This is on purpose. gemini-2.5-flash-preview is treated as an invalid model ID, and whenever it finds an invalid model ID it automatically routes to qwen3-32b. It's a bit of an issue when you need a specific model, since it doesn't give an error or a notification that it routed to qwen3-32b,

@Mostlime12195 I don't understand why this is purposefully done? Isn't it better to route gemini-2.5-flash-preview to google/gemini-2.5-flash-preview and give the user the model they actually wanted.

@Efe-Cal This is done, probably, to maintain functionality in apps when models might be removed from the API. I personally think it is better just to return an error.

As for routing gemini-2.5-flash-preview to google/gemini-2.5-flash-preview, it is done for the sake of standardization and simplicity. This is how openrouter, or any other inference provider works (afaik). Just add the AI lab before the model name and you'll be fine. A lot of the time, this is also done for disambiguation. In the future, a model from another lab with the name gemini-2.5-flash may exist, and that would clash with the existing gemini-2.5-flash. Company/lab prefixes prevent that.

Either way, @skyfallwastaken is the guy maintaining this project. I am only attempting to provide a possible explanation as for why these choices were made, and I have no real power in changing these standards.