59API
🌐 Français
äž­æ–‡EnglishEspañolà€čà€żà€šà„à€Šà„€PortuguĂȘsFrançais
ConnexionS'inscrire

← Retour aux guides

Claude pricing on 59API vs official list prices: real numbers

API · EN · 2026-11-04

Why your Claude and GPT bills look the way they do

If you call Claude or GPT through the providers' own endpoints, you already know the shape of the invoice: official list prices per 1 million tokens, billed in USD, with input and output priced separately and cache reads priced on top. For agentic workloads — long context windows, repeated prompts, batch jobs, multi-step coding sessions — those list prices compound quickly. A single run that moves a few million tokens can cost more than a month of your other tooling, and on the official channel there is no way to negotiate the per-token rate down.

That is the problem 59API was built around. The question for a developer is not whether you need Claude or GPT; it is whether you have to pay official list price for every single token. This article puts the two side by side with real numbers from the current rate card, so you can judge for yourself whether routing your requests through 59API makes sense for your workload. Every figure below is quoted exactly as published, in both USD and CNY, so the comparison is arithmetic rather than marketing.

What 59API is and how billing works

59API is an independent third-party AI API relay. One account and one API key give you access to one API host, https://api.59api.com, for Claude, GPT, Gemini, DeepSeek and other models. There is no monthly fee: you top up a prepaid balance and you are charged per request by real token usage. The balance never expires, which makes it easy to load once and spend steadily over time without losing anything.

Billing is in CNY, and USD figures are shown at „7.2 = $1, so the dollar numbers below are directly comparable to official list prices. Top-ups come in fixed amounts — „10, „50, „100, „200, „500, „1000 — paid by WeChat Pay or bank card, and invoices are available. All prices are quoted per 1 million tokens, and cache reads are billed separately at a much lower rate, which matters when you run repeated prompts or agents that re-send the same context. Optional monthly and weekly cards exist for steady workloads, but they are never required: the pay-as-you-go balance is the default way to use 59API.

Rates per 1 million tokens next to official list prices

Here is where the comparison gets concrete. A selection of models from the current rate card, with 59API prices next to the official list prices where those exist:

Read those pairs again: on 59API the same models run at a fraction of the official list price, while billing stays per request by real token usage — you pay for the tokens you actually consume, nothing more.

Worked examples with real token counts

Per-token rates only matter once you multiply them by real usage, so here are the worked examples from the rate card. All of them assume no cache hits:

These are not hypothetical discounts; they are arithmetic on published rates. A heavy session of Claude Opus 5.5 work that would cost $18 at list price costs $6.67 through 59API, and a large GPT-6.1 Sol batch costs $1.85 instead of $20.

How to point your client at 59API

Connecting takes two settings. For Claude Code or Claude Desktop, set ANTHROPIC_BASE_URL=https://api.59api.com and ANTHROPIC_AUTH_TOKEN=YOUR_59API_KEY; Claude models work only through the genuine Claude Code or Claude Desktop client, version 2.1.285 or newer, never through the OpenAI SDK, cURL or other clients. For Codex, OpenAI SDKs, cURL and OpenAI-compatible clients such as Cursor, Cherry Studio, Cline and OpenCode, use base URL https://api.59api.com/v1 and the chat endpoint https://api.59api.com/v1/chat/completions with GPT, DeepSeek and other non-Claude, non-Gemini models. Gemini models use Google's native protocol only, via Gemini CLI with GOOGLE_GEMINI_BASE_URL=https://api.59api.com and GEMINI_API_KEY=YOUR_59API_KEY.

Once connected, every request is metered by real token usage against your prepaid balance, and cache reads bill at the lower cache rate shown above — a meaningful saving for agents that re-send the same context.

Cards for steady workloads, and a small first top-up

If your usage is steady, optional monthly or weekly cards add fixed quotas for the Claude or GPT family. For example, the C-PRO ć…„é—šç‰ˆ card for Claude costs $83.19 („599) for 30 days with quota $1300 and a weekly cap $500, a reference value at official list prices of about $361.11 („2600); the G-PRO ć…„é—šç‰ˆ card for GPT costs $83.19 („599) with quota $3250, weekly cap $1250, and a reference value at official list prices of about $902.78 („6500). 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. Unused quota does not roll over.

To test the pay-as-you-go numbers above, you do not need a card at all: top-ups start at „10, there is no monthly fee, and the balance never expires. Whichever way you pay, 59API charges per request by real token usage, so the price breakdown you just read is the price you actually pay. Try 59API with your own workload and compare the invoice to the official list price yourself.

Articles connexes

Docs

PrĂȘt Ă  commencer ?

Une seule clĂ© pour Claude, GPT et d’autres modĂšles, facturĂ©e aux tokens rĂ©els, sans abonnement mensuel. Inscrivez-vous pour crĂ©er votre clĂ© API ; les tarifs de chaque modĂšle sont sur la page Tarifs.

Inscription gratuite Tarifs