Agent memory is not always better with more: eight-model evaluation shows dosage must be calibrated to capability
Original title:How Much Memory Does Your Agent Actually Need?
AI summary
Agent memory is not a feature that can be switched on at will, but a dosage that must be calibrated to model capability. Strong models benefit from injecting the full guide set, with DeepSeek-V3.2 (671B MoE) improving task completion by +9.5 percentage points; weaker models perform best with curated retrieval, with gpt-oss-120b (117B MoE) gaining +16.1 pp while adding only +5% tokens. The method requires no weight updates or human annotation, distilling guides from agents' past trajectories and injecting them at inference time.
Key Takeaways
- Agent memory is not a feature that can be switched on at will,but a dosage that must be calibrated…
- Strong models benefit from injecting the full guide set,with DeepSeek-V3.
- 2 (671B MoE) improving task completion by +9.
- 5 percentage points
- weaker models perform best with curated retrieval,with gpt-oss-120b (117B MoE) gaining +16.
Quick Facts
- Published
- Updated
- Source
- Hugging Face 社区博客(混合发现)
- Category
- Papers & Research
- AI score
- 66
把智能体记忆的配置从越多越好修正为按模型能力校准,弱模型用检索核心比全量更准且成本更低,为上下文预算提供量化依据。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Agent memory is not always better with more: eight-model evaluation shows dosage must be calibrated to capability. August 19, 2026. https://new.nat6.net/en/news/cmsz1974l06hbro20gosn5d78/