Gemini 2.5 Pro is great, it's just doesn't work

[Yovel_Cohen]

I’ve been using gemini 2.5 pro on the Gemini developer API for few weeks now.
While I acknowledge the fact it’s in preview, it’s almost not possible to even evaluate it, since in almost every session I try to use it with multi-modal prompting (images + text), it just returns a 5XX error.
Am I the only one who is experiencing this?
When it works, its results are good, the latency’s fine, and it seems great.
But currently, it’s barely testable as you get 500 errors constantly.

[Govind_Keshari]

Hi @Yovel_Cohen, Welcome to forum!!

Thanks for the feedback. Sometimes we see 5xx error due to server side error or migration or overload issue. This is an intermittent issue which will resolve by itself. This will be mitigated once it becomes a stable model.

I recommend Temporarily switch to another model like 2.5 flash.

[Yovel_Cohen]

Hi @Govind_Keshari, thanks for your reply.
It’s not only that 5xx errors are almost the norm for the model, but also empty responses, as stated here:

Gemini 2.5 Pro with empty response.text

Running Gemini 2.5 Pro with grounded search sometimes returns empty response.text with finish_reason of STOP and no other reason. When I inspect the response dict it shows evidence of the search with some web meta information, but nothing else. Any ideas on what is going on? Here is my code: def get_response(seed, model, system_prompt, user_prompt): api_key = os.environ.get('GEMINI_API_KEY') google_search_tool = Tool( google_search=GoogleSearch() ) response = None …

Makes me wonder when and if I’ll be able to test the model, on the one hand Google’s pushing it, including through it’s sales. On the other, I can’t even test it.

I don’t think using another model like 2.5 flash is the answer, because they don’t serve the same purpose…

[Yasar_Arafath]

"finishReason" : "MAX_TOKENS" - But Text is Empty

Here text is empty but giving finish reason as MAX_TOKEN { "candidates" : [ { "content" : { "parts" : [ { "text" : "" } ], "role" : "model" }, "finishReason" : "MAX_TOKENS", "index" : 0 } ], "usageMetadata" : { "promptTokenCount" : 25090, "totalTokenCount" : 25090, "promptTokensDetails" : [ { "modality" : "TEXT", "tokenCount" : 25090 } ] }, "modelVersion" : "models/gemini-2.5-flash-preview-04-17" } I’ve mentioned m…

true, I also facing same issue for long time

[Govind_Keshari]

Hey @Yovel_Cohen, Are you also using grounding by any chance or any other tool??

[Govind_Keshari]

Hey @Yasar_Arafath, I will check on this and get back to you.

[Yovel_Cohen]

no, I use structured output schema, but it’s still none even if I remove it from the request.
my input includes text + images

[Govind_Keshari]

Thanks for confirming, I will check on this.

[Govind_Keshari]

Hey @Yasar_Arafath, This is a different case, if the output token count is more than max output token count then we see empty response.

[Govind_Keshari]

Our team is working on 5xx error, meanwhile if you can share your code if possible so i can repro the issue.

[nilnor]

I will have to agree with OP here. I’ve been trying to create a coding CLI agent for Gemini, put it’s been tough going to be honest.. 500 errors ever so often, and on top of that the model does not produce any ouput at times and just stops, and it just plains refuses to actually format strings correctly for tool calls. Well, I’m just frustrated at this point, but I’ll jack in Claude to see how that handles for comparison

[Rodger_Blom]

IF Gemini 2.5 outputs at all, it is absolutely great!. This “issue” truly is a showstopper to use it in production. Please resolve it asap, I love to use this in production.