aiqbench analysis
This article is about AI: an arXiv paper on adaptive sparse-attention index reuse for language agents, with the primary subject being machine learning methods for efficient long-context inference. It is a research preprint, so its claims rest on benchmark results (SWE-bench Verified, Terminal-Bench 2.0, τ²-bench) for one model, Qwen3-4B-Instruct-2507, and should be read as promising but unverified across other backbones. For readers, the practical takeaway is that learned, per-step decisions about when to recompute attention indices may cut inference cost for long-horizon agents without hurting task success, though the gains are modest and the 44-of-50 result leaves room for settings where a static schedule does as well or better.
When Should an Agent Refresh? Adaptive Index Reuse for Efficient Long-Horizon Reasoning
arXiv:2611.01429v1 Announce Type: new Abstract: Language agents read code, tool output, and dialogue in contexts that grow at every step. Dynamic sparse attention lets a small indexer choose the earlier tokens each query attends to, but the indexer still scores the whole prefix at every layer. Cross-layer reuse methods such as IndexCache rerun the indexer at only some layers, with one fixed pattern for every input. In an agent, the relevant evidence can change abruptly when a traceback or test result arrives, so fresh selections matter more at some steps than at others. We propose Adaptive Index Reuse (AIR), which keeps IndexCache's execution and learns, for each segment of an agent episode, how many layers refresh their selection. A small controller reads signals from completed computation and the latest tool event, chooses a pattern from a nested library, and learns from terminal task reward under an expected compute budget. On Qwen3-4B-Instruct-2507 with a 128K context, AIR improves success over the strongest static schedule at matched cost on SWE-bench Verified (27.0% to 32.4%), Terminal-Bench 2.0, and three $\tau^2$-bench domains, and it lowers time per successful task on all five. Across 50 backbone, context, and benchmark settings, it matches or exceeds that schedule in 44.
Title:When Should an Agent Refresh? Adaptive Index Reuse for Efficient Long-Horizon Reasoning
Abstract:Language agents read code, tool output, and dialogue in contexts that grow at every step. Dynamic sparse attention lets a small indexer choose the earlier tokens each query attends to, but the indexer still scores the whole prefix at every layer. Cross-layer reuse methods such as IndexCache rerun the indexer at only some layers, with one fixed pattern for every input. In an agent, the relevant evidence can change abruptly when a traceback or test result arrives, so fresh selections matter more at some steps than at others. We propose Adaptive Index Reuse (AIR), which keeps IndexCache's execution and learns, for each segment of an agent episode, how many layers refresh their selection. A small controller reads signals from completed computation and the latest tool event, chooses a pattern from a nested library, and learns from terminal task reward under an expected compute budget. On Qwen3-4B-Instruct-2507 with a 128K context, AIR improves success over the strongest static schedule at matched cost on SWE-bench Verified (27.0% to 32.4%), Terminal-Bench 2.0, and three $\tau^2$-bench domains, and it lowers time per successful task on all five. Across 50 backbone, context, and benchmark settings, it matches or exceeds that schedule in 44.
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2611.01429 [cs.LG]
(or arXiv:2611.01429v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2611.01429 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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 · 308 words