I’m using Claude Opus until I reach the limit quota (so, it does not last long). When I switch to Gemini 3.1 Pro, the performance is way worse. I have to loop again and again to reach something Claude would do in one step.
Guys, do you have the same feeling/experience with this? I’m surprised about the difference.
EDIT: Writing this post, Gemini just spoiled some functionality that has been working properly! What a mess.
Gemini pro is not that bas, need a bit more guidance. Try Gemini flash the only really usable model using a pro plan you will see it’s garage and chaos
Yeah, same here. I barely use Gemini for anything big.
Opus writes the main code and handles almost everything. Gemini is more of a “tiny targeted bug fix” tool for me. Like, if there’s some small annoying issue and using Opus tokens on it feels wasteful, I’ll throw it at Gemini.
I don’t think Gemini is terrible. It’s actually pretty good in a lot of cases. It just has the bad luck of sitting right next to Opus in the model picker lol.
So yeah, Gemini for small fixes. Not as my first choice. If I hit the Opus limit, I usually just wait until it comes back.
Thanks for your time.
Honestly talking, the difference that I was feeling is really notorius.
I’ve been having problems even with small fixes. It fixes one but it messes up two! I’m killing time right now waiting the refresh of the quota 
나도 동일한 경험을 하고 있습니다
제미나이3.1프로 가 너프 된걸 체감 하고 있어요
I wonder if it’s even 3.1 custom tools ( https://www.reddit.com/r/GeminiAI/comments/1rcplac/i_had_to_switch_to_31_pro_preview_custom_tools/ ) since I’ve had a number of baffling failures with commandline tool syntax too.
I don’t know. But definitely I can’t order Gemini to do anything because it starts to spoil every progress I’ve done.
I think Gemini 3.1 pro was not THIS dumb. Did they change its reasoning level or something like that? It sounds to me like they acknowledged this fact.
Yep Gemini Pro 3.1 seems to just talk with itself endlessly about what tools to use and not for every iteration it does, often many times during the same query. It is using way too much energy on this and hence is extremely slow to actually do anything as well. I hardly use it for anything serious. I save up for Claude Opus for the more major jobs and just use Gemini Flash for simpler well defined things as that is actually very fast and seems to do the job well enough.
Yes, I’m testing Flash 3 and it works faster and better for little tasks than 3.1 Pro.
They changed something in this model. I wish I had the same quota for Claude, which is the best one for complex tasks by far.
New and supposedly better quality Flash model is likely being announced soon. If they dont park Antigravity it might be good enough to keep the AI Pro subscription.
Yeah they downgraded there model so the new one feel good ahah but it will be the same