Docs
Pricing and units
You pay per unit, counted on our servers. See the current price list.
| Unit | How it's counted |
|---|---|
Audio second (audio_sec) | Duration of your file after we convert it to 16 kHz mono, measured by our metering service. Chunk overlaps are not billed to you. |
Token (token) | Counted with the model's own tokenizer on the text you send, before any device sees it. |
Image (image) | One per image sent (not yet offered). |
- The API reports usage in
usage.prompt_tokens(embeddings) andunitson jobs. Both come from the same meter that pays contributors. - Device-reported counts are never used for billing.
- Balances are prepaid in USDC (minimum top-up $100). Charges are reserved when a job starts and settled when it's delivered. Cancelled or failed work isn't charged.
- Prices in this market fall about 10x a year at fixed capability. We change prices with notice and publish the price-sheet version.
How prices and amounts appear in the API
- Prices are published per quantum, not per single unit, so small prices are exact:
GET /v1/pricingreturnsquantum_units(3600 for audio seconds, so per audio hour; 1,000,000 for tokens; 1,000 for images) andusd_per_quantum. - Every dollar amount (prices, balances, charges) is a decimal string in USD with up to 9 decimal places, for example
"0.080000000". Parse it with a decimal type, not a float. - A charge is
units × usd_per_quantum ÷ quantum_units, rounded up to the nearest $0.000000001. A 7-token request costs a fraction of a cent, not a whole cent.
{ "model_id": "whisper-large-v3-turbo", "unit_type": "audio_sec", "quantum_units": 3600, "usd_per_quantum": "0.080000000", "min_prepay_usd": "100.000000000" }