XiuRouter exposes OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and Gemini GenerateContent routes. Match the route to the protocol your client actually sends.
Base URLs
| Client behavior | Configure this base URL |
|---|---|
| OpenAI-compatible SDK or application | https://router-api.xiu.ai/v1 |
Claude client that appends /v1/messages |
https://router-api.xiu.ai |
Gemini client that appends /v1beta/models/... |
https://router-api.xiu.ai |
| XiuRouter console | https://router.xiu.ai |
These values are client base URLs. A direct curl request must include the full path, such as https://router-api.xiu.ai/v1/responses.
Text endpoints
| Method and path | Use |
|---|---|
GET /v1/models |
List models visible to the current API key |
POST /v1/chat/completions |
Existing OpenAI-compatible clients and tools that only support Chat Completions |
POST /v1/responses |
Codex, agents, and new OpenAI-style applications |
POST /v1/responses/compact |
Responses compaction for upstream types that support it |
POST /v1/messages |
Claude Code, Anthropic SDKs, and Messages clients |
GET /v1beta/models |
List models through the Gemini protocol |
POST /v1beta/models/{model}:generateContent |
Gemini-native text generation |
The gateway also has routes for other modalities and tasks. Do not add one to a production dependency until the intended model and service group have passed a small request on that exact route.
Authentication
OpenAI-compatible requests use:
Authorization: Bearer YOUR_XIUROUTER_API_KEYAnthropic Messages can use:
x-api-key: YOUR_XIUROUTER_API_KEY
anthropic-version: 2023-06-01XiuRouter also accepts a Bearer token on the Messages route for gateway clients such as Claude Code.
Gemini GenerateContent can use:
x-goog-api-key: YOUR_XIUROUTER_API_KEYThe Gemini key query parameter is also accepted. Avoid retaining complete request URLs in logs when the key is in the query string.
Compatibility boundaries
/v1/messages/count_tokensdoes not have a dedicated XiuRouter route. Claude Code documents token counting as optional and can fall back through the inference endpoint, but the final task must still complete.- The Responses route is stateless. Stored conversations,
previous_response_id, background mode, and provider-hosted tools are outside the current compatibility scope. - XiuRouter exposes Gemini GenerateContent, not the Gemini Interactions API.
- Files, fine-tuning, image variations, and some legacy endpoints are not implemented by the current gateway.
- A model working through Chat Completions does not prove it works through Responses, Messages, or Gemini.
- Public model protocol metadata is a useful hint, but it is not a substitute for a request with the intended model and service group.
- Tool calls, structured output, prompt caching, streaming events, and token accounting can differ when the gateway converts between an inbound protocol and an upstream provider format.
Choose a protocol
| Protocol | Prefer it for |
|---|---|
| Responses | Codex, agents, and new OpenAI-style applications |
| Anthropic Messages | Claude Code, Anthropic SDKs, and Claude-native clients |
| Chat Completions | Existing OpenAI-compatible applications and tools that do not support Responses |
| Gemini GenerateContent | Gemini SDKs and Gemini-native clients |
Follow the client’s native protocol. Do not force a client onto another route solely because it is described as “OpenAI compatible.”
Sources and review date
This page was checked against the current XiuRouter routing and authentication behavior, the public model catalogue, and the linked vendor protocol references on August 22, 2026. No real API key was used for a billable cross-protocol test during this review.