Multiple-choice Reasoning on GPQA (test)
65.7AccuracyRouteGoT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RouteGoTBackbone={Qwen3-4B, 8B, 30B}2026.03 | 65.7 | 3,352 | |
| AGoTBackbone=Qwen3-30B2026.03 | 64.6 | 12,179 | |
| CoTBackbone=Qwen3-30B2026.03 | 63.1 | 4,965 | |
| KNNBackbone={Qwen3-4B, 8B, 30B}2026.03 | 61.1 | 3,292 | |
| RandomBackbone={Qwen3-4B, 8B, 30B}2026.03 | 60.1 | 5,658 | |
| GoT*Backbone=Qwen3-30B2026.03 | 59.6 | 9,468 | |
| EmbedLLMBackbone={Qwen3-4B, 8B, 30B}2026.03 | 59.6 | 11,369 | |
| RouteLLMBackbone={Qwen3-4B, 8B, 30B}2026.03 | 57.6 | 3,640 | |
| RTRBackbone={Qwen3-4B, 8B, 30B}2026.03 | 56.6 | 3,224 | |
| ToTBackbone=Qwen3-30B2026.03 | 44.9 | 9,077 | |
| IOBackbone=Qwen3-30B2026.03 | 41.4 | 7 |