randomly I am seeing prompts with gemini 3 flash preview that timeout after 5 minutes with the error: fetch failed sending request. seems to be pretty random with no specific prompt causing it
Same happens for me, approx 5% of the requests. I implemented a retry handler for that but this shouldn’t be the solution.
Up, I am facing the same problem, this is pretty random
I tried to put the maxOutputTokens to 1024, just to test, after several days, I don’t face this fetch failed sending request anymore, but Gemini 3 flash preview just return truncated response. Others models seems fine.
I think this is a good explanation Fetch failed sending request
google, any response or help here? the consistent random bugs are REALLY annoying to use gemini for things. It feels like you guys move on to new models without resolving previous issues, 2.5 still randomly truncates responses, and now 3 randomly doesnt work.
I’m running into timeouts with Gemini 3 Flash preview as well. I’m wondering if it is being made worse by setting a temperature < 1 in the generation config, which is not recommended for Gemini 3. I’m curious if others who are having this issue are setting the temperature parameter and if you’ve done any experimentation of whether that’s having an effect on the frequency of hitting these timeout issues.
1 Like
I am currently not setting the temperature and still having the problem.
Hi
I am working on reproducing this, Would it be possible to share the prompts and the model configuration used ?
Hi Mustan, The prompts that we had were for gemini to extract items from a pdf into a complex json schema. I did not modify thinking or temperature from defaults, I did try those settings to see if it made a difference and it did not. (exact same prompt and schema in 2.5 works fine) If you need the exact schema and prompts please DM me
I increased our timeout from 5 to 10 minutes and the problem stopped happening, so it’s really just sometimes taking longer than 5 minutes for our use-case (complex pdf processing).