fak · live paid inference

Every token is metered in dollars

Run a real inference request, inspect its settled bill, and compare the observed business with explicit serving scenarios. Model identity comes from server metadata and receipts.

Try the live demo

One prompt. A real response. A receipt with the actual charge.

Use your API key below to begin.

Price list

Model	Prompt / 1M	Cached prompt / 1M	Completion / 1M
`RadixArk/Qwen3.8-27B-NVFP4`	$0.20	$0.20	$0.60

Call POST /v1/chat/completions with Authorization: Bearer <your key>.

Your account

The key stays in this browser tab (session storage). It is sent only to this server.

No account loaded

Credit

Balance—

Spent—

Requests—

Prompt cache hit rate—billed at the cached price

Try it: one real, billed request

Requested model

Prices and available models: server configuration. Served model will appear after Send.

Send twice to inspect whether the server reports cached prompt tokens.

Unit economics: price vs. cost to serve

Token class	We charge / 1M	Est. cost / 1M	Markup

Fleet money · cumulative receipt history

All tenants, anonymous. Estimated serving spread excludes unknown business costs.

Cumulative receipt value—

Est. cost to serve—

Estimated serving spread—

Baseline · full utilization—

Cost envelope: amortized hardware and electricity. Upstream costs, failed attempts, fees and other operations are unknown or excluded; this is not net profit.

Actual rolling 24 hours

Waiting for settled rolling-window evidence.

Real settled revenue is separate from full-utilization projections.

What if?

Compare an explicit scenario with the server baseline. Estimates use a 730-hour month and 8,760-hour year; they are not earned revenue.

Utilization 100%Prompt cache hits 0%Price multiplier

Projection	Baseline	Scenario	Change

Awaiting a verified financial baseline. Controls do not change actual bills or pricing.

Your recent requests

Time	Fresh in	Cached in	Out	Billed	Est. cost	Est. spread %