titleRead Less, Solve More: Token-Efficient Sparse Reading of External Content for AI Agents
authorsZedong Liu, Jiaan Wu, Xinyang Ma, Le Xu, Kai Wang, Yuanchao Hu, Dingwen Tao, Guangming Tan
date2026-11-07
venuearXiv (preprint, not peer reviewed)
idarXiv:2608.22237
affiliationInstitute of Computing Technology
upvotes21 (Hugging Face Daily Papers, 2026-11-14)
linksabs pdf hf code
cite
relatedDoc2Harness: Benchmarking LLMs on Long-Horizon Harness Engineering of Coding Agents
IndusAgent: Reinforcing Open-Vocabulary Industrial Anomaly Detection with Agentic Tools
SkillForge: Co-Evolving Skills and Agents via Dynamic Skill Lifecycles
KVServe: Service-Aware KV Cache Compression for Communication-Efficient Disaggregated LLM Serving
abstractWeb-facing AI agents repeatedly access webpages, documents, repositories, and tool responses, yet reading interfaces often expose entire objects when only sparse evidence is needed. This over-reading increases token and latency costs and can dilute relevant evidence. We present SparseRead, a reusable reading control layer that governs content admission without modifying the agent model. Its Read Intent Protocol tracks evolving needs and source coverage; Adaptive Reader Backends select how evidence is located and exposed; and the Budgeted Read Gate bounds cumulative raw exposure per episode with faithful full-view fallback. A unified Reader interface supports interchangeable evidence acquisition across heterogeneous resources. Across six models, including Claude Opus 5, and five scenarios, SparseRead reduces total tokens by up to 92.9% and wall time by up to 89.0%, preserving or improving task quality in 97% settings. On Web deep-search tasks, it reduces tokens by 31.0--43.0% and task completion time by 23.0--37.0%. Ablations validate the three designs, while integration with three agent frameworks demonstrates portability.

bibtex

@misc{liu2026read,
  title={Read Less, Solve More: Token-Efficient Sparse Reading of External Content for AI Agents},
  author={Zedong Liu and Jiaan Wu and Xinyang Ma and Le Xu and Kai Wang and Yuanchao Hu and Dingwen Tao and Guangming Tan},
  year={2026},
  eprint={2608.22237},
  archivePrefix={arXiv},
  url={https://arxiv.org/abs/2608.22237}
}