Gemini 3.8 Flash API relay pricing and 120 RPM rate limits

01 · What happens to API calls when the Quanta balance reaches zero

Quanta is pay-as-you-go: a prepaid balance, no monthly fee and no minimum spend. Every call is metered on real tokens, and the dashboard lists the tokens and the charge for each individual call. So when the balance runs out, the answer is direct. There is no credit line behind the prepaid balance, and the next call is not funded until the account is topped up. Nothing is estimated from a seat count or an assumed monthly volume, because usage is what gets charged.

That makes running out of balance a funding gap rather than a billing surprise. A team can look at the dashboard, see which calls consumed the balance, and decide how much to add. New accounts receive $0.50 in trial credit on sign-up, and top-ups start at $1.39, so checking the relay with a small amount and then funding a sustained workload are both modest steps.

02 · Why a balance runs out: token-level metering on Quanta

Quanta prices input, output, cache writes and cache reads separately, per 1M tokens. For Gemini 3.8 Flash the listed figures are $1.35 input, $6.10 output, $1.35 cache write and $0.1347 cache read. That separation is what makes balance planning possible. A run that reads a large context once and then works mostly on answers behaves very differently from a run that writes and reads cache repeatedly, and the two show up as different lines in the account record.

Examples make the arithmetic concrete. At a 3:1 input-to-output ratio with no cache hits, Gemini 2.5 Flash costs about $6.49 for 10 million tokens a month and $64.93 for 100 million. Claude Sonnet 5.5 comes to about $25.07 and $250.69 on the same example, Claude Opus 5.5 to about $34.58 and $345.83, and GPT-6.1 Sol to about $4.10 and $40.97. These are examples, not quotations, and they show why a team should read the dashboard rather than estimate from intuition.

Quanta lists these prices in $ as approximate USD figures, and charges are made in CNY. For teams running heavy, sustained workloads, that is the practical point: the balance tracks measured tokens, so topping up is a decision made against observed consumption instead of a forecast built in the dark.

03 · Gemini 3.8 Flash, Claude models and one shared balance

Gemini 3.8 Flash is served over the native protocol through Gemini CLI, and that is the supported pairing for it. Quanta also speaks the OpenAI-compatible protocol on /v1 and Anthropic native, and one key and one balance cover all three protocols. 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 with GPT, DeepSeek and the other non-Claude, non-Gemini models.

For a team running several supported tools, the shared balance matters when it gets low. Usage lands in the same account record whichever route produced it, and the dashboard still breaks the account down call by call, with tokens and charge shown for each one.

Quanta keeps the text users send for 30 days for abuse handling and billing disputes, then deletes it. Model answers are not stored. When a team reviews a charge after the balance has run down, the per-call token record is the place to look, and no monthly reconciliation is required because there is no monthly fee.

04 · Concurrency, rate limits and the speed at which a balance drains

Balance and capacity are related. Quanta publishes and enforces 120 requests per minute per key, 8 requests in flight per account, a 25 MB request body limit and 20 live keys per account. The in-flight figure is the one that matters most for a team working with a thin balance: parallel calls consume tokens at the same time, so a burst can use up a small balance faster than a sequential queue would.

Streams have no total-time cap, and they end only after 300 seconds of upstream silence. A long stream can therefore keep a call open while metered tokens keep accruing. With 20 live keys available per account and 8 requests in flight per account, a team can size its concurrency against the balance it has actually funded, rather than discovering the boundary in the middle of a run.

Because the limits are published and enforced, planning is arithmetic rather than guesswork. Multiply the request rate a workflow actually needs by the token cost per call, compare that with the funded balance, and top up before the gap appears.

05 · Refunds, top-up amounts and company invoices on Quanta

Quanta's top-up amounts are $1.39, $6.94, $13.89, $27.78, $69.44 and $138.89. A team that needs only a little capacity for the week is not paying for capacity it does not use, and a team with heavy, sustained workloads can fund the account in proportion to measured demand.

Unused balance can be refunded within 15 days of the top-up, which keeps a large prepaid commitment from becoming a stranded one. Referral rewards are 19% of a referred account's top-ups and 5% one level further; rewards can be withdrawn from $27.78 or moved into the balance.

For company accounts, bank transfer from a company account is available, and invoices and advance quotes are arranged through the Contact page or the Messages page of the dashboard. That is the route for procurement teams who need a document to accompany the spend, and it is handled in the same place the rest of the account is managed.

06 · A practical checklist before the balance runs out

Five steps keep a Quanta account funded and predictable:

• › Open an account and use the $0.50 trial credit on sign-up. Check token counts and per-call charges in the dashboard from the first request onward.
• › Price the actual mix of usage. Input, output, cache writes and cache reads are billed separately, so include Gemini 3.8 Flash over Gemini CLI and any Claude models used in Claude Code or Claude Desktop in the calculation.
• › Match request rate to the published limits. 120 requests per minute per key and 8 requests in flight per account are enforced, not advisory.
• › Fund the balance in one of the published top-up amounts. Note the 15-day refund window for unused balance before committing a larger amount.
• › Arrange company billing early. Bank transfer, invoices and advance quotes are handled through the Contact page or the Messages page of the dashboard.

When the balance does run out, the fix is straightforward: top up, and calls can be funded again on the same key and the same balance. Quanta's approach — a prepaid balance, no monthly fee, per-call token accounting and published rate limits — is built for teams with heavy, sustained workloads, and it makes the cost of a run something a team can read rather than guess. Open an account and check the dashboard on the first call.