Long-context Reasoning on LongBench 2
50.3P@4MPD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MPD2026.05 | 50.3 | 700 | |
| CRISP2026.05 | 49.9 | 1,000 | |
| Vanilla LLM2026.05 | 48.7 | 900 | |
| Direct Comp.2026.05 | 48.1 | 800 | |
| Chain-of-Draft2026.05 | 40.8 | 1,800 | |
| LiteCoT2026.05 | 36.6 | 2,000 |
Loading the SOTA2 catalog…