Logical Reasoning on LogicBench (with Skill Token Tracking)
85.26AccuracyFull Text
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Full TextBackbone Model=Qwen3-8B2026.06 | 85.26 | 1,089 | |
| LLMLingua-2 (L)Backbone Model=Qwen3-8B2026.06 | 85.13 | 816 | |
| SKIM (Fix-512)Backbone Model=Qwen3-8B2026.06 | 85 | 512 | |
| SKIM (Adaptive)Backbone Model=Qwen3-8B2026.06 | 83.68 | 438 | |
| LLMLingua-2 (M)Backbone Model=Qwen3-8B2026.06 | 83.55 | 604 | |
| SKIM (Fix-256)Backbone Model=Qwen3-8B2026.06 | 83.16 | 256 | |
| LLMLingua-2 (S)Backbone Model=Qwen3-8B2026.06 | 82.5 | 318 | |
| NaiveBackbone Model=Qwen3-8B2026.06 | 77.11 | 0 | |
| Full TextBackbone Model=Phi-42026.06 | 75.66 | 1,093 | |
| SKIM (Fix-512)Backbone Model=Phi-42026.06 | 74.74 | 512 | |
| LLMLingua-2 (L)Backbone Model=Phi-42026.06 | 68.03 | 820 | |
| SKIM (Adaptive)Backbone Model=Phi-42026.06 | 66.97 | 335 | |
| LLMLingua-2 (M)Backbone Model=Phi-42026.06 | 63.16 | 607 | |
| SKIM (Fix-256)Backbone Model=Phi-42026.06 | 63.16 | 256 | |
| ICAEBackbone Model=Qwen3-8B2026.06 | 62.11 | 256 | |
| ICAEBackbone Model=Phi-42026.06 | 60.66 | 256 | |
| 500xCompressorBackbone Model=Phi-42026.06 | 60.53 | 256 | |
| LLMLingua-2 (S)Backbone Model=Phi-42026.06 | 53.42 | 319 | |
| NaiveBackbone Model=Phi-42026.06 | 48.55 | 0 | |
| 500xCompressorBackbone Model=Qwen3-8B2026.06 | 33.29 | 256 |