Newest papers

MemoryTax: The Bitter Lesson of Agent Memory

Zedong Peng · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.29823 · cited by 0 works (OpenAlex) · open-access PDF

Indonesian

Sistem memori konteks agen seperti Mem0 dan Zep memutuskan apa yang harus disimpan sebelum pertanyaan datang; agen pengkodean menyimpan sejarah mentah dan membiarkan model mencarinya dengan alat-alatnya sendiri. Kami bertanya apakah memori yang dirancang masih membantu ketika model itu cukup kuat untuk mencari dirinya sendiri. Pada satu percakapan LoCoMo (419 putar, 152 pertanyaan), kami menjalankan empat metode memori (konteks penuh, BM25, Mem0, Zep) dan tiga agen pengkodean (Pi, Codex, Claude Code), masing-masing dengan GPT-4o-mini, GPT-6 Luna dan Claude Opus 5.5, dan merekam apa yang harganya setiap rolllout. Ini adalah studi empiris singkat, dengan satu percakapan dan satu run per sel. Memori agen adalah patch mahal untuk model yang lemah: Mem0 mengalahkan bukti lengkap hanya dengan GPT-4o-mini, dan

Machine translation (Meta M2M100); the English original follows.

Abstract

Agent memory systems such as Mem0 and Zep decide what to keep before any question arrives; coding agents keep the raw history and let the model search it with its own tools. We ask whether designed memory still helps once the model is strong enough to search for itself. On one LoCoMo conversation (419 turns, 152 questions), we run four memory methods (full context, BM25, Mem0, Zep) and three coding agents (Pi, Codex, Claude Code), each with GPT-4o-mini, GPT-6 Luna and Claude Opus 5.5, and record what every rollout costs. This is a short empirical study, with one conversation and one run per cell. Agent memory is a costly patch for a weak model: Mem0 beats full context only with GPT-4o-mini, and its build alone costs more than a whole full-context rollout. Agent memory also has a ceiling, set by the model that writes it: a stronger reader does not recover what a weaker writer left out. With Claude Opus 5.5, Pi tops the cost-accuracy frontier, scoring 99% for less than half the cost of full context; its one miss is a wrong answer key. Part of that lead comes from photo captions that only the agents could read: on the 88 questions whose evidence has no photo, Pi and full context differ by one answer.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Conversation, Computer science, Coding (social sciences), Context (archaeology), Ask price, Set (abstract data type)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange