List your company

Math Reasoning on GAM-hard

12.3AccCoT-SFT
5.4367.218910.782Jan 29, 2026

Evaluation results

MethodLinks
CoT-SFTReasoning Type=CoT, Training=SFT2026.0112.335.6
ATP-LatentReasoning Type=Latent, Training=RL2026.019.810
ATP-Latent w Only Coherence as RewardReasoning Type=Latent, Training=RL, Reward Mechanism=Unsupervised Coherence2026.019.59.7
SIM-CoT*Reasoning Type=Latent, Source=Reported in Wei et al. (2025)2026.019.3—
ATP-Latent -w/o RL (Ours-SFT)Reasoning Type=Latent, Training=SFT2026.019.39.7
ATP-Latent -w/o VAEReasoning Type=Latent, Training=RL, VAE=Excluded2026.019.210.1
ATP-Latent -w/o Stop HeadReasoning Type=Latent, Training=RL, Stop Head=Excluded2026.018.89.8
SIM-CoTReasoning Type=Latent2026.018.69.8
CoconutReasoning Type=Latent2026.018.29.8
Coconut*Reasoning Type=Latent, Source=Reported in Wei et al. (2025)2026.018.1—
Answer-SFTReasoning Type=Answer-only, Training=SFT2026.016.44.8
CoLaRReasoning Type=Latent2026.015.711.5
iCoT*Reasoning Type=Latent, Source=Reported in Wei et al. (2025)2026.015.7—