Mathematical Reasoning on Aggregate AIME24, AMC, MATH500, Minerva, Olympiad bench
22.7Accuracy Delta (%)DLER-R1-1.5B-Research
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DLER-R1-1.5B-ResearchBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 22.7 | 51.7 | 1.2 | |
| DLER-R1-7B-ResearchBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 17 | 55.5 | 1.06 | |
| APR-1.5B (Ours)Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 16.3 | 52.8 | 1.02 | |
| Laser-L8192-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 15.7 | 20.4 | 0.68 | |
| L1-Qwen-1.5B-MaxBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 14.9 | 53.1 | 0.98 | |
| Laser-DE-L4096-7BBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 14.5 | 43.6 | 0.87 | |
| APR-7B (Ours)Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 11.5 | 56.7 | 0.91 | |
| DS-1.5B-thinkprune-iter2kBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 11.3 | 43.3 | 0.77 | |
| Laser-DE-L4096-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 10.7 | 38.2 | 0.7 | |
| L1-Qwen-7B-MaxBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 10 | 43.8 | 0.74 | |
| TrainingEfficient DS-7BBackbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 7.2 | 18.1 | 0.4 | |
| AdaptThink-7B-delta0.05Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 6.7 | 21.8 | 0.42 | |
| TrainingEfficient DS-1.5BBackbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 6.4 | 21.8 | 0.41 | |
| AdaptThink-1.5B-delta0.05Backbone=DeepSeek-R1-Distill-Qwen-1.5B2026.01 | 4.4 | 36.9 | 0.5 | |
| SB DS7B alpha 2Backbone=DeepSeek-R1-Distill-Qwen-7B2026.01 | 2.7 | 57.3 | 0.65 |