Claude Haiku 4.5 weekly cards on 59API: an API relay guide
Why pay-as-you-go beats a subscription for most developers
Most developers do not need another recurring bill. They need a way to pay for what they actually send and receive. A subscription looks tidy on paper, but it rewards inactivity: you pay for a month whether you ship one small script or run an agent loop every hour. Pay-as-you-go flips that logic. You top up a balance, and each request is charged by real token usage. There is no monthly fee, and the balance never expires.
59API supports that model directly. It is an independent third-party API relay where one account and one API key give you one API host for Claude, GPT, Gemini, DeepSeek and other models. Top-ups start at ¥10 and go up through ¥50, ¥100, ¥200, ¥500 and ¥1000, paid by WeChat Pay or bank card, with invoices available.
Alongside that balance, 59API offers optional monthly and weekly cards. A card is not a subscription in the usual sense. It is a prepaid, time-boxed allowance for a single model family, with its own key and its own metering rules. So the real decision is not cards versus pay-as-you-go in the abstract. It is whether your next seven days contain concentrated, predictable usage worth locking in, or whether ordinary balance top-ups fit more honestly. This guide covers the three weekly cards, who each one suits, and how to connect to 59API when you decide to try one.
What a weekly card actually is on 59API
A card is a separate credential with a separate allowance. It has its own key, it can call only that card family of models, and it is paid exclusively from the card. A card never falls back to wallet balance, and a balance key never draws on a card. That separation is deliberate: it makes your spending predictable and keeps a card from quietly draining prepaid funds you were saving for something else.
Three weekly cards are available, each valid for 7 days. The G-PRO $1500 card covers GPT models and costs $40.56 (¥292). The C-PRO $600 card covers Claude models and costs $40.14 (¥289). The C-MAX $600 card also covers Claude models and costs $77.64 (¥559). Both Claude cards carry a quota of $600 and a weekly cap of $600; the GPT card carries a quota of $1500 and a weekly cap of $1500.
Two details deserve attention before you buy. First, unused quota does not roll over, so a card that expires with allowance left simply ends. Second, quota and caps are written in card metering units, not official list-price dollars, so they should not be read as a promise of a specific dollar amount at official provider rates. Card balance drains at the card metering price multiplied by 2.4. The reference value at official list prices for the GPT card is about $416.67 (¥3000), and for each Claude card it is about $166.67 (¥1200).
Who should buy the weekly G-PRO $1500 card
The G-PRO $1500 card is aimed at developers who know exactly what they will run for the next seven days. If you have a scheduled batch, a week of agentic coding, or a predictable burst of GPT traffic, a dedicated key with a fixed weekly ceiling removes the mental overhead of checking a balance between calls. At $40.56 (¥292) for 7 days, it is a bounded commitment rather than an open-ended monthly bill.
Compare that with ordinary pay-as-you-go use. GPT-6.1 Sol is priced at $0.1847 (¥1.33) per 1 million input tokens and $0.9264 (¥6.67) per 1 million output tokens, with cache reads at $0.0375 (¥0.27). A workload of 5M input tokens plus 1M output tokens without cache costs $1.85 (¥13.32) on 59API, compared with an official list price of $20 (about ¥144), putting 59API at about 9% of list price.
That comparison is exactly why the card decision is not about rate. If your week is light, top-ups simply cost less in absolute terms and the balance carries forward. Buy the weekly G-PRO card when usage is front-loaded enough that you would rather set a ceiling, isolate the workload under its own key, and stop thinking about it. Keep using the balance when demand is uncertain, because on 59API that balance never expires and there is no monthly fee to justify spending it quickly.
Who should buy the weekly Claude cards
Claude users have two weekly options, and the choice comes down to card family and cap rather than price alone. The C-PRO $600 card costs $40.14 (¥289) and carries a quota of $600 with a weekly cap of $600. The C-MAX $600 card costs $77.64 (¥559) and also carries a quota of $600 and a weekly cap of $600. Both last 7 days, both are limited to Claude models, and both drain at the card metering price multiplied by 2.4.
Cost per request on 59API remains visible either way. Claude Haiku 4.5 is $0.3472 (¥2.5) per 1 million input tokens, $1.67 (¥12) per 1 million output tokens, and $0.0278 (¥0.2) for cache reads. A workload of 5M input tokens plus 1M output tokens without cache costs $3.4 (¥24.5) on 59API, against an official list price of $10 (about ¥72), so 59API is about 34% of list price.
One constraint shapes everything here: Claude models work only through the genuine Claude Code or Claude Desktop client, version 2.1.285 or newer. They do not work through the OpenAI SDK, cURL, Codex, Cursor, Cherry Studio, Cline, OpenCode or any other client. So a weekly Claude card suits developers who will spend the week inside Claude Code or Claude Desktop. If your workload runs through an OpenAI-compatible client, choose the weekly G-PRO card or stay on pay-as-you-go with GPT, DeepSeek and other non-Claude, non-Gemini models.
How to connect, and how to make the final call
Setting up 59API takes one account, one API key and one host. For Claude Code or Claude Desktop, set ANTHROPIC_BASE_URL=https://api.59api.com and ANTHROPIC_AUTH_TOKEN=YOUR_59API_KEY. For Codex, OpenAI SDKs, cURL and OpenAI-compatible clients, use base_url https://api.59api.com/v1 and the chat endpoint https://api.59api.com/v1/chat/completions. For Gemini CLI, set GOOGLE_GEMINI_BASE_URL=https://api.59api.com and GEMINI_API_KEY=YOUR_59API_KEY; Gemini models use Google native protocol only and do not work on /v1/chat/completions.
With that in place, the final call is simple. Buy the weekly card when you can name the workload and roughly bound it: a seven-day sprint, a scheduled job, a burst of Claude Code sessions in Claude Code, or a fixed run of GPT calls through your own tooling. The card gives you an isolated key, a hard weekly cap, and no surprise as the week unfolds.
Stay on pay-as-you-go when usage is spiky, experimental, or shared across families, because on 59API prepaid balance never expires, there is no monthly fee, and every request is charged by real token usage. You can top up by WeChat Pay or bank card and request an invoice. Either way, you are working with the same 59API account and the same API host. Start with a small top-up, watch how your tokens actually move, and let the weekly card earn its place in your workflow.