How to choose a group: lines and prices for the same model

A group is a line plus a price tier. When you create an API key you pick a group, and requests on that key go through that line at that group's prices; you can also pick several groups in priority order and let the platform switch between them. A simple rule: use a group labeled lowest price for development and for batch jobs that can be retried; prefer official or cloud groups for customer-facing production; and for critical workloads give the key several groups in priority order with cross-group retry, or keep a second key in another group to switch to when the main line has trouble.

What the labels mean

All current groups

The same model across groups

The most and least expensive groups for the same model can differ several times over. Weigh the savings against the line's stability and sourcing.

Picking by use case

Failover with two keys

For one key with several groups that the platform switches between, see LLM API high availability. When you want your own code to decide when to switch, use two keys: put the main key and the backup key in different groups. When the main line returns a 5xx, a 429 or times out, try once more with the backup:

import os
from openai import OpenAI, APIConnectionError, APIStatusError

primary = OpenAI(base_url="https://www.tutujin.com/v1", api_key=os.environ["TUTUJIN_KEY_PRIMARY"], timeout=60)
backup = OpenAI(base_url="https://www.tutujin.com/v1", api_key=os.environ["TUTUJIN_KEY_BACKUP"], timeout=60)


def chat(messages, model="gpt-5.4-mini"):
    try:
        return primary.chat.completions.create(model=model, messages=messages)
    except APIStatusError as e:
        # 400, 401 and 403 are request or account problems; another line will not fix them
        if e.status_code < 500 and e.status_code != 429:
            raise
    except APIConnectionError:  # includes timeouts
        pass
    return backup.chat.completions.create(model=model, messages=messages)

Give both keys a spending limit so the backup line cannot be run up by accident.

When a group does not offer the model

Calling a model that the key's group does not offer returns HTTP 503 with error code model_not_found, saying no channel is available. To fix it:

1. Open the model's page in Models and pricing and see which groups list a price for it.
2. Use a key in one of those groups, or create one.
3. Other error codes are explained in Errors and retries; group setup is covered in Choosing a group.

After you choose

Every record in Logs in the console shows its group, which makes reconciling by group easy. Groups affect cost as much as model tiers do; see How LLM API pricing works and How to control API costs.