GLM-5 API Without a Chinese Phone Number: OpenAI-Compatible, Pay by Card or USDT (September 2026)

GLM from Z.ai (Zhipu) is one of the stronger open model families for coding and tool use, and paying for it from outside China is awkward: account flows that want a Chinese phone number, and card checkouts that decline a lot of foreign banks. This post covers calling GLM-5 through an OpenAI-compatible endpoint on cloudgpu.app instead: prices, where requests go, and setup.

Prices, 27 September 2026

Per million tokens. The cloudgpu.app column is what you pay; the Z.ai list column is Z.ai’s own published price for the same model.

Model	Input (ours / Z.ai)	Output (ours / Z.ai)	Cached input (ours / Z.ai)
glm-5.2	$1.26 / $1.40	$3.96 / $4.40	$0.23 / $0.26
glm-5.1	$1.26 / $1.40	$3.96 / $4.40	$0.24 / $0.26
glm-5.3-flash	$0.16 / $0.075 (above list)	$0.55 / $0.25 (above list)	$0.04

GLM-5.2 and 5.1 sit under Z.ai’s list. GLM-5.3-Flash does not: since 19 September 2026 the only host we can buy it from charges more than Z.ai’s promotional price, and we chose to keep offering it rather than remove it. If Flash is all you need and your card works at Z.ai, go direct. The API page is the source of truth and moves when prices move.

Where your requests go

GLM on cloudgpu.app is currently served by DeepInfra in the United States, running Z.ai’s published GLM weights. That is one upstream, not two, so a DeepInfra outage is a GLM outage for us too. (The same is true of every model on the gateway today; we add a second upstream per model when we find one serving the same weights at a workable price.) The gateway itself runs in Hong Kong. Prompt and completion text is not stored; we keep token counts, model, latency and error codes for billing.

Set up in 90 seconds

1. Sign up at cloudgpu.app with an email or Google account.
2. Top up on the billing page: Alipay, Google Pay, Apple Pay, Visa or Mastercard from $5, or USDT (TRC-20). A first top-up of $10 or more gets a $5 bonus.
3. Create a key in the API console (it starts with cgw-sk-).
4. Point any OpenAI client at the gateway:

curl https://cloudgpu.app/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer cgw-sk-..." \
  -d '{
    "model": "glm-5.2",
    "messages": [{"role": "user", "content": "Write a unit test for this function."}]
  }'

from openai import OpenAI

client = OpenAI(base_url="https://cloudgpu.app/v1", api_key="cgw-sk-...")
resp = client.chat.completions.create(
    model="glm-5.2",
    messages=[{"role": "user", "content": "Summarise this diff."}],
)
print(resp.choices[0].message.content)

In Cursor, Cline or Continue, set the base URL to https://cloudgpu.app/v1, the API key to your cgw-sk- key, and the model to glm-5.2. Nothing else changes.

Which GLM to use

• glm-5.2 for coding agents and anything where quality matters. It is the one we would start with.
• glm-5.1 if you have prompts tuned against it and do not want to re-test; same price as 5.2 here.
• glm-5.3-flash for high-volume, low-stakes work such as classification and extraction, with the price caveat above.

The same balance also pays for DeepSeek, Kimi and MiniMax on the gateway, and for GPUs by the minute if you would rather run open weights yourself. See the full model list and prices, or the Kimi K2 and K3 guide.