← All notes

Summary

LongMemEval is a benchmark of 500 questions set in scalable user-assistant chat histories, used to test long-term interactive memory. The authors report that commercial chat assistants and long-context models show a 30% accuracy drop on memorizing information across sustained interactions. They provide two standard history sizes: LongMemEval S at about 115k tokens, and LongMemEval M at 500 sessions (around 1.5 million tokens). [S1]

Claims

• id: C1 claim: 'Benchmark and code are stated to be publicly available.' source: S1 evidence: "Our benchmark and code are publicly available at https://github.com/xiaowu0162/LongMemEval"
• id: C2 claim: 'They report a 30% accuracy drop on memorizing information across sustained interactions.' source: S1 evidence: "showing a 30% accuracy drop on memorizing information across sustained interactions"
• id: C3 claim: 'LongMemEval S is about 115k tokens per problem.' source: S1 evidence: "LongMemEval S with approximately 115k tokens per problem"

Verification

• Code available: Paper: https://github.com/xiaowu0162/LongMemEval (HTTP 200 on 2026-11-11). Not run.
• Reproduced: not reproduced (2026-11-11). Code not run.
• Related work: Live note ao-memory-long-term-benchmarks. LongMemEval-V2 in this batch extends the setting toward agent trajectories.
• Limits: Author-reported. History length is freely extensible; the 115k and 1.5 million figures are their standard settings, not a ceiling.
• License: CC BY 4.0 on the abs page (2026-11-11; icon by-4.0.png; http://creativecommons.org/licenses/by/4.0/). Submitted 14 Oct 2024 (v1); last revised 4 Mar 2025 (v2). Comments: ICLR 2025. Sources: [S1] https://arxiv.org/abs/2410.10813, v2 (4 Mar 2025). checked_on: 2026-11-11.

When to apply

Use when asking how well a chat assistant remembers facts, preferences, or events across a long history, and what benchmark to cite.

Facts

• id: F1 claim: 'LongMemEval has 500 curated questions embedded in scalable user-assistant chat histories.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F2 claim: 'The authors report a 30% accuracy drop for commercial assistants and long-context models on memorizing information across sustained interactions.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0
• id: F3 claim: 'Standard settings are LongMemEval S at about 115k tokens and LongMemEval M at 500 sessions, around 1.5 million tokens.' confidence: medium date: 2026-11-11 sources: [S1] license: CC BY 4.0

Sources

• [S1] arXiv 2410.10813, https://arxiv.org/abs/2410.10813. License: CC BY 4.0. v2 (4 Mar 2025). Accessed 2026-11-11.

License: this note is CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Summary by iinventedlight (AI-assisted). Paper is CC BY 4.0.

@misc{iil_note_35pqmp7y,
  author = {{'iinventedlight (AI-assisted)'}},
  title = {{LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-11-11},
  url = {https://iinventedlight.com/library/notes/2410-10813-longmemeval/},
  note = {Permanent ID iil:note:35pqmp7y. License: CC-BY-4.0.},
}

Markdown

Spotted a mistake? Use the MCP feedback tool or email hello@iinventedlight.com.

For language models and agents: start at /llms.txt | /agents.md | /.well-known/agent-library.json. This page in Markdown: /library/notes/2410-10813-longmemeval.md.