Mathematical Reasoning on OlympiadBench (Accuracy, Think ↓, Route%)
71.5AccuracyAT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ATBackbone=Qwen3-8B2026.06 | 71.5 | — | — | |
| SC-RouteBackbone=Qwen3-8B2026.06 | 70.4 | — | — | |
| DARTBackbone=Qwen3-8B2026.06 | 69.8 | 45 | 52.1 | |
| DARTBackbone=DeepSeek-V3.2, Hosted API=true2026.06 | 69.1 | 32 | — | |
| SC-RouteBackbone=DeepSeek-V3.2, Hosted API=true2026.06 | 67.1 | — | — | |
| ATBackbone=DeepSeek-V3.2, Hosted API=true2026.06 | 66.1 | — | — | |
| NTBackbone=DeepSeek-V3.2, Hosted API=true2026.06 | 63.6 | — | — | |
| SC-RouteBackbone=Qwen3-14B2026.06 | 62 | — | — | |
| DARTBackbone=Qwen3-14B2026.06 | 62 | 15 | 58 | |
| SC-RouteBackbone=Qwen3-32B2026.06 | 58.5 | — | — | |
| DARTBackbone=Qwen3-32B2026.06 | 58.5 | 16 | 59.5 | |
| ATBackbone=Qwen3-32B2026.06 | 54 | — | — | |
| ATBackbone=Qwen3-14B2026.06 | 53 | — | — | |
| NTBackbone=Qwen3-14B2026.06 | 51.5 | — | — | |
| NTBackbone=Qwen3-32B2026.06 | 50.5 | — | — | |
| NTBackbone=Qwen3-8B2026.06 | 49.8 | — | — |