STEM Reasoning on GPQA Diamond
70.2AccuracySwiR
Evaluation Results
| Method | Links | |
|---|---|---|
| SwiRBackbone=Qwen3-32B2025.10 | 70.2 | |
| CoT (Greedy)Backbone=Qwen3-32B2025.10 | 69.7 | |
| Soft ThinkingBackbone=Qwen3-32B2025.10 | 67.17 | |
| CoTBackbone=Qwen3-32B2025.10 | 66.16 | |
| SwiRBackbone=Qwen3-8B2025.10 | 61.11 | |
| CoTBackbone=Qwen3-8B2025.10 | 59.6 | |
| Soft ThinkingBackbone=Qwen3-8B2025.10 | 59.6 | |
| CoT (Greedy)Backbone=Qwen3-8B2025.10 | 56.57 | |
| SwiRBackbone=DeepSeek-R1-Distill-Llama-8B2025.10 | 47.98 | |
| CoTBackbone=DeepSeek-R1-Distill-Llama-8B2025.10 | 46.46 | |
| SwiRBackbone=Qwen3-1.7B2025.10 | 41.41 | |
| CoTBackbone=Qwen3-1.7B2025.10 | 39.39 | |
| Soft ThinkingBackbone=Qwen3-1.7B2025.10 | 34.34 | |
| Soft ThinkingBackbone=DeepSeek-R1-Distill-Llama-8B2025.10 | 33.33 | |
| CoT (Greedy)Backbone=Qwen3-1.7B2025.10 | 31.82 | |
| CoT (Greedy)Backbone=DeepSeek-R1-Distill-Llama-8B2025.10 | 31.81 |