The Gemini API docs show that OpenAI compatibility has been implemented for the Gemini API. However, according to the OpenAI URL, it still in a v1 beta. Because of this, it seems like a lot of parameters and functionality from the OpenAI API standard is missing in Gemini’s implementation of it.
For example, when I use the open-source Qwen Agent framework with the v1 beta URL provided by Gemini, I get an error saying that an undefined parameter “seed” was passed. Diving into the codebase, it looks like Qwen agent is passing a seed parameter in the generation config. All other OpenAI API providers support this parameter and never had any issue but since Gemini is in beta, I’m assuming this parameter is not a part of its implementation.
I think its high time these fixes were implemented. OpenAI API compatibility has been a thing since November 2024 and should be well out of beta by now.
Thanks!
Hi @Kabeer2004 , welcome to the forum.
Thanks for your feedback. Gemini OpenAI-compatible API is still in beta, so not all features are supported yet. Please refer to the official docs for the currently supported features.
Thank you
Yes, I’m aware that it is in beta - the launch of OpenAI compatibility was several months ago, should be well out of beta by now.
I’d be happy to help in any way possible to implement the gaps in functionality, but the Gemini OpenAI API does not seem to be open-source, like the Python and Java GenAI SDKs
1 Like
For anyone looking for an OpenAI-compatible way to access Gemini while the official API is under development, you can use OpenRouter.
Set your own Gemini API key using BYOK in OpenRouter and make it always use your key.
And thats it, you can use the OpenRouter API with an OpenRouter API Key + a Gemini API key to access the Gemini series of models via a fully implemented OpenAI interface.