| Benchmark | Model | Rank | Results |
|---|---|---|---|
| code-generation-on-humaneval | Llama 3.2–3B–Instruct (mid-trained, n=64, RL) | – | Pass@1: 52.82 |
| mathematical-reasoning-on-aime-2025 | Llama 3.2–3B–Instruct (mid-trained, n=64) | – | Accuracy: 0.44 |
| mathematical-reasoning-on-aime24 | Llama 3.2–3B–Instruct (mid-trained, n=32) | – | Acc: 2.64 |
| mathematical-reasoning-on-math-500 | Llama 3.2–3B–Instruct (mid-trained, n=64) | – | Accuracy: 38.11 |
| mathematical-reasoning-on-olympiadbench | Llama 3.2–3B–Instruct (mid-trained, n=64) | – | Accuracy: 8.66 |