https://tedfactory.com/en/tags/cost-optimization/Recent content in Cost Optimization on Ted FactoryHugoenThu, 20 Aug 2026 01:41:07 +0900https://tedfactory.com/en/notes/essays/why-ai-subscriptions-are-not-cheap-apis/Thu, 20 Aug 2026 00:00:00 +0900https://tedfactory.com/en/notes/essays/why-ai-subscriptions-are-not-cheap-apis/<h1 id="why-an-ai-subscription-isnt-a-cheap-api">Why an AI Subscription Isn&rsquo;t a Cheap API<a class="anchor" href="#why-an-ai-subscription-isnt-a-cheap-api">#</a></h1> <p>2026-08-20</p> <p><img src="https://tedfactory.com/images/notes/why-ai-subscriptions-are-not-cheap-apis-cover.png" alt="A personal subscription path and a metered API path lead toward the same AI through separate gates and billing systems" /></p> <h2 id="the-better-you-get-at-using-ai-the-more-cost-matters">The Better You Get at Using AI, the More Cost Matters<a class="anchor" href="#the-better-you-get-at-using-ai-the-more-cost-matters">#</a></h2> <p>When I first started using AI, model quality was what mattered most. I compared which model wrote better code, read long documents more accurately, and reasoned more deeply. But once AI became part of my actual work and my usage grew, my attention shifted. <strong>Using AI well now matters as much as being able to keep using it sustainably.</strong></p>https://tedfactory.com/en/notes/essays/prompt-caching-deep-dive/Wed, 05 Aug 2026 00:00:00 +0900https://tedfactory.com/en/notes/essays/prompt-caching-deep-dive/<h1 id="i-dug-into-prompt-caching-and-found-that-hit-rate-isnt-the-goal">I Dug Into Prompt Caching and Found That Hit Rate Isn&rsquo;t the Goal<a class="anchor" href="#i-dug-into-prompt-caching-and-found-that-hit-rate-isnt-the-goal">#</a></h1> <p>2026-08-05</p> <p><img src="https://tedfactory.com/images/notes/prompt-caching-deep-dive-cover.png" alt="Prompt content stacked like geological strata by change frequency, with the lower layers reused from cache" /></p> <h2 id="it-started-with-where-does-this-actually-happen">It Started With &ldquo;Where Does This Actually Happen?&rdquo;<a class="anchor" href="#it-started-with-where-does-this-actually-happen">#</a></h2> <p>I knew prompt caching reduced costs. My understanding went about as far as: when you send the same system prompt repeatedly, it gets cheaper from the second request onward.</p> <p>But the more I thought about it, the stranger it seemed. If caching happens in <strong>some relay layer outside the model</strong>, then from the model&rsquo;s point of view nothing is saved. And yet the savings are said to reach 90%. Would that mean most of the cost of using an LLM (Large Language Model) comes not from the model itself but from the layer around it?</p>