The short answer
In an OpenAI API request, use the model’s exact model ID—the id value returned by the API—not necessarily its human-readable display name. To see the IDs returned as available to your API, make an authenticated request to GET https://api.openai.com/v1/models:
curl https://api.openai.com/v1/models \
-H "Authorization: Bearer $OPENAI_API_KEY"
OpenAI’s List models reference defines each model’s id as the identifier that can be referenced in API endpoints. The example IDs in its sample response, such as model-id-0, are placeholders—not model names to copy.
Which OpenAI model ID should you start with?
OpenAI’s model documentation gives these starting points. Treat them as the documentation’s selection guidance, not as a guarantee about performance on your particular workload.
| Documented model name | API model ID | Stated starting point | Listed input / output price* |
|---|---|---|---|
| GPT-6 Astra | gpt-6-astra |
Complex reasoning and coding | $10 / $50 per million tokens |
| GPT-6.1 Sol | gpt-6.1-sol |
A balance of intelligence and cost; complex work | $2 / $10 per million tokens |
| GPT-6.1 Terra | gpt-6.1-terra |
Visual reasoning, precise computer use, and clear, natural writing | $1 / $5 per million tokens |
| GPT-6 Luna | gpt-6-luna |
Cost-sensitive, high-volume tasks | $0.10 / $0.50 per million tokens |
* The model cards list separate input and output rates in USD per million tokens (MTok). These are the rates shown in the captured model documentation, not a total-cost estimate. Confirm current pricing and any applicable billing details before budgeting or launch.
For example, if your priority is coding or difficult reasoning, Astra is the documented starting point; if you want a lower listed token rate for high-volume work, Luna is the documented option. If your workload involves visual reasoning or computer use, consider Terra. Your own request mix, required features, and evaluation results may lead you to a different choice.
Verify the ID before production use
- List the models. Run the authenticated request above and inspect the
dataarray in the response. - Copy an exact
id. Use the returned identifier as written; don’t substitute a display name or a sample placeholder. - Check capabilities and terms. Consult the official model documentation for the intended task and relevant API access, features, limits, and pricing. OpenAI’s model page says the latest models it describes are available through the Responses API and OpenAI Client SDKs, but the list endpoint itself is not a capability comparison.
- Test your actual request path. Before relying on an ID in production, test the model with your intended endpoint, features, and workload in your own environment. A returned ID helps answer what the API lists; it does not establish that the model is suitable for your task.
The List models response can also include basic metadata such as the model owner and an optional shutdown date. Use the official catalog for model descriptions and the authenticated API response for the IDs it returns. Third-party client libraries may have their own version-dependent supported-model lists; those lists are not a substitute for checking OpenAI’s API and model documentation.