Developer gateway

Bring GLM 5.3 into your product.

Create an account to generate a private API key, then connect through the same OpenAI-compatible patterns used by familiar SDKs.

Last updated August 23, 2026

Generate an API key

OpenAI-compatible API

One endpoint, familiar tools.

Use your GLM 5.3 Online key with standard Chat Completions clients. Requests draw from the same credit balance as the playground and failed upstream requests are refunded.

Endpoint

/api/v1/chat/completions

Model

glm-5.3

Authentication

Bearer API key

cURL

curl https://glm53.online/api/v1/chat/completions \
  -H "Authorization: Bearer glm_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3",
    "messages": [{"role": "user", "content": "Explain this function"}]
  }'

Python

from openai import OpenAI

client = OpenAI(
    api_key="glm_live_YOUR_KEY",
    base_url="https://glm53.online/api/v1"
)

response = client.chat.completions.create(
    model="glm-5.3",
    messages=[{"role": "user", "content": "Write a unit test"}]
)
print(response.choices[0].message.content)

1,000 credits equal $1 of usage

Input, cached input and output are metered separately

HTTP 402 means the account needs more credits

Authentication

Keep keys on the server.

Generate a key after signing in, store it in an environment variable and send it as a Bearer token. Never expose a live key in browser JavaScript, a public repository or a client-side mobile bundle. Create separate keys for separate environments so a compromised integration can be revoked without interrupting every client.

Send messages as an ordered conversation and keep system instructions concise. OpenAI-compatible describes the request shape; it does not guarantee identical behavior for every provider-specific SDK option.

Errors and retries

Retry only safe failures.

A 401 response indicates an invalid or revoked key; 402 means the account needs credits; 429 means requests are arriving too quickly. Treat 4xx validation errors as instructions to fix the request. Use bounded exponential backoff for transient 429 or 5xx responses, and do not automatically repeat a tool action that could create a duplicate side effect.

Expose a stable message to users while retaining the status and request identifier in server logs. Cap retry counts, add jitter under concurrency and close the upstream connection when a stream is cancelled.

Usage and limits

Meter the whole request.

Credits cover input, output and optional search. Record the response usage fields and provider request identifier for support, but avoid logging secrets or sensitive prompt text. Apply a client timeout, output limit and per-user request budget. Current prices appear on the pricing page and may change independently of a model release.

Maintain a per-user allowance and alert on unusual consumption. Test realistic input and output sizes before a batch job because reasoning effort, retrieved context and answer length all affect the final total.

curl https://glm53.online/api/v1/chat/completions \ -H "Authorization: Bearer glm_live_YOUR_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "glm-5.3", "messages": [{"role": "user", "content": "Explain this function"}] }'

from openai import OpenAI client = OpenAI( api_key="glm_live_YOUR_KEY", base_url="https://glm53.online/api/v1" ) response = client.chat.completions.create( model="glm-5.3", messages=[{"role": "user", "content": "Write a unit test"}] ) print(response.choices[0].message.content)