Public Examples / Research Literature Notes / Findings MemGPT improves dialogue recall in the reported test

Public Examples / Research Literature Notes / Findings

MemGPT improves dialogue recall in the reported test

Findings · Oct 20, 2026 · #finding

One reported finding from MemGPT’s dialogue experiment, linked to its source. It is specific to the paper’s evaluation and is not a Hjarni benchmark.

With GPT-4, MemGPT scored 92.5% dialogue-recall accuracy versus 32.1% for the compressed-history baseline.

Source: Packer 2023: MemGPT, §3.1.1, Table 2 (paper).

Scope: the systems use different history-access mechanisms. This is the authors’ reported experiment, not an independent replication or a Hjarni benchmark.

Interpretation: Retrieval before synthesis.