Choose an API
Token360 exposes several model APIs. Choose the protocol that matches your model, SDK, and workload instead of treating every endpoint as interchangeable.
Decision table
| OpenAI-compatible text or multimodal chat | POST /v1/chat/completions | Best default for existing OpenAI SDK integrations. Supports streaming where the selected model supports it. |
| An application built on the OpenAI Responses API | POST /v1/responses | Use for Responses-style input, output items, reasoning, and tool fields supported by the selected model. Stateful follow-up operations are not supported unless explicitly documented. |
| An application built on the Anthropic SDK | POST /v1/messages | Preserves the Anthropic Messages request and response shape for supported models. |
| A single chat request that should run in the background | POST /v1/async/chat/completions | Poll the returned job. This is not the Batch API and does not support callback_url. |
| Many independent requests in one file | Batch API | Upload JSONL, create a batch, then poll or configure a webhook. |
| Image generation | POST /v1/images/generations | Parameters vary by model. Use the model detail page as the source of truth. |
| Video generation | POST /v1/videos | Asynchronous. Poll the video resource or configure a supported callback. |
| Speech or transcription | /v1/audio/* | Select the endpoint that matches the audio direction. |
Compatibility rule
Token360 normalizes common fields, but compatibility does not mean that every model supports every field. Before shipping:
- Open the model detail page and check its modality, supported parameters, limits, and pricing.
- Use the protocol expected by your SDK.
- Test both non-streaming and streaming behavior when your application uses both.
- Test tool calls, structured output, reasoning, images, or files separately; these capabilities are model-specific.
- Preserve the original assistant fields in multi-turn history. Do not rebuild an assistant message from visible text alone when it also contains reasoning or tool-call data.
Normalized versus native requests
The default mode accepts Token360's documented OpenAI-style fields and maps them to the selected upstream route. For image, video, and speech models that require a vendor-native body, use Token360-Native-Params: true only when the model documentation says to do so.
Native passthrough changes request validation and field mapping. Token360 still authenticates, routes, meters, and bills the request. See Models → Native Params Passthrough before enabling it.
Base URLs
For OpenAI-compatible clients:
1https://api.token360.ai/v1For raw HTTP, use the complete endpoint path, for example:
1https://api.token360.ai/v1/chat/completionsDo not append /chat/completions to an SDK base_url; the SDK adds the resource path.
Production checklist
- Use a public model name from the live catalog instead of an upstream deployment ID.
- Keep API keys on your server, not in browser or mobile code.
- Set an application timeout appropriate to the workload.
- Record the response request ID and trace ID.
- Reconcile production charges by request ID.
- Handle
429,5xx, and interrupted streams according to Rate Limits, Timeouts, and Retries.
Related documentation
- Quickstart — send the first request.
- Models — browse the live catalog and model-specific schemas.
- Model Capabilities — tools, structured output, reasoning, and multimodal support.
- Async Chat and Batch Jobs — choose a background execution mode.
- API Reference — exact request and response fields.