Gemini 4 Argon API relay pricing with metered tokens on Quanta

01 · What prepaid, metered spend changes for a finance team

Finance teams rarely have a problem with AI spend they cannot explain. They have a problem with spend that arrives as a single line item, adjusted by factors nobody can reconstruct a month later. On Quanta, the alternative is straightforward: you fund a prepaid balance, calls are metered on real tokens, and the dashboard records the tokens and the charge for every call. There is no monthly fee and no minimum spend, so the commitment is the top-up you choose, not a contract with a floor.

Top-ups on Quanta start at $1.39 and step through $6.94, $13.89, $27.78, $69.44 and $138.89. A team can begin small with $0.50 in trial credit on sign-up, test the workflow against real documents, and then size its next top-up against observed usage rather than a forecast. Unused balance can be refunded within 15 days of the top-up, which keeps an evaluation budget recoverable if the evaluation does not proceed.

Referral economics are visible too: 19% of a referred account's top-ups, and 5% one level further, can be withdrawn from $27.78 or moved into the balance. For a finance lead, that is a rule that can be written into a policy note, not an exception that needs a ticket.

02 · How Quanta's token-level charges add up

Quanta meters each call on real tokens. Input, output, cache writes and cache reads are priced separately, so a finance team can see which component is driving the bill instead of treating the total as one blended rate. The dashboard lists the tokens and the charge of every call, which turns a month-end review into a filtering exercise rather than an investigation.

Prices are quoted in $ as approximate USD figures, and charges are made in CNY. That distinction is stated plainly so the treasury team can reconcile the two figures against the invoice instead of assuming they match.

Consider Gemini 4 Argon, priced on Quanta at $1.72 per 1M input tokens, $8.26 per 1M output tokens, $1.72 per 1M cache write tokens and $0.0861 per 1M cache read tokens. A team that caches heavily will see the cache read line behave very differently from the input line, and the dashboard lets that split be checked call by call rather than inferred at the end of the month.

Quanta also publishes worked examples at a 3:1 input-to-output ratio with no cache hits. Gemini 2.5 Flash is shown at about $6.49 for 10 million tokens a month and $64.93 for 100 million. GPT-6.1 Sol is shown at about $4.10 for 10 million tokens and $40.97 for 100 million. Claude Opus 5.5 is shown at about $34.58 for 10 million tokens and $345.83 for 100 million. These are examples, not quotations, but they give a finance team a defensible starting shape for a budget conversation before any real traffic is sent.

03 · Which operating limits finance can plan around

Heavy, sustained workloads are where undocumented constraints hurt. Quanta publishes and enforces the limits that matter: 120 requests per minute per key, 8 requests in flight per account, 25 MB per request body and 20 live keys per account.

Each of those numbers becomes a planning assumption. Concurrency of 8 requests in flight per account tells a finance lead what a single account can absorb at once; 20 live keys per account defines how many separate workloads can run under one balance; 120 requests per minute per key sets the pace any one integration can sustain; and the 25 MB request body limit marks the practical ceiling for a single payload. When these are published, capacity planning stops being guesswork.

Streams also behave predictably. They have no total-time cap and only end after 300 seconds of upstream silence, so a long-running generation is not cut off by a wall-clock rule that nobody can anticipate.

04 · How company payment and invoices work on Quanta

Quanta supports company payment by bank transfer from a company account. Invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard, so procurement paperwork follows a defined route and leaves a record inside the same place where usage is tracked.

Because Quanta is pay-as-you-go only, with no monthly fee and no minimum spend, the invoice follows the top-ups you make and the tokens your teams actually consume. There is no recurring subscription line to reconcile and no unspent committed amount sitting on the balance. A budget owner can match a top-up to a period, and a reviewer can match that period to the calls recorded in the dashboard.

Data retention is stated in the same practical terms. The text users send is kept for 30 days for abuse handling and billing disputes, then deleted; model answers are not stored. A finance or compliance reviewer can be told exactly what is retained, for how long, and why, without having to infer it from policy language.

05 · How to set up Gemini 4 Argon and other models on Quanta

Quanta exposes three protocols: OpenAI-compatible (/v1), Anthropic native and Gemini native, with one key and one balance across all three. Gemini 4 Argon runs over the native protocol through Gemini CLI, and Gemini models are not served over the OpenAI-compatible path.

Claude models work only from the official Claude Code and Claude Desktop clients, with Claude Code 2.1.285 or newer. Codex, Cursor, Cherry Studio, Cline, OpenCode, OpenAI SDKs and cURL pair only with GPT, DeepSeek and the other non-Claude, non-Gemini models available on Quanta, including GPT-6.1 Sol, Qwen3.8 Max, Grok 4.7 and Kimi K3.

For a finance team, the practical point is that protocol choice is a routing decision, not a separate billing decision. One key, one balance and one dashboard cover the OpenAI-compatible path, the Anthropic native path and the Gemini native path, so a team running a mix of models still reports against a single prepaid balance and a single list of metered calls.

The sensible sequence is short. Open an account on Quanta, add a modest top-up, run a representative workload, and read the first entries in the dashboard, where each call shows its tokens and its charge. Adjust the top-up size against what the dashboard actually shows, then scale into the published limits of 120 requests per minute per key and 8 requests in flight per account. For teams with heavy, sustained workloads, that is a short distance between a budget question and a defensible answer.