🪙 How Not to Waste Tokens in Language Models

If you actively use language models like Claude or GPT, you've probably noticed how quickly token limits run out. This is especially true for premium models like Opus 4.6, whose usage is limited even on paid subscriptions. The reason is that each request sends the entire conversation history, and the longer it is, the more tokens are consumed. But there is a way to save.

Try making a brief summary. At a certain point in the conversation with the model (for example, in the middle of the dialogue), ask it to formulate a summary in markdown format. This summary should contain all the key information so that another agent can continue the conversation from the same point. Then open a new agent, paste this summary, and ask the next question. The new dialogue will start with minimal history but with the context preserved 👍

👩‍💻 Data Flow