Skip to content
AI News HubLIVE
Original source2 min read

MEMENTO: Learning Context Management for Long Context Processing Agents

Summary

Researchers present MEMENTO, a long-context agent built on Qwen3-8B and trained to manage its context through compaction. It first bootstraps note-taking by fine-tuning on synthetic trajectories annotated with hindsight-generated notes, then applies RL with a reward combining answer correctness, note-quality heuristics, and confidence calibration. On book-level QA benchmarks, MEMENTO lifts Qwen3-8B's accuracy from 36.9% to 67.8% while never holding more than 20K tokens in context, outperforming GPT-5.5 agents and nearing a far larger GLM-5.3 agent at a fraction of the cost.

SourcearXiv Computational LinguisticsAuthor: Yunfan Zhang, Taehoon Hwang, Kathleen McKeown, Smaranda Muresan
MEMENTO: Learning Context Management for Long Context Processing Agents
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 30 Oct 2026]

Title:MEMENTO: Learning Context Management for Long Context Processing Agents

View a PDF of the paper titled MEMENTO: Learning Context Management for Long Context Processing Agents, by Yunfan Zhang and 3 other authors

View PDF HTML (experimental)

Abstract:LLM agents that operate over long inputs must perform context management: deciding what to retain in a fixed-size context window. A common approach is compaction, where the model replaces its history with a self-generated note. Yet off-the-shelf agentic models, including GPT-5.5 and GLM-5.3, often write ineffective notes, and reinforcement learning (RL) alone rarely discovers useful note-taking behavior. We present MEMENTO, a long-context agent built on Qwen3-8B and trained to manage its context through compaction. We first bootstrap note-taking by fine-tuning on synthetic trajectories annotated with notes generated in hindsight, then reinforce it with RL using a reward that combines answer correctness, note-quality heuristics, and confidence calibration. Although trained only to compact into a fresh window, MEMENTO extends at test time to strategies such as context folding and confidence-based backtracking. On book-level question answering across four benchmarks, with books of up to 2.5M tokens, MEMENTO raises Qwen3-8B's accuracy from 36.9% to 67.8% while never holding more than 20K tokens in context. It outperforms GPT-5.5 agent in the same harness (13.1%) and full book context (56.7%) at less than 1/50 the cost. It also nearly matches a ~100x larger GLM-5.3 agent (69.3%) at 1/9 the cost, and comes within 6.4 points of GLM-5.3 full context at 1/26 the cost.

Subjects:

Computation and Language (cs.CL); Information Retrieval (cs.IR)

Cite as: arXiv:2611.00244 [cs.CL]

(or arXiv:2611.00244v1 [cs.CL] for this version)

https://doi.org/10.48550/arXiv.2611.00244

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Yunfan Zhang [view email] [v1] Fri, 30 Oct 2026 18:25:08 UTC (195 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled MEMENTO: Learning Context Management for Long Context Processing Agents, by Yunfan Zhang and 3 other authors

View PDF

HTML (experimental)

TeX Source

view license

Additional Features

Audio Summary

Current browse context:

cs.CL

new | recent | 2026-11

Change to browse by:

cs cs.IR

References & Citations

SciX

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • MEMENTO bootstraps note-taking via fine-tuning on synthetic trajectories with hindsight-generated notes, then optimizes with RL using a reward that combines answer correctness, note-quality heuristics, and confidence calibration.
  • On book-level QA with books up to 2.5M tokens, accuracy rises from 36.9% to 67.8% while context never exceeds 20K tokens.
  • It beats a GPT-5.5 agent (13.1%) at under 1/50 the cost and nearly matches a ~100x larger GLM-5.3 agent (69.3%) at 1/9 the cost.
  • Trained only to compact into a fresh window, it extends at test time to context folding and confidence-based backtracking.

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Agents