Gemini service providers cannot invoke the custom gemini-3.5-flash model.

[notoccupy2023-design]

When using a custom gemini-3.5-flash model, the invoked model unexpectedly changes to gemini-3.1-pro-preview.

[proyectoauraorg]

I'd like to take this one. Root cause: GeminiHandler.getModel() discards any apiModelId that isn't in the hardcoded geminiModels map and falls back to geminiDefaultModelId (gemini-3.1-pro-preview) — even though the settings ModelPicker has a "use custom model" option and the useSelectedModel hook keeps the configured id as-is. So the frontend honors the custom id but the backend swaps it.

PR incoming that aligns the backend with the frontend by honoring custom gemini-* ids (with a baseline ModelInfo), while still falling back for ids that don't look like Gemini models. 🦘