Mathematical Reasoning on AIME 2025 (Pass@1, Pass@K)
9.2Pass@1GRPO + HORA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRPO + HORAScale=7B2026.05 | 9.2 | 73.3 | |
| GRPOScale=7B2026.05 | 8.3 | 66.7 | |
| BaseScale=7B2026.05 | 2.6 | 70 | |
| GRPOScale=3B2026.05 | 2.4 | 63.3 | |
| GRPO + HORAScale=3B2026.05 | 2 | 70 | |
| GRPOScale=1.5B-Inst.2026.05 | 1 | 56.7 | |
| GRPO + HORAScale=1.5B-Inst.2026.05 | 1 | 60 | |
| BaseScale=1.5B-Inst.2026.05 | 0.9 | 63.3 | |
| BaseScale=3B2026.05 | 0.7 | 56.7 |