Question Answering on SuperGPQA (test)
40.2AccuracySFT
Evaluation Results
| Method | Links | |
|---|---|---|
| SFTBackbone=Qwen32026.01 | 40.2 | |
| 5-shotBackbone=Qwen32026.01 | 35.56 | |
| kNN-MoEBackbone=Qwen32026.01 | 35.15 | |
| Zero-shotBackbone=Qwen32026.01 | 34.85 | |
| SFT (Router Only)Backbone=Qwen32026.01 | 33.15 | |
| SFTBackbone=GPT-OSS2026.01 | 24.83 | |
| kNN-MoEBackbone=GPT-OSS2026.01 | 24.35 | |
| Zero-shotBackbone=GPT-OSS2026.01 | 23.52 | |
| 5-shotBackbone=GPT-OSS2026.01 | 19.95 | |
| SFT (Router Only)Backbone=GPT-OSS2026.01 | 18.83 | |
| SFTBackbone=OLMoE2026.01 | 13.67 | |
| kNN-MoEBackbone=OLMoE2026.01 | 13.27 | |
| Zero-shotBackbone=OLMoE2026.01 | 13.02 | |
| SFT (Router Only)Backbone=OLMoE2026.01 | 11.66 | |
| 5-shotBackbone=OLMoE2026.01 | 11.09 |