aiqbench analysis
DeepControl introduces a novel framework that optimizes information acquisition in search-augmented reasoning agents by regulating both the extent and resolution of retrieved evidence. This approach addresses critical issues like context saturation and RL instability, enabling models to internalize efficient search behaviors without external control during inference. The method demonstrates significant performance gains on standard benchmarks, marking a meaningful advancement in how LLMs leverage external knowledge sources.
Adaptive Information Control for Search-Augmented LLM Reasoning
arXiv:2602.01672v3 Announce Type: replace Abstract: Search-augmented reasoning agents interleave multi-step reasoning with external retrieval, but uncontrolled retrieval can introduce redundant evidence, saturate the context, and destabilize reinforcement learning (RL). Existing outcome-based RL methods provide only sparse terminal rewards, offering limited guidance for intermediate information-acquisition decisions. We propose DeepControl, an adaptive information-control framework based on information utility, a state-dependent estimate of the marginal value of retrieved evidence. The framework regulates information acquisition along two axes: extent, i.e., whether retrieval should continue, and resolution, i.e., how much retrieved detail should be exposed. It implements these controls through retrieval-continuation guidance, hierarchical granularity control, and an annealed control-forcing scheme. This enables the policy to internalize effective acquisition behavior during training and operate without external control at test time. Across seven benchmarks, DeepControl consistently outperforms strong RL and retrieval baselines without explicit information control; compared with Search-R1, it improves average performance by +9.4 and +8.6 points on Qwen2.5-7B and Qwen2.5-3B, respectively. Additional analyses show improved search effectiveness, training stability, and evidence utilization.
Abstract:Search-augmented reasoning agents interleave multi-step reasoning with external retrieval, but uncontrolled retrieval can introduce redundant evidence, saturate the context, and destabilize reinforcement learning (RL). Existing outcome-based RL methods provide only sparse terminal rewards, offering limited guidance for intermediate information-acquisition decisions. We propose DeepControl, an adaptive information-control framework based on information utility, a state-dependent estimate of the marginal value of retrieved evidence. The framework regulates information acquisition along two axes: extent, i.e., whether retrieval should continue, and resolution, i.e., how much retrieved detail should be exposed. It implements these controls through retrieval-continuation guidance, hierarchical granularity control, and an annealed control-forcing scheme. This enables the policy to internalize effective acquisition behavior during training and operate without external control at test time. Across seven benchmarks, DeepControl consistently outperforms strong RL and retrieval baselines without explicit information control; compared with Search-R1, it improves average performance by +9.4 and +8.6 points on Qwen2.5-7B and Qwen2.5-3B, respectively. Additional analyses show improved search effectiveness, training stability, and evidence utilization.
Computation and Language (cs.CL)
arXiv:2602.01672 [cs.CL]
(or arXiv:2602.01672v3 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2602.01672 Focus to learn more arXiv-issued DOI via DataCite
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.
Full text under the publisher's open licence · 255 words
Related coverage
Adaptive Information Control for Search-Augmented LLM Reasoning
arXiv · Jun 4, 2026
From Vector Search to Agentic RAG: Building an Enterprise Research Analyst with hipVS
AMD · Jul 15, 2026
Problems of Proliferation
LessWrong · Sep 29, 2026
AgentOps: Operationalize agentic AI at scale with Amazon Bedrock AgentCore
AWS · Jun 1, 2026