Capabilities
Model capabilities change as models and provider routes are updated. Use the live model catalog and generated Model API Reference instead of relying on a hand-written capability matrix.
Current sources of truth
- Browse Models to compare currently available models.
- Open the Model API Reference for model-specific request fields and declared capabilities.
- Call
GET /v1/modelswhen an integration needs to discover enabled model IDs programmatically. - Call
GET /v1/models/{id_or_name}before depending on detailed capability or pricing metadata. For a logical model name, keep the fullvendor/modelpath (for example,/v1/models/{vendor}/{model}).
Treat model IDs and returned metadata as opaque values. Do not infer vision, tools, reasoning, structured-output, streaming, context, or caching support from the vendor name alone.
Match the request to the capability
| Application need | Verify on the selected model |
|---|---|
| Text or chat | Chat or text-generation support and the required endpoint |
| Image input | Vision or multimodal-input support and accepted content-part formats |
| Tool use | Function/tool-calling support and the accepted tool schema |
| Structured output | JSON or schema-constrained output support |
| Reasoning | Supported reasoning controls, values, and token behavior |
| Streaming | Streaming support on the exact compatibility surface you call |
| Long context | Current context and output limits from the model card |
| Prompt caching | Model-specific cache behavior and returned usage fields |
Unsupported parameters may be rejected, ignored, or handled differently across compatibility surfaces. Send only fields documented by the selected model page.
Production guidance
- Discover availability at startup or before accepting a newly selected model.
- Pin a specific logical model ID when stability matters.
- Handle unavailable-model and unsupported-capability errors even after successful discovery.
- Read current pricing from the detailed model card; do not copy a price or cache discount into application logic.
- Test the exact model, endpoint, and parameters used in production.