Availability belongs to the exact model ID
A provider family on this page describes the catalog scope; it does not guarantee that every upstream model or version is available. Open Models and Pricing to verify the exact ID, supported endpoints, current availability and billing.
Copy, do not guess
Model IDs can include dates, capability labels and provider-specific suffixes. Copy the full value from the console before sending a request.
Choose the request format separately
| Format | Primary route | Guide |
|---|---|---|
| OpenAI-compatible | /v1/responses or /v1/chat/completions | Responses and Chat Completions |
| Anthropic Messages | /v1/messages | Messages API guide |
| Gemini generateContent | /v1beta/models/{model}:generateContent | generateContent guide |
Evaluate models by completed task
- Define a redacted evaluation set that matches real usage.
- Compare quality and failure modes, not only benchmark summaries.
- Record time to first token and total latency at realistic concurrency.
- Measure total cost per successful task, including retries and long outputs.
- Keep a fallback model and test it with the same request format.