Credits & Billing

How charging works

• Every inference request is priced from the model's configured sell rates — typically input, output, cache-read and cache-write per million tokens.
• A model with no explicit price falls back to upstream cost × (1 + markup), with an optional minimum charge per request.
• Cost is written to the ledger with the model id, token counts and a reason string, so any balance can be traced to individual calls.
• Requests are refused with 402 insufficient_credits before they reach a provider, so an empty balance never produces upstream spend.

Your current balance, total spend and itemised ledger are available in Dashboard → Usage & Analytics. Because the account API is private, do not build scripts or integrations that depend on its routes.

Paid and free balance

Top up your paid balance with crypto without buying a plan. Paid credits can cover metered models. Daily rewards, referral rewards and free balance codes are restricted to models classified as free; they cannot pay for paid models, plans or cash withdrawals. Code redemption identifies whether the grant is paid or free.

Keeping spend predictable

• Watch Usage & Analytics for spend per model over the last 30 days.
• Set a spend cap on experimental keys before letting an agent loop run.
• Send max_tokens on every request so a runaway completion has a hard ceiling.
• Cache deterministic answers in your own layer — the cheapest token is the one you never send.