Models
Capability Matrix
How to read Xenodia model capabilities, pricing hints, modalities, and async support from generated catalog data.
The model capability matrix should be generated from the existing Xenodia backend, not maintained by hand in MDX.
For the MVP, generated/model-catalog.json can be a checked-in public snapshot. Later, the backend should export the file during release or CI.
Minimum catalog fields
{
"id": "openai/gpt-4o-mini",
"name": "gpt-4o-mini",
"modalities": ["text"],
"pricing": {
"input_per_m": 0,
"output_per_m": 0,
"unit": "tokens"
},
"capabilities": {
"streaming": true,
"tools": true,
"async_supported": false
},
"docs_path": "/docs/api/chat-completions"
}Capability groups
| Group | Examples |
|---|---|
| Text | Chat completions, streaming, tools, reasoning controls. |
| Image | Prompt-only image generation, reference image generation, output formats. |
| Video | Text-to-video, image-to-video, task polling, generation duration. |
| Audio | Speech, transcription, translation, voice controls. |
| Async | Task mode support and polling expectations. |
Rule for docs
If a model capability can change without a code deploy, it should come from generated catalog data.