Pricing and billing

What you pay for#

Every price is on Pricing, with a calculator for a month's use.

Box time and box fees#

A box's running time costs its hourly price divided into seconds: a 90-second run is charged 90 seconds, a 20-second run is charged the one-minute minimum. A stopped or sleeping box costs nothing to run. Box time is usage, so the plan allowance pays it first, then your balance.

Your boxes run in the location nearest you, automatically, and are billed at that location's price; the prices Pricing shows you are the ones you pay. A persistent box adds a flat monthly fee with storage included, and an always-on box is one flat monthly price. Monthly box fees are taken from your balance.

How a request is charged#

Each call costs its input tokens times the model's input price, plus its output tokens times the output price (prices are per million tokens, on Models), rounded up to the next millionth of a dollar.

deepseek-v4.1-flash: $0.20 / 1M input, $0.60 / 1M output1,200 input tokens  × $0.20 / 1M = $0.000240  350 output tokens × $0.60 / 1M = $0.000210                           total = $0.000450

• The token counts are the usage in the response (prompt_tokens and completion_tokens, or input_tokens and output_tokens).
• Reasoning tokens are output tokens. Images count as input tokens, as the model reports them.
• Cached input is reported in prompt_tokens_details.cached_tokens and billed at the model's cached-input price, shown as pricing.cached_input in /v1/models (the normal input price where that's empty).
• Streams are charged when they end. Stop one early and you pay only for what was generated; if the model never sent a final count, it's estimated from the text (about four characters a token).
• Refused and failed calls (any 4xx or 5xx before the model answers) cost nothing.

Plan allowance and balance#

Every account has a plan. Each plan includes usage every month, spent on AI tokens and box time at the usual prices:

1. The plan's monthly allowance pays first. It resets each plan month and doesn't roll over.
2. Your prepaid balance pays the rest. It never expires.
3. When both are empty, API requests are refused with 402 insufficient_quota and new box time can't start until you add credit. Admission is checked before each call, so the last call can take the balance a few cents below zero; your next top-up covers it.

Compare plans on Pricing and change yours under Plan. If a card plan's renewal fails, its allowance pauses until the payment goes through; your balance keeps working.

Adding credit#

Auto top-up and alerts#

• Auto top-up: save a card, then pick a threshold (up to $500) and an amount ($5 to $1,000). When a request takes the balance under the threshold, the card is charged and the credit added, so an agent never stops on a 402.
• Low-balance email: pick a level under Balance and payments and you're warned once when the balance drops under it.
• Caps: monthly spend caps per key and budgets per project; see Spending limits.

Usage and receipts#

Usage breaks spend down by day, key, project and model, with a CSV download. Logs lists every call with its tokens and cost. Balance and payments has a PDF receipt for every payment, with your company name, address and VAT number on it.

Request logging#