Blog · Engineering notes

Kimi API, minus the payment headache

Kimi (Moonshot) has been one of our favourite models for long-context work. Getting it to accept money from outside mainland China is a different story. This post is the short version of how we route Kimi through one OpenAI-compatible key and pay for it with PayPal.

A few months ago I needed Kimi's 128k context for a project, signed up, and immediately hit the usual wall: the models are excellent, but the billing flow assumes you have a mainland payment method. No card, no top-up, no API. If you're building from overseas — or from a team that bills in USD — that wall is real.

We solved it the boring way: instead of calling each provider directly, we go through an OpenAI-compatible gateway that already holds the upstream connections, and we pay the gateway with PayPal. One key, one dashboard, one invoice line.

Why not just call Kimi directly?

If you're in China with a working local payment method, by all means call Moonshot directly — it's a solid API. The pain shows up when you're outside that billing system, or when you want Kimi and a couple of other models behind the same code path. Maintaining separate SDKs, separate keys, and separate rate limits for every model family gets old fast.

A single OpenAI-compatible endpoint means your existing openai client keeps working; you just change base_url and the key.

The setup we use

One API key for Kimi, DeepSeek, Qwen, GLM, GPT-5, Claude and Gemini-family models on the platform.
OpenAI-compatible: base_url + Authorization: Bearer — no vendor SDK required.
Smart routing picks the cheapest/healthiest upstream for the model you asked for, so one model name keeps working even if an upstream wobbles.
PayPal top-up, so no local card is needed for most international users.

This is exactly what AI24X does — one key, PayPal billing, 20+ models.

Create a free account Open Console

Pricing (read the live page)

Entry models start at $0.35 per million input tokens — but prices move, promo packs change, and per-model rates differ. Treat any number here as a ballpark and check the official pricing page before you commit. We deliberately don't copy-paste a full rate table into blog posts, because stale numbers are worse than no numbers.

Quick start in five steps

1) Register and create a key. Sign up at the registration page, then create an API key in the Console.

2) Top up with PayPal. Add a small balance in Console — PayPal USD billing, no card required for most accounts.

3) Point your OpenAI client at the gateway.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.ai24x.com/v1",
    api_key="sk-YOUR_AI24X_KEY",
)

resp = client.chat.completions.create(
    model="kimi",   # or vip-kimi / flash / other names from the model list
    messages=[{"role": "user", "content": "Summarise a 100-page PDF for me"}],
)
print(resp.choices[0].message.content)

4) Or test with curl before writing any code:

curl https://api.ai24x.com/v1/chat/completions \
  -H "Authorization: Bearer $AI24X_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "kimi", "messages": [{"role": "user", "content": "Hello from Kimi via AI24X"}]}'

5) Pick your model name from the Console model list. Names like kimi, vip-kimi, flash, pro exist today, but the catalog is the source of truth — always check it after major releases.

Pitfalls we hit (so you don't have to)

base_url must end with /v1. Miss it and requests hit wrong paths like /chat/completions instead of /v1/chat/completions.
Paste the full key. A truncated key gives a clean 401 every time and costs you 20 minutes.
402 usually means zero balance. Top up in Console, then retry — no need to redeploy anything.
vip-* models need an active membership. If you don't have one, start with the non-VIP names; the API behaviour is identical.
Long-context requests are heavier than they look. Token counts include your system prompt and full history. Start small, measure, then scale.

FAQ

Is this a replacement for the official Moonshot API? No — it's an alternative billing and routing path. If the official API already works for you, keep using it.

Do I need a mainland payment method? No, that's the point. PayPal top-up covers most international users; see the Console for the current payment options.

Is my data sent somewhere extra? Your requests are forwarded to the upstream model provider you asked for. Read the platform's privacy policy and pick models accordingly — standard API-consumer judgement applies.

Which models are available? The Console model list is the live answer. Kimi, DeepSeek, Qwen, GLM and several GPT/Claude models have been on it recently, but the list evolves.

One key, PayPal, OpenAI-compatible — try it with your next side project.

Create account Console & billing

This article reflects our own setup and experience at the time of writing. Model availability, pricing and payment options change — the live site always wins. Nothing here is financial or investment advice.