List your company

Mathematical Reasoning on AMC 2023 (Accuracy (%))

94.7Accuracy (AMC 2023)Token Zones + KL Annealing
19.851239.283158.71578.1469May 25, 2026May 30, 2026Jun 4, 2026Jun 10, 2026Jun 15, 2026Jun 20, 2026Jun 26, 2026

Evaluation results

MethodLinks
Token Zones + KL AnnealingSEAD Components=Token Zones + KL Annealing, Category=Proposed2026.0694.7
GRPOCategory=Baselines2026.0694.6
KL Annealing onlySEAD Components=KL Annealing only, Category=Proposed2026.0694.3
Nemotron-49B (Teacher)Model Role=Teacher, Category=Off-the-shelf Models2026.0694.2
OPD (RKL)Category=Baselines2026.0693.9
Token Zones onlySEAD Components=Token Zones only, Category=Proposed2026.0693.8
OPSDCategory=Baselines2026.0693.4
Nemotron-8B (Student)Model Role=Student, Category=Off-the-shelf Models2026.0693.3
GTModel=Qwen3-8B-Base2026.0570.31
OracleModel=Qwen3-8B-Base2026.0566.41
CAREModel=Qwen3-8B-Base2026.0565.23
EntropyModel=Qwen3-8B-Base2026.0561.09
CAREModel=Qwen3-4B-Base2026.0561.02
GTModel=Qwen3-4B-Base2026.0560.62
ProbModel=Qwen3-8B-Base2026.0560.47
TTRLModel=Qwen3-8B-Base2026.0560.16
OracleModel=Qwen3-4B-Base2026.0560.08
RandomModel=Qwen3-8B-Base2026.0558.91
RandomModel=Qwen3-4B-Base2026.0555.23
ProbModel=Qwen3-4B-Base2026.0554.45
TTRLModel=Qwen3-4B-Base2026.0553.52
EntropyModel=Qwen3-4B-Base2026.0551.09
GTModel=Qwen3-1.7B-Base2026.0547.73
VanillaModel=Qwen3-8B-Base2026.0543.13
CAREModel=Qwen3-1.7B-Base2026.0540.08
TTRLModel=Qwen3-1.7B-Base2026.0536.33
OracleModel=Qwen3-1.7B-Base2026.0536.02
RandomModel=Qwen3-1.7B-Base2026.0535.7
EntropyModel=Qwen3-1.7B-Base2026.0535.47
ProbModel=Qwen3-1.7B-Base2026.0533.98
VanillaModel=Qwen3-1.7B-Base2026.0525.62
VanillaModel=Qwen3-4B-Base2026.0522.73