LongMemEval-V2: Evaluating Long-Term Agent Memory Toward Experienced Colleagues

Hugging Face Daily Papers Papers

Summary

This paper introduces LongMemEval-V2, a benchmark for evaluating long-term memory systems in web agents, along with two memory methods: AgentRunbook-R and AgentRunbook-C.

Long-term memory is crucial for agents in specialized web environments, where success depends on recalling interface affordances, state dynamics, workflows, and recurring failure modes. However, existing memory benchmarks for agents mostly focus on user histories, short traces, or downstream task success, leaving open how to directly evaluate whether memory systems effectively internalize environment-specific experience. To address this gap, we introduce LongMemEval-V2 (LME-V2), a benchmark for evaluating whether memory systems can help agents acquire the experience needed to become knowledgeable colleagues in customized environments. LME-V2 contains 451 manually curated questions covering five core memory abilities for web agents: static state recall, dynamic state tracking, workflow knowledge, environment gotchas, and premise awareness. Questions are paired with history trajectories containing up to 500 trajectories and 115M tokens. We use a context gathering formulation: memory systems consume history trajectories and return compact evidence for downstream question answering. We propose a suite of two memory methods: AgentRunbook-R, an efficient RAG-based memory with knowledge pools for raw state observations, events, and strategy notes, and AgentRunbook-C, which stores trajectories as files and invokes a coding agent to gather evidence in an augmented sandbox. Experiments show that AgentRunbook-C achieves the best performance with 72.5% average accuracy, outperforming the strongest RAG baseline (48.5%) and the off-the-shelf coding agent baseline (69.3%). Despite the strong performance gains, coding agent based methods have high latency costs. While AgentRunbook-C advances the accuracy-latency Pareto frontier, substantial room for improvement remains. Together, these results establish LME-V2 as a challenging testbed for developing long-term memory systems for environment experience.
Original Article
View Cached Full Text

Cached at: 05/13/26, 04:10 AM

Paper page - LongMemEval-V2: Evaluating Long-Term Agent Memory Toward Experienced Colleagues

Source: https://huggingface.co/papers/2605.12493 Published on May 12

·

Submitted byhttps://huggingface.co/xiaowu0162

Di Wuon May 13

Abstract

A new benchmark called LongMemEval-V2 is introduced to evaluate memory systems’ ability to help agents acquire environment-specific experience in web environments, featuring a suite of memory methods including AgentRunbook-R and AgentRunbook-C that demonstrate varying performance in accuracy and latency.

Long-term memoryis crucial for agents in specializedweb environments, where success depends on recalling interface affordances, state dynamics, workflows, and recurring failure modes. However, existingmemory benchmarks for agents mostly focus on user histories, short traces, or downstream task success, leaving open how to directly evaluate whether memory systems effectively internalize environment-specific experience. To address this gap, we introduce LongMemEval-V2 (LME-V2), a benchmark for evaluating whether memory systems can help agents acquire the experience needed to become knowledgeable colleagues in customized environments. LME-V2 contains 451 manually curated questions covering five core memory abilities for web agents: static state recall, dynamic state tracking, workflow knowledge, environment gotchas, and premise awareness. Questions are paired with history trajectories containing up to 500 trajectories and 115M tokens. We use a context gathering formulation: memory systems consume history trajectories and return compact evidence for downstreamquestion answering. We propose a suite of two memory methods:AgentRunbook-R, an efficientRAG-based memorywithknowledge poolsfor raw state observations, events, and strategy notes, andAgentRunbook-C, which stores trajectories as files and invokes a coding agent to gather evidence in an augmented sandbox. Experiments show thatAgentRunbook-Cachieves the best performance with 72.5% average accuracy, outperforming the strongest RAG baseline (48.5%) and the off-the-shelf coding agent baseline (69.3%). Despite the strong performance gains, coding agent based methods have high latency costs. WhileAgentRunbook-Cadvances the accuracy-latency Pareto frontier, substantial room for improvement remains. Together, these results establish LME-V2 as a challenging testbed for developinglong-term memorysystems for environment experience.

View arXiv pageView PDFProject pageGitHub2Add to collection

Get this paper in your agent:

hf papers read 2605\.12493

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.12493 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.12493 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.12493 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

MemGym: a Long-Horizon Memory Environment for LLM Agents

arXiv cs.CL

MemGym is a benchmark for evaluating memory formation in LLM agents over long-horizon tasks, unifying existing agent gyms and synthetic pipelines with memory-isolated scores. It spans tool-use dialogue, multi-turn search, coding, and computer use, and includes a lightweight reward model (MemRM) for efficient evaluation.