Pricing
One model, one flat per-token rate, prepaid. No subscriptions, no seats, no platform fee — you pay for tokens the engine actually reports.
Per-token rate
kwaipilot/kat-coder-v2.5-dev| Direction | Per 1M tokens |
|---|---|
| Input | $1.00 |
| Output | $4.00 |
Introductory — final prices publish when the launch performance gate passes. Rate changes apply from a dated announcement, never retroactively.
Minimum request fee
Each request carries a small dynamic minimum fee that covers cold starts: when your request has to wake capacity, the minimum absorbs the startup cost instead of hiding it in the token rate. It is sized from the measured p95 startup time of the fleet, shown before you top up, and applies only when the token charge would fall below it. If capacity fails to start, you are not charged at all.
Prepaid wallet
- · Top up from $20 via Stripe Checkout (presets $20 / $50 / $100 / $250).
- · Credits don't expire during the beta.
- · The balance never goes negative — a request that can't be covered is refused up front.
- · Capacity errors never debit: a capacity 429 costs you nothing.
- · No auto-charge without an explicit opt-in; top-ups are always deliberate.
What you are not buying
No subscription. No per-seat pricing. No platform or gateway fee. No paid tiers that change the model's behavior — the model is identical for every account; only the release ID on the response tells you what served it.
Later — Ready: reserved warm capacity with a monthly minimum, for teams that need cold starts off their latency path.
Later — Batch: discounted off-peak processing for workloads that can wait.
Pricing FAQ
How is usage metered?
From engine-reported token counts on your request — the same usage object you see in
the response. If the engine fails to report usage, the request is not charged: metering fails closed
in your favor, never ours.
What price applies to a request?
The price version in force when the request was accepted, recorded on the usage row itself. A published price change never rewrites history, and your usage export shows exactly which version each request settled under.
Refunds?
During the beta, contact support — we refund unused balance and sort out disputed charges by hand rather than hiding behind policy.