Mathematical Reasoning on AIME24 (Accuracy, Latency, Step Metrics)
56.7AccuracyLRM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LRMModel Backbone=DeepSeek-R1 (1.5B/32B)2026.04 | 56.7 | 97 | 268 | 19.5 | |
| RankGuideModel Backbone=DeepSeek-R1 (1.5B/32B)2026.04 | 56.7 | 69 | 247 | 15.68 | |
| GlimpRouterModel Backbone=DeepSeek-R1 (1.5B/32B)2026.04 | 53.3 | 86 | 325 | 19.93 | |
| SpecReasonModel Backbone=DeepSeek-R1 (1.5B/32B)2026.04 | 43.3 | 114 | 261 | 17.21 | |
| SRMModel Backbone=DeepSeek-R1 (1.5B/32B)2026.04 | 26.7 | 31 | 295 | 24.7 |