Reasoning on ARC Challenge (accuracy, length)
418Solution Length+ShorterBetter
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| +ShorterBetterBackbone=Qwen3-4B2026.05 | 418 | 0.9297 | |
| +LP-G (Kimi)Backbone=Qwen3-4B2026.05 | 464 | 0.923 | |
| +LC-R1Backbone=Qwen3-4B2026.05 | 505 | 0.9163 | |
| +ICR /w LP-FBackbone=Qwen3-8B2026.05 | 602 | 0.9397 | |
| +ICR /w LP-FBackbone=DSQW-7B2026.05 | 603 | 0.8428 | |
| +ICR /w LP-FBackbone=Qwen3-4B2026.05 | 608 | 0.9397 | |
| +ShorterBetterBackbone=Qwen3-8B2026.05 | 614 | 0.9197 | |
| +ICRBackbone=Qwen3-4B2026.05 | 622 | 0.9364 | |
| +LP-G (Kimi)Backbone=Qwen3-8B2026.05 | 646 | 0.9264 | |
| +LP-FBackbone=Qwen3-8B2026.05 | 699 | 0.9364 | |
| +LC-R1Backbone=Qwen3-8B2026.05 | 701 | 0.9264 | |
| +ShorterBetterBackbone=DSQW-7B2026.05 | 707 | 0.8127 | |
| +LP-FBackbone=Qwen3-4B2026.05 | 732 | 0.9364 | |
| +LP-FBackbone=DSQW-7B2026.05 | 792 | 0.826 | |
| +ICRBackbone=Qwen3-8B2026.05 | 804 | 0.9331 | |
| +LP-G (Kimi)Backbone=DSQW-7B2026.05 | 855 | 0.796 | |
| +LC-R1Backbone=DSQW-7B2026.05 | 878 | 0.8327 | |
| +GRPOBackbone=Qwen3-4B2026.05 | 886 | 0.9257 | |
| +GRPOBackbone=Qwen3-8B2026.05 | 981 | 0.9257 | |
| +ICRBackbone=DSQW-7B2026.05 | 1,177 | 0.8261 | |
| BaseBackbone=Qwen3-4B2026.05 | 1,231 | 0.9331 | |
| BaseBackbone=DSQW-7B2026.05 | 1,321 | 0.786 | |
| +GRPOBackbone=DSQW-7B2026.05 | 1,386 | 0.816 | |
| BaseBackbone=Qwen3-8B2026.05 | 1,398 | 0.9264 |