# Cut your LLM bill by 30 to 70%: the levers that work > Source: > Published: 2026-07-12 07:54:52+00:00 On the bills I audit, the problem is almost never the price per token. It is useless context sent on every call and the most expensive model plugged in everywhere by default. Here is what I cut first. In nearly every engagement, the problem is not the price per token but the way the tokens are spent: useless context sent back on every call, the most expensive model used everywhere by default, answers regenerated when they already existed. So I start by measuring where each euro goes, not by cutting at random. Yes, and it is almost always the first lever: in production, a large share of calls are near-duplicates. Caching the answers on identical inputs removes that waste without changing anything for your users, often within a few days. No. Routing each request to the cheapest model capable of handling it is enough in most cases: a simple classification or extraction does not need the most powerful model. I never cut a bill by degrading quality. I cut it by no longer paying for what adds nothing. Taken together, these levers bring a bill down by 30 to 70% on most of the products I audit, at equal quality, and the first gains usually land within two to three weeks. **How fast does the LLM bill drop?** The first levers (caching, model routing) show results within two to three weeks. **Does cutting the bill mean degrading quality?** No. I remove the useless context, the duplicates and the use of the most expensive model where a lighter one is enough. The quality measured by my tests stays stable, only the cost goes down. **Do I need to switch model provider to save?** Rarely. Most of the savings come from how you call your current models. I write about shipping AI to production at [guinat.ai](https://www.guinat.ai). Honest advice, no hype.