Skip to main content
OrcaRouter’s /v1beta/ surface is a direct passthrough to Google’s Gemini API. Pick this path when you want Gemini features that don’t fit OpenAI shape — inline audio/video data, native built-in tools like googleSearch, fully-formed generationConfig.thinkingConfig, or simply because you’re already on the google-genai SDK. For everyday chat the OpenAI-compatible path (POST /v1/chat/completions with model=google/...) is usually simpler and works across providers — see Compatibility / Google GenAI SDK for SDK setup.

​
Path layout

POST https://api.orcarouter.ai/v1beta/models/{model}:{action}
Recognized actions:
ActionWhat it does
:generateContentOne-shot response (JSON)
:streamGenerateContentSSE stream of partial chunks
Other Gemini actions (countTokens, tunedModels.*, etc.) are not routed through this surface today.

​
Authentication

The gateway accepts three header forms on this surface, so the official google-genai SDK works without header hacks:
  • Authorization: Bearer sk-orca-...
  • x-goog-api-key: sk-orca-...
  • ?key=sk-orca-... (query string, last resort for tools that don’t support setting headers)

​
Examples

curl "https://api.orcarouter.ai/v1beta/models/google/gemini-2.5-flash:generateContent" \
  -H "Authorization: Bearer sk-orca-..." \
  -H "Content-Type: application/json" \
  -d '{
    "contents": [{"role": "user", "parts": [{"text": "Write a one-sentence haiku about the sea."}]}]
  }'

​
Built-in tools

Gemini’s three native built-in tools work as documented by Google — pass them in the request body’s tools array:
ToolEffect
{"googleSearch": {}}Web grounding. The response carries groundingMetadata. OrcaRouter captures webSearchQueries for billing.
{"codeExecution": {}}Lets the model run sandboxed Python.
{"urlContext": {}}Lets the model fetch and summarize URLs you mention.
The same three tools are reachable from OpenAI-shape Chat Completions via reserved function names — see Advanced / Web search and Advanced / Tool calling.

​
See also