Pricing
You buy credit and spend it per token. There's no subscription, and a request that credit can't cover doesn't start.
Token prices
| Model | Input | Cached input | Output | Minimum per request |
|---|---|---|---|---|
| qwen2.5-0.5b-abliterated | $0.25 | $0.025 | $1.00 | $0.0002 |
| qwen3-1.7b-abliterated | $0.25 | $0.025 | $2.00 | $0.00025 |
Prices are US dollars per million tokens. Repeated prompt prefixes are cached automatically and kept separate for each account.
Two fees when a GPU isn't busy
GPU workers start on demand and stop when idle. These cover the time a worker isn't running your tokens.
Start fee
$0.04
Paid once, by the request that starts a worker. That request waits a few minutes for it.
Idle fee
$0.0003 / second
Paid by a request that reuses a worker which sat idle, for up to 60 seconds. Gaps under a second are free, so it falls to nothing under steady traffic.
Both appear on every response, in X-Cold-Start-Fee-USD and X-Idle-Fee-USD. Waiting in the queue is free and failed requests cost nothing.
What a request costs
qwen2.5-0.5b-abliterated costs $0.000325 in tokens. If it's the request that starts a worker, it also pays the $0.04 start fee.Prepaid credit
Add from $5 to $1,000 at a time.
- Before a request runs, the most it could cost is reserved. It then pays what it used, and the rest returns straight away. Setting
max_tokensreserves less. - Auto-reload starts on: below $2 of credit it adds $10, at most $25 a month. You can change all three or turn it off.
Monthly plans, if you use it regularly
Optional. The fee becomes credit, and token usage past the fee in a calendar month is discounted. Start and idle fees aren't discounted.
Starter
$20 / month
$20 of credit each month, and 2.5% off tokens past that.
Growth
$50 / month
$50 of credit each month, and 5% off tokens past that.
Scale
$200 / month
$200 of credit each month, and 10% off tokens past that.