Top-k Memory - LongMemEval (LEDGER, 479 Questions): leaderboard
Top-k Retrieval Leaves Memory Value Undetermined v1 Appendix Table7, 479 LongMemEval questions under the LEDGER retrieval policy, fixed gemini-2.5-flash-lite reader at temperature 0 and 24000-character limit. LoCoMo stores are shared within each conversation; LongMemEval pools the oracle histories and retains 479 questions with gold turns. Each policy has its own board; the 479-question population is distinct from Table9's 300-question reader comparison. Reported metric: Accuracy (fraction).
Metric: Accuracy (fraction). Source: arxiv.org. 1 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | gemini-2.5-flash-lite | 0.27 |
Interactive version: theaggregate.ai/benchmark?slug=top-k-memory-longmemeval-ledger-479-questions · How It Works · Data refreshed daily, snapshot 2026-11-05.