---
title: Niyra's memory, measured: 98.2% recall on LongMemEval
description: We ran the production memory stack against the complete LongMemEval benchmark, all 500 questions, and published every number and every caveat. 98.2% recall@10, 100% on knowledge updates, zero benchmark tuning.
url: /changelog/longmemeval-benchmark
lastUpdated: 2026-07-22
---

# Niyra's memory, measured: 98.2% recall on LongMemEval


The memory system that serves every Niyra conversation now has a public, reproducible number behind it. On the complete LongMemEval benchmark (500 questions, 246,738 conversation turns), the production retrieval stack scored 98.2% recall@10 and 92.0% recall@5, measured at turn level with zero benchmark-specific tuning. Knowledge-update questions, the "this fact changed, do you know the current truth?" category, came in at 100%. The [full write-up](/blog/niyra-longmemeval-benchmark) publishes the methodology, the control arm, and every caveat, including what these numbers do not mean.