> Credits
What credits are
Ocai work is measured in credits. Every model call spends some, and the number depends on how much work happened: a quick question costs a fraction of what a long run over a folder of files costs.
Your balance lives in your Wallet, alongside everything you have bought and everything you have spent.
The three kinds of token
Credits are counted in tokens, and there are three kinds. Knowing which is which makes a surprising number explainable.
- Input — what you send. Your message, the files you attached, the conversation so far, and the instructions Ocai is working under.
- Cache read — context the system has already processed. When the same material comes round again, most of it is read from a cache instead of being processed anew, which costs much less than fresh input.
- Output — what Ocai writes back. The answer you see, and the thinking that produced it.
Input usually dominates a normal exchange. Output grows with the length of the answer, and with how much thinking a step did.
What makes a task expensive
- Long context. A big attachment or a long conversation is sent as input on every turn.
- Many steps. Agentic work multiplies: each step carries its own context, its own thinking, and its own output.
- High thinking effort. More effort inside a step means more output tokens.
- Images. Images are priced differently from text, as a per-image cost.
The cache is what keeps a long conversation from becoming expensive every turn. Much of the earlier context is read rather than reprocessed, which is why the tenth message in a chat does not cost ten times what the first one did.
Checking what you have
- Open your Wallet to see your balance, what you have bought, and what you have spent.
- Read the usage on a conversation to see where a particular chat stands.
Buying more
Add credits from your Wallet, the same place you check the balance. If you are on a subscription, your allowance renews on its own; credits bought on top sit alongside it.
Questions people ask
Why did one task cost more than another?
Longer context, more steps, and higher thinking effort all add up. A one-line question and a run over a folder of files are not the same amount of work.
Why is the tenth message in a chat cheaper than the first?
It usually is not, but it is a fraction of the cost. Most of the earlier conversation comes back as a cache read rather than fresh input.
Do I pay for thinking?
Thinking is part of the output a step produces, so it counts. Lower thinking effort is the most direct way to spend less on work that does not need it.
What happens when I run out?
Check the Wallet before a long autonomous run. If you are about to hand over a big job, knowing your balance first is cheaper than finding out midway.