Knowledge-Orthogonal Reasoning on KOR-Bench
56.56Overall Score (%)Base Model
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Base ModelArchitecture=Qwen3-30B-A3B2026.05 | 56.56 | 1,173 | 1,163 | 1,163 | |
| Base ModelArchitecture=Qwen3-4B2026.05 | 52.64 | 1,186 | 1,182 | 1,182 | |
| SxS InterleavedArchitecture=Qwen3-4B, Stage=RL Final2026.05 | 32.96 | 1,438 | 1,491 | 1,321 | |
| Standard CoTArchitecture=Qwen3-30B-A3B, Stage=RL Final2026.05 | 32.08 | 1,468 | 1,429 | 1,429 | |
| SxS InterleavedArchitecture=Qwen3-30B-A3B, Stage=RL Final2026.05 | 31.76 | 1,352 | 1,382 | 1,304 | |
| SxS InterleavedArchitecture=Qwen3-4B, Stage=RL Recovery2026.05 | 27.76 | 1,225 | 1,324 | 1,193 | |
| SxS InterleavedArchitecture=Qwen3-30B-A3B, Stage=RL Recovery2026.05 | 23.44 | 1,178 | 1,293 | 1,169 | |
| Standard CoTArchitecture=Qwen3-30B-A3B, Stage=SFT2026.05 | 22.56 | 1,580 | 1,533 | 1,533 | |
| SxS InterleavedArchitecture=Qwen3-4B, Stage=SFT2026.05 | 22.16 | 942 | 1,151 | 960 | |
| Standard CoTArchitecture=Qwen3-4B, Stage=SFT2026.05 | 21.68 | 1,598 | 1,556 | 1,556 | |
| SxS InterleavedArchitecture=Qwen3-30B-A3B, Stage=SFT2026.05 | 19.6 | 910 | 1,117 | 885 | |
| Standard CoTArchitecture=Qwen3-4B, Stage=RL Final2026.05 | 19.52 | 1,715 | 1,681 | 1,681 |