For me I have around 60% error rate
It got really worse recently, just impossible to use gemma 4
Had suspicions that the clawbots using the gemma model too much that it errors out with 500 error to other users.
I think infinite TPM wasn’t the best idea for free users. People can run millions of tokens per request and just spam them. And I noticed that errors appeared when gemma 3 was disabled so could be that too many users are using it (including abusers)
Issues are still here, it’s crazy how post was created almost month ago
I am getting like 15% - 20% success rate.
any updates regarding with this issue? i hope they fix it asap
Feels like it errors less than before, but errors are still there
Yeah sometimes error rate is low and almost every response is good, until it goes crazy again (up to 70% error rate, which is 30% success rate). Feels like somebody abuses API like crazy then hits rate limit and then does that again.
They even deprecated the stable gemma 3 endpoints.
Yeah, it’s crazy to deprecate gemma 3 when alternative is still being unstable and not the best
Errors are still there
Errors still persist. <50% success rate - is anyone working on this?
issue still persists with massive internal error 500.
Why anyone is not working on this?
when is this issue going to be fixed?
I am also getting error 500 on gemma 4 31b, passing PDF with chunk size 1000 with overlap of 300
AI Studio
Errors are still there, it’s been 10 days since my last reply, nothing has changed
The issues are still here, and we are not receiving detailed error reports explaining what is causing the 500 Internal Server Error. It happens with medium and large contexts/questions, but there were also cases where even a simple “hello” resulted in a 500 Internal Server Error.
Errors near to 100% on some daytime. Someone has tried the openrouter endpoint? Is it the same?
thanks for the reports and please keep them coming!!!
I think the servers were overloaded. It seems to be working now.