Supported Models
Upthos Relay supports three upstream protocols. Use the same model names as the official APIs.
Anthropic (Claude)
| Model | Endpoint |
|---|---|
| claude-opus-4-7 | /v1/messages or /v1/chat/completions |
| claude-opus-4-6 | /v1/messages or /v1/chat/completions |
| claude-sonnet-4-6 | /v1/messages or /v1/chat/completions |
| claude-haiku-4-5 | /v1/messages or /v1/chat/completions |
OpenAI
| Model | Endpoint |
|---|---|
| gpt-5.5 | /v1/chat/completions |
| gpt-5.4 | /v1/chat/completions |
| gpt-5.3-codex | /v1/chat/completions |
| o3 | /v1/chat/completions |
| o4-mini | /v1/chat/completions |
| codex-mini | /v1/responses |
Google (Gemini)
| Model | Endpoint |
|---|---|
| gemini-2.5-pro | /v1beta/models/{model}:generateContent |
| gemini-2.5-flash | /v1beta/models/{model}:generateContent |
Protocol Translation
You can call any model using the OpenAI format (/v1/chat/completions). The gateway automatically translates the request to the correct upstream protocol.
For example, calling claude-sonnet-4-6 via the OpenAI SDK works out of the box -- no need to use the Anthropic SDK.