← Voltar aos guias

Codex pricing on 59API: what happens to unused balance?

Why a monthly subscription often wastes your budget

If you are searching for Codex pricing, you have probably already felt the awkward part of AI API billing: you pay for a month, and then you pay again for the days you did not use. A subscription is simple on paper, but it treats every month as identical. Your workload is not. A week of heavy agent runs is not the same as a quiet fortnight of small edits, and most developers end up paying for capacity they never touch.

59API takes a different route. It is an independent third-party AI API relay where you keep a prepaid balance, and each request is charged according to the real token usage it produces. There is no monthly fee. Instead of renewing a plan, you top up only when you need more, and the balance you have already paid for does not expire. Nothing quietly drains away while a project is paused.

That matters for the practical question behind most pricing comparisons: what happens to money you have already committed? With a wallet balance, the answer is simple. It stays there. You decide later whether to spend it on a larger model, a cheaper model, or a burst of requests during a busy release week.

The rest of this guide covers the balance in detail, including the one place where the rule changes: optional cards.

Pay-as-you-go on 59API: the balance never expires

With 59API you get one account, one API key and one API host, https://api.59api.com. GPT, DeepSeek and other non-Claude, non-Gemini models are reachable through that same setup, and billing follows real token usage per request. Prices are quoted per 1 million tokens, with billing in CNY and USD figures shown at ¥7.2 = $1.

Top-ups are deliberately small-grained: ¥10, ¥50, ¥100, ¥200, ¥500 and ¥1000, paid by WeChat Pay or bank card. Invoices are available. If you are comparing Codex pricing against a subscription, this is the core difference. A subscription asks for a recurring commitment; 59API asks for a small, one-off purchase that stays available until you spend it.

The per-model rates make the balance easy to reason about. GPT-6.1 Sol is $0.1847 (¥1.33) for input, $0.9264 (¥6.67) for output and $0.0375 (¥0.27) for a cache read, per 1 million tokens. GPT-5.6 Sol is $0.3472 (¥2.5) input, $1.94 (¥14) output and $0.0278 (¥0.2) cache read. GPT-6 Astra is $0.6944 (¥5) input and $3.47 (¥25) output.

To see the shape of a real month, consider a hypothetical run of 5M input tokens plus 1M output tokens without cache on GPT-6.1 Sol. That costs $1.85 (¥13.32) on 59API. At the official list price of $20 (about ¥144), 59API is about 9% of list price. Unspent funds simply remain in the balance.

Cards are a separate bucket, and their quota does not roll over

59API also offers optional monthly and weekly cards, and this is where the unused-money question changes. A card has its own key, can only call that model family, and is paid only from the card. It never falls back to wallet balance, and a balance key never draws on a card. The two pools stay completely separate.

The second difference matters just as much: unused quota does not roll over. A card is a 30-day or 7-day commitment, and whatever quota remains when it expires is gone. So if your goal is to avoid losing money to inactivity, the balance is the instrument built for it, and a card is a separate choice for predictable, heavy usage.

For example, the G-PRO 入门版 card for GPT runs 30 days at $83.19 (¥599), with quota $3250 and a weekly cap $1250. Its reference value at official list prices is about $902.78 (¥6500). The G-PRO 标准版 is $160.42 (¥1155) for 30 days, with quota $6500 and a weekly cap $2500.

There is also a shorter option: 【周卡】G-PRO $1500 for GPT runs 7 days at $40.56 (¥292), with quota $1500 and a weekly cap $1500. One important note: card quota and caps are metering units written in $, not official list-price dollars, so they should not be converted into official prices.

Balance or card: choose based on how predictable your usage is

The choice is not really about which option has a nicer headline number. It is about how you behave. If your usage is spiky and uncertain, a prepaid balance on 59API is the safer fit, because unused balance never expires and there is no monthly fee. You can start with a ¥10 top-up, watch how a week of Codex runs actually consumes tokens, and top up again when the balance gets low.

If your usage is steady and heavy enough that you already know how much you will call each month, a card can be a more structured way to pay, with a defined quota and a weekly cap that keeps the metering predictable across a 30-day window.

Numbers make the balance case concrete. A hypothetical month of 2M input tokens plus 0.5M output tokens without cache on GPT-5.6 Sol costs $1.67 (¥12) on 59API. At the official list price of $18 (about ¥129.6), 59API is about 9% of list price. Multiply that by a few projects and the money you have already committed becomes a working buffer rather than a deadline.

So the honest summary is this. Balance gives you durability: money stays until you spend it. Cards give you structure: a fixed quota, a weekly cap and a defined term, with no rollover. Many developers keep the balance as their default and reach for a card only when a predictable burst justifies it.

Your safety net: refunds on unused top-up balance

Prepaid money can feel risky if you have never used an API relay before, so it is worth knowing what happens if you top up and change your mind. On 59API, 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.

That window is worth planning around. If you are evaluating 59API for the first time, keep your early top-ups modest, run a few real Codex requests, and check the balance movement before committing to a larger amount. The point of the refund policy is not to make every decision reversible, but to give you room to test the setup properly.

Two practical notes. First, refunds are described for top-ups and unused balance, so do not assume a card purchase behaves the same way; a card is a separate product with its own key, its own family restriction and its own no-rollover rule. Second, unused balance that simply sits in your account is not at risk from expiry, because balance never expires. The refund route exists for a different situation: a recent top-up that you no longer want to hold.

For developers comparing Codex pricing, this combination is unusually straightforward. No monthly fee, no expiring wallet, per-request billing, invoices available, and a defined refund window for recent unused top-ups.

Connect Codex to 59API and start with a small balance

Getting started is short. For Codex, the OpenAI SDKs, cURL and OpenAI-compatible clients such as Cursor, Cherry Studio, Cline and OpenCode, point at base_url https://api.59api.com/v1 and use the chat endpoint https://api.59api.com/v1/chat/completions. One account, one API key, one host. If you are connecting Claude models instead, they work only through the genuine Claude Code or Claude Desktop client, version 2.1.285 or newer, never through Codex, cURL or other clients.

Once the key works, make your first top-up small. ¥10 is enough to prove the connection and see how a few requests move the balance. Then read the per-request charges against the model rates, and watch the balance rather than a calendar date. That is the whole pay-as-you-go idea: you are not renting access to a fixed monthly allowance, you are holding a prepaid balance that never expires and spending it only when your code actually calls.

If a burst of predictable usage arrives, look at the optional cards and their quotas and weekly caps. If you would rather keep the flexibility, stay on the balance. Either way, you know exactly what happens to the money you have not used, and you can try 59API on terms that fit how you actually ship.

Docs