List your company

Mathematical Reasoning on Average Across all Benchmarks

0.8Delta AccuracyDPOshortest
-3.256-2.203-1.15-0.097Feb 1, 2026Feb 8, 2026Feb 16, 2026Feb 23, 2026Mar 3, 2026Mar 10, 2026Mar 18, 2026

Evaluation results

MethodLinks
DPOshortestBackbone=R1-Distill-Qwen-1.5B2026.020.8-32.2—
RFTshortestBackbone=R1-Distill-Qwen-7B2026.020.60.5—
RFTshortestBackbone=Qwen3-4B-Thinking2026.020.2-5.9—
ConPressBackbone=R1-Distill-Qwen-7B2026.020-33.9—
RFTshortestBackbone=R1-Distill-Qwen-1.5B2026.02-0.10.1—
ConPressBackbone=R1-Distill-Qwen-1.5B2026.02-0.2-40.2—
ThinkPruneBackbone=R1-Distill-Qwen-1.5B2026.02-0.4-45.9—
ConPressBackbone=Qwen3-4B-Thinking2026.02-0.6-48.7—
TRiMSBackbone=Qwen3-4B2026.03-0.63-88.370.6
AdaptThinkBackbone=R1-Distill-Qwen-1.5B2026.02-0.7-55.8—
DPOshortestBackbone=Qwen3-4B-Thinking2026.02-0.9-13.1—
AdaptThinkBackbone=R1-Distill-Qwen-7B2026.02-1.2-36.3—
LC-R1Backbone=R1-Distill-Qwen-1.5B2026.02-1.7-53.1—
DPOshortestBackbone=R1-Distill-Qwen-7B2026.02-2.3-31.4—
LC-R1Backbone=R1-Distill-Qwen-7B2026.02-3.1-45.4—
VanillaBackbone=Qwen3-4B2026.03——8.8