Adaptive Consistency Graph for Long-Horizon Agents
Summary
The paper introduces Adaptive Consistency Graph (ACG), a method that enhances long-horizon decision-making in LLM agents by structuring evidence and context, leading to improved success rates in benchmarks.
View Cached Full Text
Cached at: 09/29/26, 04:11 AM
Paper page - Adaptive Consistency Graph for Long-Horizon Agents
Source: https://huggingface.co/papers/2609.32754 Published on Sep 26
·
Submitted byhttps://huggingface.co/yunsaijc
jcon Sep 29
Abstract
Largelanguagemodelagentscanoftenmakereasonablelocaldecisionsonshorttasks,yettheirperformancedegradeswhensuccessrequireslongsequencesofdependentactionsandtoolcalls.Duringexecution,taskrequirements,historicalevidence,andthecurrentexecutionstatemaygraduallybecomedisconnected,solaterdecisionscandriftfromtheoriginalobjective.WestudythisproblembyintroducingtheAdaptiveConsistencyGraph(ACG)forlong-horizonexecution.ACGincrementallyorganizesexecutionevidenceanditsprovenanceinapersistentgraph,thenconstructsatemporaryrequirement-centeredviewforeachdecisionunderaboundedcontextbudget.Ratherthanreplacingthebaseagent’splannerortoolexecutor,ACGprovidesastructuredandtraceablecontextviewforeachdecision.Inthematchedevaluation,ACGimprovesGPT-5.6-luna’saveragesuccessfrom44.5\%withReActto50.2\%,withthelargestgainonBrowseComp-Plus(73.5\%versus62.4\%).Wefurtheranalyzetrajectorystructureandinferencecosttocharacterizethisimprovement.
View arXiv pageView PDFAdd to collection
Get this paper in your agent:
hf papers read 2609\.32754
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/2609.32754 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2609.32754 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2609.32754 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
Granularity-Adaptive Credit Assignment for Long-Horizon LLM Agent Reinforcement Learning
The paper proposes GACA, a granularity-adaptive credit assignment method for long-horizon LLM agent reinforcement learning that improves task success by adapting resolution to step importance.
ACC: Compiling Agent Trajectories for Long-Context Training
ACC converts multi-turn agent trajectories into long-context QA pairs to train LLMs on long-range reasoning without additional annotation, achieving significant gains on MRCR and GraphWalks benchmarks while preserving general capabilities.
Learning Agent-Compatible Context Management for Long-Horizon Tasks
Introduces AdaCoM, an external LLM-based context manager for frozen agents, using reinforcement learning to improve long-horizon task performance by preserving task constraints and pruning stale content, with experiments on web search and deep research benchmarks.
ExpGraph: Model-Agnostic Experience Learning with Graph-Structured Memory for LLM Agents
ExpGraph is a model-agnostic framework that enables LLM agents to reuse past experiences via a self-evolving graph of skills and failures, improving task performance by 12–21% without retraining the executor.
AgentWorld: Benchmarking Long-Horizon Collaboration of Multi-agent LLMs
AgentWorld introduces a benchmark for evaluating long-horizon collaboration in multi-agent LLM systems with 100 tasks and a new causal collaboration effectiveness metric. Experiments show that even top models achieve only 52% task success, highlighting failure modes like communication breakdowns.