API keys and chat subscriptions are not the same thing

The most common question from people setting up a self-hosted workspace is some version of: "I already pay for ChatGPT. Why do I need a key?" The answer is that providers sell two separate products, and only one of them can be used from other software.

Two products with one brand name

A chat subscription is a flat monthly fee for using the provider's own app. It comes with usage caps that the provider manages, and it is tied to their interface.

An API account is a developer product. You create a key in a console, software sends requests with that key, and you pay for what those requests consume. There is no monthly fee and no app.

The two are billed separately even when they share a login. Paying for one gives you nothing on the other side.

How API billing works

API usage is measured in tokens, which are pieces of words. A rough guide for English is that 1,000 tokens is about 750 words. Each request is charged for:

• Input tokens: everything sent to the model. In a chat that means the whole conversation so far, not just your latest message.
• Output tokens: what the model writes back. These usually cost several times more per token than input.

That first point surprises people. A long conversation resends its history with every message, so the tenth message costs more than the first even if you type the same amount. Starting a new chat for a new topic is the simplest way to keep costs down.

What does it cost in practice?

It depends heavily on the model. Small, fast models cost a fraction of a cent per exchange. The largest reasoning models can cost many cents for one long answer, more with large attachments. For light personal use of mid-range models, many people spend less than a subscription would cost. Heavy daily use of top models can cost more.

The honest answer is to try it with a low limit and look at the numbers. Max Harness shows token counts for each reply, and a cost estimate when a reliable price is available.

Set limits before you connect a key

Every major provider lets you cap spending. Do this before pasting a key into any application, this one included.

1. Use prepaid credit where it is offered. When the balance runs out, requests stop.
2. Set a monthly budget or hard limit in the billing settings.
3. Create a separate key for each application. If something goes wrong you can revoke one key without disturbing the rest.
4. Turn on usage alerts by email.

The exceptions

A few subscriptions do include API access, usually ones aimed at coding tools. ClinePass, Command Code and OpenCode Zen each publish an API that works with a key from their dashboard, and Max Harness has presets for them. Their plan limits still apply, and some restrict which models the subscription covers.

Where a gateway fits

Services such as OpenRouter sit in front of many providers and give you one key and one bill. That is convenient if you want to try many models without opening an account with each lab. The trade-off is that your requests pass through one more company. Direct keys keep the path as short as possible. Both approaches work in Max Harness, and you can mix them.

If you are starting from nothing, one direct key for the model family you use most, plus a gateway key for everything else, covers nearly every case.