Credit
The dollar balance that pays for requests. Add it with USDC or INFERENCE on Home.
Credit is a dollar balance on your Deference account. Every request spends it. Free credit is a separate balance for open-weight text models, spent before your credit.
USDC in · INFERENCE out
USDC buys INFERENCE
The selected purchase route delivers INFERENCE. Its quote includes the purchase fees.
Burn tokens into credit
Activation burns INFERENCE. The amount after the activation fee becomes service credit that cannot be sent, sold or withdrawn.
Hold, then charge
A request holds its prompt plus max_tokens. When it ends, the hold is released and the measured cost is charged.
Add credit
Open Add credit, choose USDC or INFERENCE, and enter an amount. Review the estimated credit before confirming.
| Pay with | What happens |
|---|---|
| USDC | The selected route buys INFERENCE, then activates it. The quote estimates credit after purchase and activation fees |
| INFERENCE | The INFERENCE is burned and the amount after its activation fee becomes service credit |
Eligible launch token holders can also claim their rewards straight into credit.
Confirmed activation receipts determine the credit added. A purchase's minimum INFERENCE output applies before activation fees. See Order book for quote details.
Credit stays on your account. It cannot be sent, sold, withdrawn or turned back into INFERENCE. If you may want to sell later, keep INFERENCE and activate it when you need credit.
How a request spends it
- When a request starts, Deference holds an estimated cost for the prompt and the tokens the model's chat template adds, reserved output, and any per-request price. Server tools add to the hold. A file, or an image or video given as a URL, is held at the larger of the text estimate and the model's context window, or 1,000,000 tokens when that window is unknown. Held credit shows beside your balance.
- When the response ends, the hold is released and the real cost is charged. A non-streamed timeout keeps the hold reserved until the cost is reviewed.
- A few seconds later the cost is checked against the provider's record, and Activity shows the final amount.
Your available credit is your balance minus what running requests hold. A lower max_tokens holds less. The hold is an estimate, so a request can cost more than its hold. The difference is charged from the rest of your credit balance, within the key's limit, and no request takes your credit balance below zero. Pricing shows how each cost is worked out.
When a request is refused
A request is refused when the credit it must hold does not fit in your available credit or in its key's remaining limit. Chat Completions and legacy Completions that omit an output cap and use no server tools can still run: Deference lowers the cap it sends, down to the prompt plus one output token. A cap you set is kept. A request that free credit pays for can run on a smaller hold.
| Error | Cause | Fix |
|---|---|---|
402 insufficient_credit | Not enough available credit for the hold | Add credit, or lower max_tokens |
402 free_ | Only free credit is left, and this model or request does not qualify | Use an open-weight text model, lower max_tokens, or add credit |
402 key_limit_reached | The key has less left than the hold | Raise the key's limit in API keys, or use another key |
Requests already running finish.
Limit what a key can spend
Give each key its own credit limit and expiry, so one tool cannot spend your whole balance. See Keys and security.