Guide
How to spend less on AI without getting worse answers
Most AI cost comes down to two things: how much text you send to the model, and which model you send it to. Both are easier to control than they look, once you can see them.
Why long chats get expensive
Models are billed by tokens, and each new prompt is sent along with the conversation so far. In a long thread, your short follow-up question travels with everything that came before it. The tenth question in a conversation costs more than the first, even if both are one line long.
This is why splitting work into branches saves money as well as attention. A branch carries only the messages above it, so two separate lines of analysis each pay for their own history instead of each other’s.
Match the model to the question
The strongest models are priced for hard problems. Reformatting a table, shortening a paragraph or answering a factual follow-up does not need one. Switching to an affordable model for that kind of work, and keeping the expensive one for the reasoning that needs it, is the single biggest lever most people have.
In Bonscape you can change models mid-conversation and the context comes with you, so the switch costs nothing in setup.
See the number while you work
A monthly total tells you that you overspent, not where. Bonscape shows cost, time spent, branches and prompts at the bottom of each project, so you can tell which piece of work used what while there is still time to change how you do it.
Subscriptions, credits and your own keys
Paying for several AI subscriptions to reach different models is a common way to overspend. A single plan with access to many models removes that. If you already pay a provider, you can go further and connect your own API keys or your existing Claude or ChatGPT subscription. Plans and what they include are on the pricing page.