Mathematical Reasoning on OlympiadBench (Avg@16)
50.7Avg@16HEAPA
Evaluation Results
| Method | Links | |
|---|---|---|
| HEAPABackbone=Qwen3-8B2026.01 | 50.7 | |
| HEAPABackbone=Qwen3-4B2026.01 | 46.5 | |
| GRPO (w. PS)Backbone=Qwen3-8B2026.01 | 45.6 | |
| GRPO (w. PS)Backbone=Qwen3-4B2026.01 | 41.1 | |
| GRPOBackbone=Qwen3-8B2026.01 | 39.4 | |
| GRPOBackbone=Qwen3-4B2026.01 | 33.1 | |
| HEAPABackbone=Qwen3-1.7B2026.01 | 29.4 | |
| GRPO (w. PS)Backbone=Qwen3-1.7B2026.01 | 28.2 | |
| GRPOBackbone=Qwen3-1.7B2026.01 | 26.5 | |
| HEAPABackbone=Qwen3-0.6B2026.01 | 14.9 | |
| GRPO (w. PS)Backbone=Qwen3-0.6B2026.01 | 14.6 | |
| GRPOBackbone=Qwen3-0.6B2026.01 | 14.2 |