No Literal Matching — a needle-in-a-haystack variant that removes lexical overlap to prevent shortcut matching, measuring true long-context recall accuracy.
Eliminates lexical matching shortcuts that inflate scores on traditional NIAH tests. Tests genuine comprehension rather than surface-level retrieval.
Measures only retrieval recall, not reasoning over retrieved content. Scores can vary with haystack composition.
Higher is better. Accuracy percentage.
Max score: 100
Measures only retrieval recall, not reasoning over retrieved content.
| # | Model | Score |
|---|---|---|
| 1 | 99.3 | |
| 2 | 99.3 | |
| 3 | 97.3 | |
| 4 | 97.3 | |
| 5 | 97.0 | |
| 6 | 97.0 | |
| 7 | 96.7 | |
| 8 | 96.5 | |
| 9 | 96.3 | |
| 10 | 96.1 | |
| 11 | 96.1 | |
| 12 | 95.8 | |
| 13 | 95.8 | |
| 14 | 95.5 | |
| 15 | 95.5 | |
| 16 | 95.5 | |
| 17 | 95.4 | |
| 18 | 95.4 | |
| 19 | 95.3 | |
| 20 | 95.3 | |
| 21 | 95.3 | |
| 22 | 95.2 | |
| 23 | 95.2 | |
| 24 | 95.1 | |
| 25 | 95.1 | |
| 26 | 95.1 | |
| 27 | 95.1 | |
| 28 | 95.1 | |
| 29 | 95.0 | |
| 30 | 95.0 | |
| 31 | 95.0 | |
| 32 | 95.0 | |
| 33 | 95.0 | |
| 34 | 94.9 | |
| 35 | 94.9 | |
| 36 | 94.9 | |
| 37 | 94.9 | |
| 38 | 94.9 | |
| 39 | 94.8 | |
| 40 | 94.8 | |
| 41 | 94.8 | |
| 42 | 94.8 | |
| 43 | 94.7 | |
| 44 | 94.7 | |
| 45 | 94.7 | |
| 46 | 94.7 | |
| 47 | 94.7 | |
| 48 | 94.7 | |
| 49 | 94.7 | |
| 50 | 94.7 |