Scientific Reasoning on GPQA Diamond (Accuracy MV, Accuracy WMV, Required Tokens)
72.7Accuracy (MV)DeepConf
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DeepConfBackbone=Phi-4-RP-14B, branching factor n=322026.06 | 72.7 | 72.3 | 49.3 | |
| Flex-Fork-thinkBackbone=Phi-4-RP-14B, branching factor n=32, seed path length l0=2,0482026.06 | 70.7 | 63.2 | 37.8 | |
| DeepConfBackbone=DeepSeek-8B, branching factor n=322026.06 | 67.3 | 66.3 | 73.8 | |
| ASCBackbone=DeepSeek-8B, branching factor n=322026.06 | 67.1 | — | 42.1 | |
| Flex-Fork-thinkBackbone=Qwen3-8B, branching factor n=32, seed path length l0=2,0482026.06 | 63.1 | 63.2 | 35 | |
| DeepConfBackbone=Qwen3-8B, branching factor n=322026.06 | 62.8 | 62.3 | 64.2 | |
| Flex-Fork-thinkBackbone=DeepSeek-8B, branching factor n=32, seed path length l0=2,0482026.06 | 62.1 | 63.1 | 34.5 | |
| ASCBackbone=Qwen3-8B, branching factor n=322026.06 | 61.7 | — | 58.5 | |
| ASCBackbone=Phi-4-RP-14B, branching factor n=322026.06 | 58.1 | — | 41.8 |