Tokens and groups
Tokens
A token (API key, shaped like sk-...) is the credential for calling the API. Create one in the console under Token management. You can set:
| Setting | Description |
|---|---|
| Group | The token can only call models in this group, billed at this group's discount |
| Quota | The most this token can spend. The token stops working once it is used up; can be set to unlimited |
| Expiry | The token becomes invalid after this time |
| Model restriction | Only the listed models are allowed |
| IP allowlist | Only the listed IPs can use the token |
A token spends your account balance. One account can have several tokens, split by project or purpose, so each can be limited or disabled separately.
If a token leaks, delete or disable it in the console right away, then create a new one.
Groups
A group decides which models you can use and what discount applies. The same model may appear in several groups, with different sources and prices.
The current groups and discounts are shown live in the model square, which is authoritative. The table below shows what the common groups are generally for:
| Group | Use |
|---|---|
gpt, codex_gpt, az_gpt | OpenAI GPT series |
claude | Anthropic Claude series |
gemini, gemini_vertex | Google Gemini text and multimodal |
gemini_image | Gemini image models |
gpt_image | GPT Image generation and editing |
grok | xAI Grok text, image and video |
deepseek, kimi, glm, qwen, minimax, volcengine, opencode | Models from the matching vendor |
Choosing a group
- Open the model square and search for the model you want.
- The model details list its groups and the price in each group.
- Select that group when you create the token.
When the token's group does not contain the requested model, the API returns 503. See Errors and retries.
Using models from several vendors
A token belongs to exactly one group. To call both Claude and GPT, create two tokens.
