Gemini-2.0-flash-thinking-exp-01-21: Urgent Fix Needed for Google Search Integration + Paid API Access Request

I’ve been experimenting with the gemini-2.0-flash-thinking-exp-01-21 model, and I have to say it’s incredibly impressive! I’ve noticed Google’s own Gemini application has leveraged this model extensively, achieving outstanding results. As a developer, I’m eager to integrate this powerful tool, especially in combination with Google’s search capabilities, to expand my own projects and explore innovative application ideas.

However, today I’ve encountered an issue—the model currently doesn’t support Grounding with Google Search for web-based testing. My application is at a critical testing stage, and this functionality is essential. Could your team please look into this issue and provide a fix or an update on the timeline?

Additionally, I’d like to strongly suggest transitioning this model into a paid service as soon as possible. The current free usage limitations, particularly around parallel processing, are significantly restricting our ability to test and implement the model effectively within our company. Moving to a paid model would greatly benefit developers by lifting these API constraints.

Moreover, incorporating the ability to call Google Search tools within this paid model would greatly enhance its utility and unlock even greater potential for developers.

Thanks for your fantastic work—I look forward to your updates!

Hi @milkmilk, Thanks for the feedback!!

There are some limitation of thinking model, you can refer this doc. Currently the grounding is not supported.
But, I will raise this feature request with our team.

Thanks.

1 Like

Hi @Govind_Keshari

The current response is an HTTP 500

"code": 500,
    "message": "An internal error has occurred. Please retry or report in https://developers.generativeai.google/guide/troubleshooting",

If there are restrictions, the API should return a better message.

Thanks.

2 Likes

Thanks for getting back to me!

I’m a bit confused because, for the past two weeks, gemini-2.0-flash-thinking-exp-01-21 has actually been able to use the grounding feature without any issues. In my tests, when search was enabled, the model’s capabilities were unexpectedly strong. Thanks to its reasoning power and ability to generate long-form responses, it could explore different angles of a user’s query, conduct extensive searches, and provide highly comprehensive and well-structured answers—something even Gemini 2.0 Pro struggled with.

I can confidently say that gemini-2.0-flash-thinking with search enabled is already highly capable and well-developed. I truly believe it’s ready for a paid model. If needed, I’d be happy to provide some examples showcasing how powerful the model becomes with search functionality enabled.

That being said, I’ve noticed that recently, the search function is no longer available, which is why I wanted to report this issue. Would it be possible to restore it?

Thanks again for your help!

Hey @milkmilk, I have already raised it with the team. It depends on their decision. Recently, showing the thought through the sdk was stopped.
Hopefully, all the tools will be supported with the stable version this is what my guess is.

Hey @jkirstaetter, Sometimes we see so many reports related to “internal error” , “unknown error occured”. Thanks for flagging. There is always a scope for improvement. Hopefully, will provide better clarified response in such cases.

1 Like