AI used to feel free. Pay a monthly fee, run everything in Opus, never think about it again.

That ended. In June 2026 GitHub Copilot stopped charging per request and started charging per token. Claude started working with sessions. These changes left us with a paradox: you pay for the tokens the model spends thinking, and you cannot see or control how many that is.

Lets talk about our options. Let's talk about how to optimize our input and instructions, how caching and context windows work and the difference a model really makes.

Coding agents will never be as cheap as they were, but you will leave able to cut the bill without cutting what you get out of it.