You prepay a balance and spend it per token, with no subscription. Each model is priced per million input, cached, and output tokens. Open a model on the models page for its benchmarks and the one-line switch.
Refusals removed from the weights.
| Details | |||||||
|---|---|---|---|---|---|---|---|
| $1.00 | $2.00 | $0.15 | 1M | 85 tok/s | |||
| $0.12 | $0.50 | $0.025 | 1M | 120 tok/s | |||
| $0.12 | $0.45 | $0.025 | 1M | 155 tok/s |
The vendor's original weights, unmodified.
| Details | |||||||
|---|---|---|---|---|---|---|---|
| $0.01 | $0.08 | $0.002 | 1M | 100 tok/s | |||
| $0.02 | $0.08 | $0.002 | 1M | — | |||
| $0.25 | $2.50 | $0.15 | 1M | 48 tok/s |
Pay as you go. Cards and crypto accepted for top-ups.
You prepay a balance and spend it per token. There is no subscription and no monthly minimum. Input, cached input, and output are each priced per million tokens, at the rates in the table.
A lower rate for input tokens the model can reuse from an earlier request with the same prefix. A stable system prompt in front of a long conversation is billed at this rate after the first call.
Your concurrency limit rises as your total deposits grow. A request over the limit returns HTTP 429; retry when one of your running requests finishes.
A card top-up can be refunded if you have used less than 5% of the credit it added. Email [email protected] from your account address. Crypto payments are non-refundable.
Visa, Mastercard, BTC, ETH, USDT, USDC, and XMR.
