Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-08-05 · high
The system sorts dialogue into three memory types and updates only the ones that change, cutting inference cost while beating every baseline tested.
Why now: As of August 5, LeanMem's benchmark numbers are the most direct comparison yet against single-pass memory summarization.
LeanMem sorts every piece of a conversation into one of three memory types, then stores each differently, per a paper posted to arXiv.
Most memory systems for AI agents run everything through one summarization process. That wastes tokens compressing facts that will never change, and loses detail on facts that need to stay exact. LeanMem beat the strongest baseline in every setting tested on LoCoMo and LongMemEval-S, gaining up to 15.1 points. It also ran at the lowest or near-lowest inference token count and latency of the methods compared.
The three types split by how a piece of dialogue behaves over time. Stable facts about a person go into compact profile memory. Things that change as the conversation evolves become temporally structured event memory. Anything that needs to stay traceable to its exact source, like quotes, numbers, or commitments, gets stored as source-grounded record memory.
The part worth taking for production agents is the maintenance rule, not the classification scheme. LeanMem only re-processes event memory when it updates. Profile and record memory get written once and left alone. Most memory implementations re-summarize the whole store on every update cycle, which is where the token cost actually comes from.
The results held across two different models, GPT-4.1-mini and Qwen3-8B, which suggests the gain comes from the architecture and not from either model's quirks.
The paper doesn't say how the routing decision gets made at write time, whether by a classifier, a prompt, or hand-coded rules. That's the detail that decides whether the approach is easy to reproduce.
Each link below shares sources, entities, or timing with this story.
NapMem benchmarked against LoCoMo / Shared entities / Same source domain / Earlier coverage
Linked by a graph relationship (NapMem benchmarked against LoCoMo); both cover LoCoMo, LongMemEval; reported by the same outlet (arxiv.org).
mempalace benchmarked against LoCoMo / Shared entities / Earlier coverage / Tension
Linked by a graph relationship (mempalace benchmarked against LoCoMo); both cover LoCoMo, LongMemEval; earlier LoCoMo coverage from 2026-07-28.
GPT competes with Claude / Shared entities / Same source domain / Earlier coverage
Linked by a graph relationship (GPT competes with Claude); both cover GPT, Qwen3; reported by the same outlet (arxiv.org).
GPT competes with Claude / Shared entity: GPT / Same source domain / Earlier coverage / Tension
Linked by a graph relationship (GPT competes with Claude); both cover GPT; reported by the same outlet (arxiv.org).
EverMemOS benchmarked against LoCoMo / Shared entities / Same source domain / Earlier coverage
Linked by a graph relationship (EverMemOS benchmarked against LoCoMo); both cover LoCoMo, LongMemEval; reported by the same outlet (arxiv.org).
GPT competes with Claude / Shared entity: Qwen3 / Shared topic / Earlier coverage
Linked by a graph relationship (GPT competes with Claude); both cover Qwen3; overlapping topics (cost, inference).
GPT competes with Claude / Shared entity: GPT / Same source domain / Earlier coverage
Linked by a graph relationship (GPT competes with Claude); both cover GPT; reported by the same outlet (arxiv.org).
Linked by a graph relationship (GPT competes with Claude); both cover GPT; reported by the same outlet (arxiv.org).