Skip to content
AMD AMD Radeon Cloud Docs

Token Factory

Token Factory hosts a set of models on Radeon GPUs and serves them through an OpenAI-compatible API. There’s nothing to launch and no credits to spend: copy your API key and send requests. Any OpenAI client, such as curl, the openai SDK, LangChain or Cherry Studio, works once you change the base URL and the key.

Token Factory is on the China site only. The platform chooses the models and changes them from time to time. To run a model that isn’t listed, or with your own vLLM settings, serve it on a GPU instance instead.

Get your API key

Section titled “Get your API key”
  1. Sign in to the console and open Token Factory.
  2. The Your API key card at the top shows the Base URL and your API key. The key is hidden; click Show key to see it or Copy to copy it.

Token Factory with the Your API key card: Base URL and the hidden API key

The key starts with rc- and works for every model in Token Factory. Keep it secret: anyone who has it can spend your daily allowance. See Authentication for rotating a key.

Pick a model

Section titled “Pick a model”

Token Factory lists its models. Filter them by Use case, Provider or Publisher, or search by name. Each card shows whether the model is busy right now.

Click a card to see the model’s description, its pricing and how to call it. The value to send in the model field is shown there. The Model catalog describes what each model accepts.

Prices are in points (pts) per million tokens, for example 0.14 pts / 1M tokens. A model billed by input characters, such as OmniVoice, is priced per million characters instead, and its card says so. Points measure usage against your daily cap. They aren’t money, and they aren’t credits.

Make a request

Section titled “Make a request”
Terminal window
export RADEON_API_KEY=rc-...


curl https://developer.amd.com.cn/radeon/api/v1/chat/completions \
  -H "Authorization: Bearer $RADEON_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"DeepSeek-V4-Flash","messages":[{"role":"user","content":"Hello"}]}'

From Python, point the OpenAI SDK at the same base URL:

from openai import OpenAI


client = OpenAI(base_url="https://developer.amd.com.cn/radeon/api/v1", api_key="rc-...")
response = client.chat.completions.create(
    model="DeepSeek-V4-Flash",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

To use another model, change model. To list the models your key can call:

Terminal window
curl https://developer.amd.com.cn/radeon/api/v1/models \
  -H "Authorization: Bearer $RADEON_API_KEY"

Token Factory also accepts the Anthropic Messages API, for clients such as Claude Code. See Messages. Every operation is described in the API reference.

Limits

Section titled “Limits”

Token Factory is free, but usage is capped:

  • Requests per minute and concurrent requests are limited per account and per key.
  • A daily usage cap, counted in points (pts), resets at 00:00 Beijing time (UTC+8).

When you hit a limit, the request returns 429 with a Retry-After header. Every response carries your remaining allowance in its headers. See Rate limits and Usage and quota. Your usage is also on Credits & Usage in the console.

If you keep hitting the limits, serve the model on your own instance instead.

Terms

Section titled “Terms”

Token Factory is a free trial and development service, not for production. Reselling access is not allowed. See API terms of use.