Pricing per million tokens
Pay per token with no minimums and no commitments. Prices are listed here and in our OpenRouter listing before a model goes live.
| Model | Context | Input $/1M | Output $/1M | Cache $/1M | tok/s | Quant |
|---|---|---|---|---|---|---|
Gemma 4 26B A4Bgoogle/gemma-4-26b-a4b-it | — | — | — | — | — | FP8 |
Qwen3.8 27Bqwen/qwen3.8-27b | — | — | — | — | — | FP8 |
Prices, served context and median tok/s (last 7 days) are published before launch.
Prices are not set yet
We set them from measured cost and speed on our own node, after the first test run.
How billing works
Three rates per model
Input, cached input and output tokens. Cached input is prompt text we have just processed, such as a repeated system prompt or chat history; it costs us less to serve, so it costs you less.
Billed through OpenRouter
OpenRouter charges your account balance at the price in our listing. Every response reports its token usage, including cached tokens.
Nothing for failures
Errors and rejected (429) requests are not billed. A stream you cancel is billed only for the tokens you received. Price changes are announced on this page before they take effect.