# API endpoint and capability matrix

Confirm the exact model and enabled endpoint in the marketplace first, then select a protocol here. A registered route only means the gateway recognizes it, not that every model or channel supports it.

## Beginner: path, model, and channel are separate

A path selects a request contract, a model ID selects a capability, and a channel selects the upstream destination. All three must match. Start in the [model marketplace](https://api.tu-zi.com/pricing), find the exact ID and its enabled endpoints, then choose a route below. A provider offering image models does not mean every image model works through a chat endpoint.

| Capability | Main gateway entry | Transport and result |
| --- | --- | --- |
| Model discovery | `GET /v1/models` | JSON list for the current token |
| Text | `POST /v1/chat/completions`, `POST /v1/responses` | JSON or SSE |
| Claude | `POST /v1/messages` | JSON or Claude SSE |
| Gemini | `POST /v1beta/models/{model}:generateContent` | JSON; streaming differs |
| Retrieval | `POST /v1/embeddings`, `POST /v1/rerank` | Synchronous JSON |
| Images | `POST /v1/images/generations`, `POST /v1/images/edits` | URL or Base64 |
| Audio | `POST /v1/audio/speech`, `transcriptions`, `translations` | JSON, multipart, or binary |
| Video | `POST /v1/videos`, `POST /v1/video/generations` | Create and poll a task |
| Realtime | `GET /v1/realtime` | WebSocket, not SSE |
| Generic async | `POST /async/*`, `GET /get-async?id=...` | /async retired (410 after authentication), /get-async removed (404); use model task protocols |

## Minimal discovery request

Use the same site token planned for production. Presence in this response still does not prove support on every endpoint, so follow it with one inexpensive contract request.

```bash
curl "$BASE_URL/v1/models"   -H "Authorization: Bearer $API_KEY"
```

## Registered but unsupported

`/v1/files`, `/v1/images/variations`, and legacy fine-tunes routes currently return not implemented. Registration provides a deterministic error; it is not a capability promise. Documentation must distinguish an HTTP route, an upstream feature, and a gateway channel adaptation.

## Intermediate verification

Record path, exact model, Request-ID, HTTP status, response object type, and usage. For 404, check spelling, token visibility, and endpoint adaptation. For 400, validate the schema. For 401 or 403, inspect the site token. For 429, distinguish request rate, token throughput, quota, and balance.

## Expert contract

Maintain an allowlist rather than concatenating arbitrary paths. For every capability, record method, content type, sync or async behavior, streaming, terminal states, body limits, result retention, retryable errors, and billing unit. Run a contract test against the real model whenever a channel or protocol changes; replacing only the base URL is never sufficient.
## MiniMax M3 video input

`POST /v1/media/video` uploads MP4/MOV/AVI/MKV (up to 50,000,000 bytes) and returns a public URL for M3 video understanding. `POST /v1/media/video/delete` deletes the uploaded object using its receipt. Site-token authentication and configured staging storage are required. See [files and media](https://api.tu-zi.com/docs/use-cases/files-media) for examples. These extensions do not enable `/v1/files`.

## Use cases

- Choose text, media, and retrieval routes
- Separate implemented, unsupported, and channel-specific behavior

## API paths

- /v1/models
- /v1/chat/completions
- /v1/responses
- /v1/messages
- /v1/embeddings
- /v1/rerank
- /v1/images/generations
- /v1/audio/speech
- /v1/videos
- /v1/videos/{video_id}/remix
- /v2/video_generation
- /v2/video_regeneration
- /v2/h3_context_ir
- /v2/query/video_generation/{task_id}
- /v1/media/video
- /v1/media/video/delete

## Related guides

- [AI API quickstart: from API key to first request](https://api.tu-zi.com/en/docs/use-cases/getting-started): An AI API call needs a matching base URL, API key, model ID, and endpoint; start with minimal curl, then add error handling, cost controls, and key security.
- [SSE streaming and WebSocket realtime connections](https://api.tu-zi.com/en/docs/use-cases/streaming): SSE is an event stream in one HTTP response; WebSocket is a bidirectional session. A client must use protocol-specific terminal events instead of treating HTTP 200 or a closed connection as success.
- [Asynchronous media tasks: submit, retrieve and verify](https://api.tu-zi.com/en/docs/use-cases/async-tasks): Use the model task protocol. Generic /async is retired (410), and /get-async was removed (404). Poll the corresponding task endpoint and verify the resulting media.
- [File and media input, output, and lifecycle](https://api.tu-zi.com/en/docs/use-cases/files-media): The gateway `/v1/files` routes are currently not implemented. Attach files in a form supported by the target model and endpoint, and treat every input or output URL as a permissioned resource with a lifecycle.

## Frequently asked questions

### Can I call a path just because it exists?

No. The model must be visible to the token, a channel must be enabled, and that channel must adapt the endpoint. The marketplace owns live availability and pricing.

### Why do capability pages not repeat provider-wide endpoints?

Text, image, video, audio, and retrieval capabilities can use different protocols. Broadcasting a provider route would mislabel media models as chat-compatible.

### Is `/v1/files` supported?

No. It currently returns not implemented. Supply files through the target endpoint using a supported URL, Base64, data URL, or multipart form.

## Sources

- OpenAI API Reference: https://developers.openai.com/api/reference/overview
- Claude Messages API: https://platform.claude.com/docs/en/api/messages
- Gemini API Reference: https://ai.google.dev/api

## Documentation source

Source: https://api.tu-zi.com/en/docs/use-cases/endpoint-matrix

Reviewed: 2026-11-02

This document is an integration reference. Provider-native APIs may differ from this site's APIs; use the corresponding API Reference and model marketplace to confirm the site URL, authentication, model ID, group and parameters. This document does not certify live success for every model or upstream service. It contains documentation and examples, without the current user's credentials or request history.