BenchmarkNeedle-in-a-Haystack / NIAH
Needle-in-a-Haystack Test
A benchmark for evaluating a large language model's ability to locate specific information within very long contexts. A target piece of information (the needle) is embedded within a large body of irrelevant text, and the model is assessed on its retrieval accuracy. The test reveals attention blind spots in long-context processing.
Timeline (last 90 days)
Oct 20
The needle-in-a-haystack test is a commonly used benchmark for evaluating the effective recall capability of long-context models.
Unverified50%