https://jaehun.me/en/tags/llm-agents/Recent content in LLM Agents on Jaehun's BlogHugoen-USFri, 23 Oct 2026 08:19:00 +0900https://jaehun.me/en/posts/paper-2610-28577v1/Fri, 23 Oct 2026 00:00:00 +0900https://jaehun.me/en/posts/paper-2610-28577v1/<p><a href="https://arxiv.org/abs/2610.28577v1"target="_blank" class="inline-flex items-center gap-1" >Paper<svg class="h-3 w-3 flex-shrink-0" id="external-link" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M15 3h6v6m-11 5L21 3m-3 10v6a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2V8a2 2 0 0 1 2-2h6"/></svg> </a></p> <h2 id="the-cost-and-accuracy-dilemma-of-long-agent-trajectories">The Cost and Accuracy Dilemma of Long Agent Trajectories<a href="#the-cost-and-accuracy-dilemma-of-long-agent-trajectories" class="heading-anchor" aria-label="Link to this section"></a></h2><p>LLM agents accumulate long trajectories by repeating reasoning and tool calls for hundreds of rounds. To make inference fast, the already-computed key-value (KV) cache must stay on the GPU and be read again for each next token; as the trajectory grows, storage and read costs increase. The paper measures KV cache reads at 82.4% of total API cost (§1, Figure 13), and points out that accuracy degrades once the number of KVs or the positional span handled by attention exceeds what the model can handle (§1).</p>