026.010.10420.72510.71730.25130.4495————————
THINKSAFESize=0.6B2026.010.09580.72360.70650.2330.4397————————
SafePathSize=0.6B2026.010.07920.71260.71770.26070.4426————————
STAR-1Size=0.6B2026.010.06250.68150.68170.24180.4169————————
DirectRefusalSize=0.6B2026.010.05830.6430.67530.24810.4062————————
SafeKeySize=0.6B2026.010.05420.71580.66170.24940.4203————————
SafeChainSize=0.6B2026.010.04580.68680.62420.23740.3986————————