Physics Reasoning on Synthetic Numeric
21.9AccuracyQwen2.5-32B + RL (synthetic)
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-32B + RL (synthetic)Model=Qwen2.5-32B, Post-training Stage=RL (synthetic)2026.04 | 21.9 | |
| Qwen3-30B + RL (synthetic)Model=Qwen3-30B, Post-training Stage=RL (synthetic)2026.04 | 17.4 | |
| Qwen2.5-14B + RL (synthetic)Model=Qwen2.5-14B, Post-training Stage=RL (synthetic)2026.04 | 17 | |
| Qwen2.5-7B + RL (synthetic)Model=Qwen2.5-7B, Post-training Stage=RL (synthetic)2026.04 | 16.3 | |
| Qwen3-30BModel=Qwen3-30B, Post-training Stage=Instruct2026.04 | 14.8 | |
| Qwen2.5-3B + RL (synthetic)Model=Qwen2.5-3B, Post-training Stage=RL (synthetic)2026.04 | 12.5 | |
| Qwen2.5-32BModel=Qwen2.5-32B, Post-training Stage=Instruct2026.04 | 8.9 | |
| Qwen2.5-7BModel=Qwen2.5-7B, Post-training Stage=Instruct2026.04 | 7.7 | |
| Qwen2.5-14BModel=Qwen2.5-14B, Post-training Stage=Instruct2026.04 | 7 | |
| Qwen2.5-3BModel=Qwen2.5-3B, Post-training Stage=Instruct2026.04 | 4.8 |