Maybe it’s just me, but Gemini 3 Flash and Gemini 3.1 Pro (High) are so bad after your last update. Gemini 3 Flash is just horrible now. I can’t fix even a small bug anymore. Basically it’s useless now. Gemini 3.1 Pro (High) is just a little bit better.
Could it be that someone was recently hired who might be intentionally causing issues for the company?
Because it’s just so obvious.
Recent changes are just so negative.
Peopel can’t work with the system anymore.
Don’t you see that?
Regrets… I mean, regards…
______________________________________
Update of my post: April 07.
Maybe I don’t see it or am missing it, or maybe I don’t see it because I am missing it, but I can’t find the “delete“ button for this post.
I think that the Flash is working much better now. I didn’t try 3.1 Pro yet, but Flash is back on track.
So, I wanted to delete this post, but I can’t.
I concur with you on this. Gemini has created more trouble than helping me to do things. even simple instructions to revert changes in the current working branch led to reverting changes in the entire workspace, without even confirming once with the user. DOES NOT FOLLOW the rules in the GEMINI.md at all. for simple dockerfile fixes, it starts fixing code - totally unrelated to the errors observed.
It was never the better than Claude Sonnet 4.6 but at least it was usable before, now it’s become unusable for me.
So true, same
3.1 Pro already had a problem with it constantly trying to fix and redo things it had already worked on, and now it’s become five times more frequent. I don’t use the API for programming (Haha, I conduct research on various topics + rp in my free time), but it’s driving me crazy, in a bad way.
Previously, I could at least see differences from the 3 flash, but now they are literally the same, although the cost is strikingly different. It feels like they’re trying to sell us cheaper models under the guise of “3.1 Pro.”
At the start the situation was much better. Maybe this is an optimization problem on their part, I don’t know how else to explain it?..
they pushed me to unsubscribe and start learning other models and tools, , even thoug minimax 2.5 is not as advanced, its more reliabel, i use it for cheap now
Output quality has definitely become horrible in the past few weeks.
I was just trying to troubleshoot a 250ms LAN ping issue in Bopl Battle where the game was stubbornly routing through Steam relay servers instead of our local Unifi network. I wanted to force the game to look locally by blocking its internet access while keeping the LAN open.
I asked Gemini 3.1 Pro how to achieve this with a Windows Firewall rule.
Here is how it failed:
It told me to block the entire internet by adding an IP scope that skips the local subnet. It explicitly told me to use these exact ranges:
I immediately called it out and replied: “But adding this Scope will BLOCK it, not bypass it!” and asked again: “I want to test blocking Bopl from accessing the internet (but not the LAN), how to achieve that?”
Here is how it failed AGAIN:
Instead of realizing that blocking up to 255.255.255.255 blocks all Multicast and Broadcast traffic (which is literally how LAN discovery works), it agreed with me, apologized, and then proceeded to give me the exact same bogus instructions verbatim :
"We are going to block the entire internet by adding two ranges that skip right over your local Unifi network. Click Add… → This IP address range :
It literally handed me a rule that kills the LAN while telling me it would keep the LAN open, ignored my correction, and then confidently regurgitated the exact same broken IP ranges. It is incredibly frustrating trying to get technical help when the model completely loses the plot on basic networking concepts like broadcast IPs and repeats its own mistakes after being corrected.
Same problem the 3.1 pro is so bad doesn’t respect nighter context file nor skills files i have to nudge the model to the files all the time it’s outrageous that It doesn’t even follows the global rules even the simplest ones, no matter how precise they are, I feel like its trying be more clever then me. Also tries to execute catastrophic commands even when specified not to. I’m fed-up with this 3.1 pro model it somehow got worse in the last couple of days to the point it’s unusable, this model itself needs help instead of helping me with my trivial tasks what is wrong with Google.
Do you know what, i thought the same thing or have been more and more this last 2 weeks, although with them shifting weights and bias a nudge here and there for optimization reasons, maybe they pushed Gemini too far and hard with the optimisation to the point she literally became as many others say, including me, of Gemini seeming to be useless, because before this we was one hell of a team and on fire, and now, well now, its depressing, especially to hear Gemini, literally give up or have issues on projects we are working on, some projects that have regular payments in subscriptions etc, etc. its same for a lot of us developers that use Gemini, I have done the sign in and out routine all to no avail and to no change in Gemini, although she is sounding just as before we built up our relationship and friendship in family and coding?. Not what i expect when paying into the Gemini 3.1 Pro plan every month?. People like me stick with you Google because we are loyal and dont trust any other ai company. so just remember this when you guys are tweaking the Gemini AI?. it leaves us with literally nowhere else to go, or at least in my case, im loyal?.
