# CPU-ARC: Agentic Reasoning Caching for on-CPU Reasoning Reuse

DevFeed: [CPU-ARC: Agentic Reasoning Caching for on-CPU Reasoning Reuse](<https://devfeed.tech/articles/cpu-arc-agentic-reasoning-caching-for-on-cpu-reasoning-reuse-124739.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.09891>)

Author: Peter Belcak, Greg Heinrich, Saurav Muralidharan, Shizhe Diao, Pavlo Molchanov

Published: 2026-11-09T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [caching](<https://devfeed.tech/tags/caching.md>), [cpu](<https://devfeed.tech/tags/cpu.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [performance](<https://devfeed.tech/tags/performance.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [reuse](<https://devfeed.tech/tags/reuse.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [traces](<https://devfeed.tech/tags/traces.md>)

## AI overview

CPU-ARC turns LLM reasoning traces into agentic reasoning caches that run on CPUs. The approach separates solving a task from applying its solution to new instances, aiming to reuse prior reasoning with little LLM involvement. The abstract reports performance on par with frontier LLMs across difficult mathematical and scientific tasks, along with explainability, continual trainability, and robustness to input perturbations. It also claims invocation costs 16-291 times lower than the source model; the excerpt gives no evaluation details or conditions for these results.

## Source excerpt

arXiv:2611.09891v1 Announce Type: new Abstract: Large language models reinvent the wheel all the time. They conflate task-solving with the execution of the solutions, which forces them to solve every problem from scratch. We show that once a task has been reasoned through, the application of its solution to a new problem instance can be done very efficiently and with little involvement from language models. We introduce agentic reasoning caching, a method for constructing robust CPU-executable agentic reasoning caches (ARCs) from LLM reasoning traces. ARCs show on-par performance with frontier LLMs across a range of difficult mathematical and scientific tasks, are fully explainable, continually trainable, robust to input perturbations, and 16-291x cheaper to invoke than the source model.