← Back to all guides

Gemini CLI pay-as-you-go on 59API: what happens to unused balance

The problem with paying monthly for AI you might not use

If you are wiring an AI model into a script, a CLI tool or a weekend project, the awkward part is rarely the code. It is the bill. A monthly subscription asks you to guess how much you will use before you have written a single request. Use it lightly and you pay for quiet weeks you never asked for. Use it heavily and you hit a cap. That is the shape of the problem 59API was built around, and it is exactly what the topic of this article is about: what happens to the money you put in but do not spend.

59API (https://59api.com) is an independent third-party AI API relay. One account, one API key and one API host (https://api.59api.com) get you to Claude, GPT, Gemini, DeepSeek and other models. The billing model is prepaid balance, charged per request by real token usage. There is no monthly fee, and balance never expires. Top-ups start at ¥10, and you can also add ¥50, ¥100, ¥200, ¥500 or ¥1000 by WeChat Pay or bank card. Invoices are available. Monthly and weekly cards exist too, but they are optional, and they behave differently from balance. That difference is the point.

What unused balance means on 59API

Here is the short answer, and it is the one worth remembering: on 59API, unused balance simply stays yours. Because 59API is pay-as-you-go, a request deducts what it actually consumed in tokens and nothing more. If you top up ¥100 and only ¥12 of it goes out this month, the rest sits there until you need it. There is no monthly fee eating into it, no expiry date quietly draining it, and no renewal you have to remember to cancel. Balance never expires.

That changes how you can plan. Suppose a hypothetical side project needs a burst of calls one weekend and then goes quiet for two months. With a subscription you would either pay twice for the quiet stretch or downgrade and lose headroom. With prepaid balance on 59API, you top up, you spend, and the remainder waits. When the project wakes up again, it is still there. Billing is per 1 million tokens, in CNY, with USD figures shown at ¥7.2 = $1, so you can see what a request costs before you make it.

If you genuinely change your mind, 59API has a route for that too: unused balance can be refunded within 7 days of the top-up, minus what was used and payment-channel fees. Requests go through the Contact page and are handled within 3 business days.

How far prepaid balance stretches on real requests

Because you are charged per token, the natural question is how far a modest top-up goes. Prices are per 1 million tokens. Take Gemini 2.5 Flash (model id gemini-2.5-flash): input is $0.2083 (¥1.5), output is $1.67 (¥12) and cache read is $0.0208 (¥0.15). Or look at Gemini 3.1 Pro (model id gemini-3.1-pro): input is $1.39 (¥10), output is $8.06 (¥58) and cache read is $0.1389 (¥1). Small, exploratory calls therefore consume small amounts, which is what makes a ¥10 or ¥50 top-up a reasonable way to test the water before committing anything larger.

Worked examples make the scale concrete. Claude Haiku 4.5 with 5M input and 1M output tokens without cache comes to $3.4 (¥24.5) on 59API, against an official list price of $10 (about ¥72), so 59API is about 34% of list price. Claude Sonnet 5.5 with 3M input and 0.5M output tokens without cache is $6.11 (¥44) on 59API, versus $16.5 (about ¥118.8) at official list price, so about 37% of list price. Official list price is only a reference price, not a quote. The takeaway is that real usage drives the deduction, and what you do not use stays put.

Optional cards behave differently, and that is by design

Alongside pay-as-you-go balance, 59API offers optional monthly and weekly cards. A card has its own key, can only call that family's models, and is paid only from the card: it never falls back to wallet balance, and a balance key never draws on a card. In other words, the two are separate purses. You can run a card for a heavy stretch and keep your balance untouched, or run balance only and skip cards entirely.

Here is the trade-off to weigh. Unused quota does not roll over. Quota and caps are in official-price USD, and a card drains at official price × 2.4. Take the C-PRO Starter card (Claude, 30 days): $83.19 (¥599) for a quota of $1300 with a weekly cap of $500, which works out to about $541.67 of usage at official list prices. There is also a G-PRO Starter card (GPT, 30 days) at $83.19 (¥599) with a quota of $3250 and a weekly cap of $1250, about $1354.17 of usage at official list prices. If you finish early, the remainder is gone when the 30 days end. If your usage is spiky and unpredictable, that is exactly the property pay-as-you-go balance avoids.

Choosing between balance and cards in practice

So which should you pick? If you are learning, prototyping, or running something whose traffic you cannot forecast, start with prepaid balance. It is the pay-as-you-go path: no monthly fee, charged per request by real token usage, and balance never expires. Top up ¥10, try a few calls, and see what a real workload costs before scaling up. Nothing expires on you while you think.

Cards make sense when you know your workload in advance and intend to push it hard within a fixed window. A weekly card is the tighter version of that idea: the G-PRO weekly card (GPT, 7 days) is $40.56 (¥292) with a quota of $1500 and a weekly cap of $1500, about $625 of usage at official list prices. If you have a deadline and a heavy week planned, that can be a deliberate fit. If not, remember the rule: unused quota does not roll over.

Either way, 59API keeps the plumbing simple. One account, one API key, one API host at https://api.59api.com. For Gemini CLI, set GOOGLE_GEMINI_BASE_URL=https://api.59api.com and GEMINI_API_KEY=YOUR_59API_KEY. Gemini models use Google's native protocol only, so they are not OpenAI-compatible and do not work on /v1/chat/completions. Pick the billing shape that matches how you actually work.

What this means for your first top-up

The mental model to carry away is simple. On 59API, money you add as balance is money you keep. It is not a subscription that resets, not a coupon with an expiry date, and not a plan you have to optimise around. It is a prepaid meter that only moves when your code makes a request, charged per request by real token usage, with prices per 1 million tokens and no monthly fee. Balance never expires, and within 7 days of a top-up you can even request a refund of unused balance through the Contact page.

That is a friendly shape for the way developers actually work: bursty, exploratory, and rarely a perfect match for a fixed monthly quota. You top up when you have something to build, you spend what the work costs, and you keep the rest for next time. If your workload later becomes predictable and heavy, the optional cards are there as a different tool for a different job, with their own key and their own rules. Until then, pay-as-you-go is the low-commitment way to find out what a real workload on 59API costs you.

Start small, look at what a request actually deducts, and scale from there.

Docs