What procurement and operations ask before signing off
Quanta is a prepaid, independent AI API relay. These are the questions procurement and operations teams bring before signing off, answered with figures from the current rate card and settings.
Our overnight batch jobs run in parallel: what concurrency and rate limit does one Quanta account get per API key?
One account holds 8 requests in flight at a time, and each API key takes 120 requests a minute. A request past either limit gets HTTP 429, so batch workers should queue and retry rather than fan out further. A request body can reach 25 MB and an account can keep 20 live keys.
Finance needs a company invoice for API top-ups: how do we pay Quanta by bank transfer and get one?
Write to us on the Contact page with your username, and the invoice, an advance quote or a bank transfer from your company account is arranged there. The dashboard itemises the tokens and charge of every call, so month-end reconciliation runs from your own records.
Our own tiktoken count differs from the usage record: which token number does Quanta bill?
Quanta bills the usage the model provider reports for the call, not a count made on your side. Input, output, cache writes and cache reads are priced apart: GPT-6.1 Sol, for instance, costs $0.2083 for input and $1.01 for output per 1M tokens. If a stream ends with no reported usage, the charge is estimated and the record is marked as estimated.
A streamed GPT or Claude call returns an overloaded error before any output: is it still charged to the balance?
No, it costs $0.00: with an error before any answer and no final usage from the provider, the call is recorded with its error code and nothing is charged. Once part of the answer has been delivered, or the provider has reported final usage, the call is billed as usual. A client that hangs up mid-stream does not cancel the call and pays for the usage reported.
Is there a monthly commitment or minimum order before a team can pay as you go?
No, Quanta is pay-as-you-go from a prepaid balance, with no monthly fee and no minimum spend. Top-ups run in steps from $1.39 to $138.89. A new account starts with $0.50 of trial credit, which pays for calls but cannot be refunded or withdrawn.
A project was cancelled with budget left on the account: can unused API balance be refunded?
Yes, if you ask within 15 days of the top-up; what was consumed and the payment provider’s fees are deducted. Ask on the Contact page, and requests are handled within 3 working days. Older top-ups and trial or promotional credit are not refunded.
Our developers use Claude Code and Cursor side by side: can one Quanta key cover both?
One key and one balance cover both, but Claude models answer only in the official Claude Code and Claude Desktop (Claude Code 2.1.285 or newer). Cursor, Codex, Cherry Studio, Cline, OpenCode and OpenAI SDKs take GPT, DeepSeek and the other non-Gemini families at https://api.quantarelay.com/v1. Gemini models use the native protocol through Gemini CLI. Claude Code reads ANTHROPIC_BASE_URL=https://api.quantarelay.com with your key in ANTHROPIC_AUTH_TOKEN; Gemini CLI reads GOOGLE_GEMINI_BASE_URL=https://api.quantarelay.com.
Our security review asks what the relay stores: how long are prompts kept, and are model answers logged?
Prompt text is kept for 30 days for abuse handling and billing disputes, then deleted automatically, and model answers are not stored. System instructions are not kept either, and the rest of each raw request is discarded once the call completes. Your IP address and browser cookies are not passed to model providers.