List your company

Mathematical Reasoning on AMC23 (avg@16 accuracy)

84.38Avg@16 AccuracyOISD
16.561634.168351.77569.3817Jan 19, 2026Feb 9, 2026Mar 2, 2026Mar 24, 2026Apr 14, 2026May 5, 2026May 27, 2026

Evaluation results

MethodLinks
OISDBackbone=Qwen3-4B2026.0584.38
Qwen3-8B-Base + PASC-GRPOmax_completion_tokens=8K2026.0182.5
Gemma-3-27B-ITmax_completion_tokens=8K2026.0180.5
Gemma-3-12B-ITmax_completion_tokens=8K2026.0177.3
OPSDBackbone=Qwen3-4B2026.0577.19
GRPOBackbone=Qwen3-4B2026.0576.88
Qwen3-8B-Base + GRP-SFTmax_completion_tokens=8K2026.0175
DyJRBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, alpha_JS=0.05, M=82026.0372
Qwen3-4B-Base + PASC-GRPOmax_completion_tokens=8K2026.0171.43
DyJR (Forward KL)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, Divergence Type=Forward KL2026.0370.7
Ex-GRPOBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.0370.2
DyJR (alpha_JS = 0.01)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, alpha_JS=0.012026.0369.3
DyJR (M = 16)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, M=162026.0369.3
RLEPBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.0368.9
DyJR (M = 32)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, M=322026.0368.9
DyJR (M = 64)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, M=642026.0367.9
DyJR (alpha_JS = 0.2)Backbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@256, alpha_JS=0.22026.0366.7
DAPOBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.0365.5
DPH-RLBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.0365.2
GRPOBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.0364.6
Qwen3-4B-Base + GRP-SFTmax_completion_tokens=8K2026.0163.33
Base ModelBackbone=Qwen3-4B-Base, Evaluation Protocol (Mean@k)=Mean@2562026.0351.5
LLaMA-3.1-8B-Instructmax_completion_tokens=8K2026.0141.2
Qwen3-8B-Basemax_completion_tokens=8K2026.0138.12
Qwen3-4B-Basemax_completion_tokens=8K2026.0119.17