Pricing
Prepaid credits. Pay only for the tokens you use.
Per-model prices
| Model | Id | Context | Precision | Input / Mtok | Cache read / Mtok | Cache write / Mtok | Output / Mtok | Status |
|---|---|---|---|---|---|---|---|---|
| Qwen3 0.6B | tokencannon/qwen3-0.6b | 32K | bfloat16 | $0.05 | $0.005 | $0.00 | $0.15 | active |
| Qwen3 1.7B | tokencannon/qwen3-1.7b | 32K | bfloat16 | $0.10 | $0.01 | $0.00 | $0.25 | active |
Prices are per million tokens, flat at any concurrency, and read live from GET /v1/models. Every request is billed on four dimensions: input, cache read, cache write, and output. Output includes any reasoning tokens; reasoning is not billed separately.
How caching is billed
Input is the uncached part of your prompt, billed once at the input price. Cache read is what the engine served from its prefix cache, at the cache-read price. Cache write is the prompt prefix the engine reports it wrote into its cache this request, priced at $0 under automatic prefix caching. Both counts come from the engine; if it does not report them, the whole prompt is billed as input and the receipt says so. Unknown is never rounded to zero. See Metering & receipts.
Buying credits
- $10
Try the API.
- $25
Build and test.
- $50
Run sustained traffic.
Buy any amount from $5 in the console; $10, $25 and $50 are suggestions, not tiers. Credits never expire.
Optional auto top-up: set a threshold and an amount in the console, and when a request leaves your balance below the threshold we charge the card from your last purchase for that amount. Off by default. Every purchase through Stripe Checkout saves the card, so you can turn it on later without buying again.
How billing works
Each request deducts its input, cache read, cache write and output token cost at the rates above, computed in exact integer micro-dollars and rounded down. Every request has a receipt you can recount yourself.
Requests stop when your balance reaches zero. Failed requests are not charged.