CliffCompaction: Cost-Efficient Compaction for Long-Horizon Coding Agents

Researchers submitted CliffCompaction, an autocompaction technique for long-horizon coding agents, to arXiv on 22 Sep 2026, reporting cost reductions of up to 50% under a bounded context while maintaining or improving performance on Terminal-Bench and setting state-of-the-art results on KernelBench. CliffCompaction keeps compacted information faithful by only truncating or dropping content rather than rephrasing it, and never compacts a compaction, which the authors say sustains continual learning over sessions exceeding a million tokens; on KernelBench it reached CUDA kernel speedups of 2.23x after 200 steps and 3.58x after 400 steps. Under parallel test-time scaling, CliffCompaction lets Kimi K2.6 match Opus 4.7 and exceed Opus 4.6 and GPT-5.3 Codex at lower cost, and the authors open-sourced a scaffold-agnostic API-proxy implementation usable with Claude Code, Codex and other harnesses.

Computer Science Artificial Intelligence Submitted on 22 Sep 2026 Title:CliffCompaction: Cost-Efficient Compaction for Long-Horizon Coding Agents View PDF http://arxiv.org/pdf/2609.26779v1 HTML experimental https://arxiv.org/html/2609.26779v1 Abstract:Agents often work on complex problems that require millions of tokens of context, which necessitates compacting across sessions due to limited context windows. We develop CliffCompaction, an autocompaction technique that reduces cost by up to 50% under a bounded context while maintaining or improving performance on Terminal-Bench and achieving new levels of efficiency for test-time scaling and state-of-the-art results on KernelBench. The per-rollout savings of CliffCompaction make the performance--cost trade-off of test-time scaling more efficient, adding over 10 percentage points on Terminal-Bench for less than the cost of two full-context runs. Under parallel test-time scaling, CliffCompaction lets Kimi K2.6 match Opus 4.7, and exceed Opus 4.6 and GPT-5.3 Codex at lower cost. The key to CliffCompaction's effectiveness is that it keeps compacted information faithful by only truncating or dropping content, never rephrasing or rewriting it. We never compact a compaction---each pass operates only on original content, and prior compacted output is discarded, preventing context drift from accumulating. These properties sustain continual learning over sessions exceeding a million tokens: on KernelBench, CliffCompaction reaches CUDA kernel speedups of $2.23\times$ after 200 steps and $3.58\times$ after 400 steps, surpassing specialized search algorithms and trained agents despite being a general-purpose compaction technique. We open-source a scaffold-agnostic API-proxy implementation of CliffCompaction usable with Claude Code, Codex and other harnesses. Submission history From: Thien Trang Nguyen Vu view email http://arxiv.org/show-email/949bd38b/2609.26779 v1 Tue, 22 Sep 2026 17:55:59 UTC 417 KB Current browse context: cs.AI References & Citations Loading... Bibliographic and Citation Tools Bibliographic Explorer What is the Explorer? https://info.arxiv.org/labs/showcase.html arxiv-bibliographic-explorer Connected Papers What is Connected Papers? https://www.connectedpapers.com/about Litmaps What is Litmaps? https://www.litmaps.co/ scite Smart Citations What are Smart Citations? https://www.scite.ai/ Code, Data and Media Associated with this Article alphaXiv What is alphaXiv? https://alphaxiv.org/ CatalyzeX Code Finder for Papers What is CatalyzeX? https://www.catalyzex.com DagsHub What is DagsHub? https://dagshub.com/ Gotit.pub What is GotitPub? http://gotit.pub/faq Hugging Face What is Huggingface? https://huggingface.co/huggingface ScienceCast What is ScienceCast? https://sciencecast.org/welcome Demos Recommenders and Search Tools Influence Flower What are Influence Flowers? https://influencemap.cmlab.dev/ CORE Recommender What is CORE? https://core.ac.uk/services/recommender arXivLabs: experimental projects with community collaborators 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 https://info.arxiv.org/labs/index.html .