Pricing and billing
Token Spark is pay-as-you-go: you top up credit and are billed per call. The prices below come from live listings on the Platform and are visible without an account.
How billing works
- Credit first, then usage: credit is deducted in real time per call, and no charge arises without a top-up.
- Text and multimodal models are billed by tokens — input tokens consumed by your prompts and output tokens generated by the model, priced per million (1M) tokens.
- Image models are billed per generated image, each converted into a fixed token equivalent, so all prices below use the per-million-tokens unit.
- We do not sell auto-renewing subscriptions and never charge your payment method on a recurring basis.
- An entirely unused top-up can be refunded in full within 7 calendar days; if a Platform-caused outage lasts more than 72 consecutive hours you may claim a proportional refund (see Section 7 of the Terms).
Amounts are in CNY. Cache hits, reasoning tokens and other special cases follow each model's own billing rules, and every invoice line is itemised.
Live price list
"Platform price" is the lowest / highest total (input + output combined, per million tokens) among resources on sale now; "Official guide price" is the provider's combined input + output price, and the discount is derived from it.
Routes are shown by anonymous route ID; supplier identity is not disclosed. Regions are self-declared by the supplier and spot-checked by the platform. Quality grades are computed from real calls over the last 7 days using success rate, TTFT, and TPS; routes with insufficient samples are marked "Observing".
| Model | Upstream provider | Type | Routes | Platform price | Official guide price | vs official |
|---|---|---|---|---|---|---|
Loading prices… | ||||||
Terms of Service · Platform Rules · Refund Policy · API Docs
Pricing and billing
Token Spark bills per token with no subscription fee. Every call is settled by actual token usage, and model, unit price and amount are itemized.
The price table on this page comes from live listings in the platform marketplace: for the same model, the lowest/highest total price is shown against the official reference price, with the discount calculated automatically. When there are no listings, the table says so instead of showing estimates.
Balance, keys, rate limits and budget caps are all managed in the console; invoice applications start from the billing page, and the refund policy follows the user agreement.
FAQ
How is usage charged?
Per token, with input and output priced separately. Actual unit prices are shown in real time in the console.
Is there a subscription fee?
No. Pay-as-you-go: top up and use what you need.
Why does the same model have multiple prices?
The same model may have listings from multiple suppliers, and prices fluctuate with upstream cost and supply; the platform routes across them by latency, availability and price.
Can I get an invoice?
Yes. The billing page in the console provides an invoice application entry; follow the guide to submit invoicing details.