Long-context retrieval and reasoning on RULER synthetic 4K-64K
100Niah Score 1 (S1)
ET
Updated 2mo ago
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
2026.05 | 100 | 91.28 | 92.92 | 90.2 | 85.48 | 62.92 | 81.67 | 82.06 | 82 | 42.82 | 83.53 | 51.88 | 41.6 | 76.03 | |
2026.05 | 100 | 99.28 | 79.68 | 96.12 | 76.48 | 45.72 | 77.81 | 83.88 | 68.18 | 53.14 | 63.01 | 53.04 | 42.84 | 72.24 | |
2026.05 | 100 | 99.44 | 86.2 | 97.36 | 77.96 | 51.92 | 81.56 | 83.79 | 65.7 | 50.97 | 58.17 | 52.4 | 42.92 | 72.95 | |
2026.05 | 97.56 | 98.12 | 72.72 | 94.52 | 90.24 | 56.2 | 62.34 | 62.56 | 68.56 | 38.32 | 58.1 | 55.68 | 45.04 | 69.23 |