Pricing

$1 of SparkAPI credit = $1 of upstream API usage. We charge exactly what the upstream provider charges us — same numbers you'd see on the official rate card, every model, every time.

How credits work

When you top up, your balance is held as USD-denominated credits. Each API call deducts credits equal to the cost of the underlying request:

credits_charged = (input_tokens / 1,000,000) × input_rate
                + (output_tokens / 1,000,000) × output_rate

Per-model rates live on the Models page. Your running balance and per-call breakdown are in the dashboard.

Worked example

A typical chat call to claude-sonnet-4-6 with 5,000 input tokens and 2,000 output tokens costs:

Call: claude-sonnet-4-6 — 5k input, 2k output

Input: 5,000 tokens × $3.00 / 1M$0.01500

Output: 2,000 tokens × $15.00 / 1M$0.03000

Total per call$0.04500

Calls per $1 of credits≈ 22

Calls per $15 (the smallest plan)≈ 333

Most production chat workloads sit between 1,000–4,000 input tokens and 200–800 output tokens, so per-call cost is usually a few tenths of a cent.

💡 Enable prompt caching to run more requests per dollar. Repeated context blocks (long system prompts, RAG chunks, code files) can be cached and billed at ~10% of the input rate. For stable workloads this typically cuts the same job 60–80%. See the caching section.

Cheaper models for high-volume workloads

If you're sending tens of thousands of calls a day, switch to the cheaper tier. Pricing scales linearly:

Model	1M in	1M out	Per $1
`gpt-4o-mini`	$0.15	$0.60	~1,300 calls*
`claude-sonnet-4-6`	$0.80	$4.00	~285 calls*
`gemini-3-flash-preview`	$0.30	$1.20	~667 calls*

* Assumes 1k input + 200 output tokens per call (a typical short chat message).

Billing cycle

Prepaid — credits are deducted per call in real time. No monthly bills, no overage charges. Top up when you run low; your balance never expires.

Refunds

Within 7 days of purchase, unused credits can be refunded. See the refund policy for the full terms.