List your company

Mathematical Reasoning on MetaMathQA (Accuracy and Length Analysis)

92.4AccuracyAdaSR-HRPO
67.85674.22880.686.972Jun 12, 2026

Evaluation results

MethodLinks
AdaSR-HRPOModel=Qwen3-4B2026.0692.4121.708115.65237.358
AdaSR-GRPOModel=Qwen3-4B2026.0690.9140.01113.545253.555
StreamingThinker (SFT)Model=Qwen3-4B2026.0686152.2121.998274.198
Read-then-thinkModel=Qwen3-1.7B2026.0684.8—1,281.7741,281.774
Read-then-thinkModel=Qwen3-4B2026.0682.8—1,379.0881,379.088
AdaSR-HRPOModel=Qwen3-1.7B2026.0682.6173.246129.356302.602
AdaSR-GRPOModel=Qwen3-1.7B2026.0682.3158.712159.285317.997
StreamingThinker (SFT)Model=Qwen3-1.7B2026.0668.8149.196121.465270.661