Mathematical Reasoning on Math tasks (Accuracy Breakdown)
81.6Easy AccuracySFT-80K (no RL)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SFT-80K (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.05 | 81.6 | 43.3 | 70.7 | |
| DPO (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.05 | 80.1 | 45.8 | 70.3 | |
| CPT + RLScale=14B2026.05 | 79.8 | 43.3 | 69.4 | |
| CPTScale=32B, Backbone=Olmo-3 32B Base, RL=No RL2026.05 | 79.1 | 44.2 | 69.1 | |
| Abs-RLScale=14B2026.05 | 78.7 | 39.4 | 67.5 | |
| SFT (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.05 | 78.4 | 45.2 | 68.9 | |
| SFT+RLScale=14B2026.05 | 78.3 | 39.6 | 67.2 | |
| DPO+RLScale=14B2026.05 | 77.2 | 42.7 | 67.3 | |
| SFT-80K+RLScale=14B2026.05 | 76.5 | 39.4 | 65.9 | |
| CPT + RLScale=8B2026.05 | 72.9 | 29.2 | 60.6 | |
| Abs-RLScale=8B2026.05 | 72.7 | 27.9 | 60 | |
| SFT-80K+RLScale=8B2026.05 | 72.4 | 28.9 | 60 | |
| SFT+RLScale=8B2026.05 | 69.8 | 29.8 | 58.4 | |
| DPO+RLScale=8B2026.05 | 69.7 | 30.2 | 58.5 | |
| CPT + RLScale=4B2026.05 | 67.5 | 26.5 | 55.8 | |
| Abs-RLScale=4B2026.05 | 67.3 | 23.1 | 54.7 | |
| SFT-80K+RLScale=4B2026.05 | 66.7 | 23 | 54.2 | |
| SFT+RLScale=4B2026.05 | 66 | 24.8 | 54.2 | |
| DPO+RLScale=4B2026.05 | 63.4 | 18.3 | 50.5 | |
| BaseScale=14B2026.05 | 60.2 | 12.9 | 46.6 | |
| BaseScale=8B2026.05 | 54 | 13.6 | 42.4 | |
| BaseScale=4B2026.05 | 43.9 | 11.7 | 41 | |
| SFT-80K+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.05 | 39.1 | 7 | 29.9 | |
| CPT + RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.05 | 29.2 | 3.3 | 21.8 | |
| SFT+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.05 | 28.4 | 4.6 | 21.6 | |
| Abs-RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.05 | 28.1 | 1 | 20.4 | |
| DPO+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.05 | 24.5 | 2.7 | 18.3 | |
| BaseScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.05 | 22.8 | 2.5 | 17 |