Overview
Last updated September 1, 2026
Endpoints, authentication, model IDs, and routing syntax for the Amux API.
Amux provides endpoints compatible with OpenAI, Anthropic, and Google Gemini, plus a unified endpoint for asynchronous generation. You can keep using your existing SDK by changing its API address and key.
Addresses and endpoints
- API host:
https://gateway.amux.ai - OpenAI SDK
base_url:https://gateway.amux.ai/v1
Endpoints are grouped by calling contract:
Text generation (convertible protocols)
| Protocol | Method | Path |
|---|---|---|
| OpenAI Chat Completions | POST | /v1/chat/completions |
| OpenAI Responses | POST | /v1/responses |
| Anthropic Messages | POST | /v1/messages |
| Google Gemini | POST | /v1beta/models/{model}:generateContent /v1/models/{model}:generateContent |
Image generation and editing
| Protocol | Method | Path |
|---|---|---|
| Google Gemini | POST | /v1beta/models/{model}:generateContent /v1/models/{model}:generateContent |
| OpenAI Images | POST | /v1/images/generations |
| OpenAI Images (edit) | POST | /v1/images/edits |
Unified asynchronous generation
| Protocol | Method | Path |
|---|---|---|
| Amux Tasks | POST | /v1/tasks |
| Amux Tasks (get) | GET | /v1/tasks/{id} |
| Amux Tasks (cancel) | POST | /v1/tasks/{id}/cancel |
The four text protocols can be converted between one another, but the available entry points still depend on the model. OpenAI Images endpoints are only for image generation and editing; they are not part of text-protocol conversion.
Gemini's generateContent appears under both text and images — it is the same address, and generationConfig.responseModalities in the body decides whether the call returns text or an image.
Amux Tasks is a unified asynchronous media API. The model determines the modality, so image, video, audio, and other media tasks share the same submit and retrieve paths. Check the model catalog for currently available models.
Authentication
Use an Amux API key and pass it in the format expected by your client:
| Protocol | Authentication |
|---|---|
| OpenAI-compatible endpoints and Amux Tasks | Authorization: Bearer <key> |
| Anthropic Messages | x-api-key: <key> |
| Google Gemini | x-goog-api-key: <key> or the URL parameter ?key=<key> |
Create and view API keys on the keys page in the console.
Model-name syntax
Use the canonical ID shown on the model detail page. Its format is manufacturer/model:
anthropic/claude-opus-5
openai/gpt-5.6
deepseek/deepseek-v3.2
Some common model names are also supported as aliases. Use canonical IDs for new integrations to avoid ambiguity.
Select a provider
By default, Amux routes each request across the available providers. To use a specific provider, append :<provider> to the model ID:
anthropic/claude-opus-5:anthropic
A pinned request does not fail over to another provider. Omit the provider suffix when you need automatic failover.
Select a routing strategy
Append :@<strategy> to the model ID to set the provider ordering for one request:
anthropic/claude-opus-5:@price
| Suffix | Behavior |
|---|---|
@balanced | Balances availability and channel weight; the default |
@price | Prefers lower prices |
@latency | Prefers lower time to first token |
@throughput | Prefers faster output |
@reliability | Prefers higher success rates |
A routing strategy only reorders the providers available to the current key; it does not expand the provider pool. Provider and strategy suffixes cannot be combined. If the API key has a fixed routing strategy, the request cannot specify another one.
Protocol and model are two separate things
Protocol and model support are independent. Each model detail page lists the protocols available for that model. Unsupported combinations return an error and are not silently downgraded.
Some parameters may be converted or ignored when you call a model through a different protocol. See Protocol conversion compatibility for the exact rules.
Next steps
- Request parameters, response fields, and examples: select an endpoint in the sidebar
- Parameter conversion rules: Protocol conversion compatibility
- Error formats and retry guidance: Errors and retries