Needle-In-A-Haystack Retrieval
Benchmarks
Dataset NameSOTA methodMetricTrendResultsLast Updated
100S-NIAH-1 (Pass-Key Retrieval)
94
May 22, 2026
76Accuracy
30
Feb 26, 2026
100Accuracy
30
Feb 26, 2026
94Accuracy
23
Jun 30, 2026
98.8Accuracy
23
Jun 30, 2026
100Accuracy
23
Jun 30, 2026
100Accuracy
23
Jun 30, 2026
95.2Accuracy
23
Jun 30, 2026
98.6Accuracy
23
Jun 30, 2026
100Accuracy
23
Jun 30, 2026
100Accuracy
23
Jun 30, 2026
100Accuracy
23
Jun 30, 2026
100Accuracy
23
Jun 30, 2026
100Accuracy (NIAH-Single-1 32K)
23
Jun 30, 2026
100NIAH Score
14
Jun 2, 2026
100Accuracy
13
Jun 30, 2026
28.2F1 Score
8
May 19, 2026
2,800Multi-Query Success Rate
8
Feb 26, 2026
100Accuracy
6
May 12, 2026
100Retrieval Success (1K)
4
May 8, 2026
83.2Success Rate (4K Context)
4
Mar 4, 2026
100Retrieval Success Rate (1K)
4
Mar 4, 2026
100Success Rate (4K Context)
4
Mar 4, 2026
100Retrieval Success Rate (1K Context)
4
Mar 4, 2026
9Accuracy
3
Mar 24, 2026