This configuration bypasses extension UI limitations by routing direct Google AI Studio requests through Google’s native OpenAI-compliant endpoint. This allows you to use newly released or unlisted models like gemma-4-26b-a4b-it completely free of charge, without needing a third-party proxy account like OpenRouter.
1. Prerequisites
VS Code with the RooCode extension installed.
- A Google AI Studio API Key (Get one for free at aistudio.google.com).
2. RooCode Profile Configuration
Open your RooCode settings menu (the gear icon at the top right of the extension pane) and fill out the fields exactly as follows:
| Setting | Value to Enter | Notes |
|---|---|---|
| API Provider | OpenAI Compatible | Standardizes data transport to match Google's bridge. |
| Base URL | https://generativelanguage.googleapis.com/v1beta/openai/ | Must include the trailing slash (/). |
| API Key | Your Google AI Studio API Key | Starts with AIzaSy... |
| Model | gemma-4-26b-a4b-it | The exact technical identifier for the model. |
| Model Reasoning Effort | None / Default | Not natively supported by Gemma models; leave it blank or default. |
3. Why This Works (Troubleshooting Context)
The Bridge Endpoint: Google provides the/v1beta/openai/URL block specifically so developers can drop Gemini models seamlessly into software written for OpenAI backends.
Bypassing Dropdowns: Choosing "OpenAI Compatible" instructs RooCode to treat the Model box as a free-form text input rather than a strict, predefined dropdown menu, preventing the extension from stripping out unlisted models.
Native Tool Calling: RooCode requires full tool-calling support to write code and read files in your editor. Becausegemma-4-26b-a4b-itsupports native function calling out of the box via this bridge, RooCode's autonomous agent actions function normally.
Pro-Tip: If Google updates this model line in the future (e.g., to a newer point release), you only need to swap out the text string in the Model field without changing any other parameters.