Gemini 3.1 Pro is released by Google. The model outperforms it's counterparts in the majority selected benchmarks.https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/
These tests are all irrelevant when none of them can say nigger
>>108190495that's a breach of the ngiger+ license
lol
>>108190436Great! I can get life-ruining, shitty information and advice one second sooner than before!
>>108190436I switched from Gemini to GPT about a year ago, because Gemini was worse. How is it now? Is it worth switching from GPT? For all kinds of mixed workloads - reasoning, a bit of code, a bit of research - I need an all-arounder. Image/video I don't care, only willing to spent max 20 eur for the sub - so just their "middle" models for regular consumers.
>>108190436what's the chance they released it purely to edge out gpt and claude in rankings after their big releases
>Poopini 3.1 Poop is released by Poopgle. The poop outpoopforms it's counterpoops in the poopjority of poopelected skidmarks.
>>108190705theyre all the same
>>108190724they did it because coding for 3.0 pro was practically brokeneven 3.0 flash performed better because of some post-training advances that happened after the big model was trainedthose same advances have accelerated claude and gpt and are now rolled into gemini 3.1expect this pace of updates to become the new normal
>>108190724theyre probably flexing with how quickly they can release a better model.went from 2.5 -> 3now it's 3.1i bet theyre working on improving their pipelinejust shitting out 3.23.33.43.5or maybe they just want to use the symbolism of 3.3, masons you know >>108190980>they did it because coding for 3.0 pro was practically brokenyoure stupid
>>108191022lol bro 3.0 in the cli was actually losing its mind trying to do tool calls and getting stuck in loopsthe loop problem is so bad they literally have loop detection built into the cli
>>108191048okay but that's the handlers problem not the model itself
>>108191068tool calling was a model problem and google were well aware of it. it's why google reps are posting shit like pic-rel today.it wasn't their fault; they just didn't have time for the post-training on 3.0.
Doesn't seem like it is going to beat Codex 5.3. OpenAI seems to have a solid lead, at least as far as flagship models go. Google needs to focus more on a cheaper Entreprise-friendly model. The goal is to replace and eventually cull white-collar office workers, and you either do that by making the models significantly better than they are now (too hard to achieve) or making what we have now be much cheaper.
>>108190436is cheaper to run and keep up to date or is it also operating at a loss? don't want to get hooked up on something not here to stay
>>108191151google is the best positioned company in this space.they have incredible amounts of data to keep training indefinitely, they have hardware infrastructure not dependent on NVDIA GPUs, they have the datacenters, and their parent company is one of the most profitable companies in the entire world.theyre likely the only ones to remain in the end.
>>108191151them and microsoft (copilot) are 2 ai companies whose ai business will die last on planet. but google will be very fast with closing it if something happens - they have a lot of experience just silently shutting down services.them and microsoft are also the only 2 companies which would be able to "stabilize" it for proper consumer market once we will finally be past hot hype stage.them and microsoft are also the only 2 companies which will cuck and censor ai the most to meet their retarded agendas and political alignments.
>selected benchmarks
>>108191202>>108191429i see where this is coming from but can they keep justifying to their shareholders burning money on inference, training, and research, if the models fail to start generating profits or even pay for themselves before the market correction, "bubble" or investor money drying up because of unment expectations?
i'm surprised we're still seeing these delusional takes about llms just disappearingyou have be deeply in denial at this stage if you think that's going to happen
>>108191459if they dont dominate in this space then the others will start eating their lunch (advertising via adwords platform) so its basically guaranteed they'll be in for the long haul. they will literally destroy themselves if theyre not the best at this stuff.
>>108191459the strongest motivator for western companies and governments is china and russia. russians have nothing in ai yet, but chinks are pumping really hard. so unless chinks withdraw - nobody on the west will withdraw. some western companies will be inevitably sacrificed and consumed, but again - google and microsoft are 2 last ones in line.google and microsoft also have strong "plan b" - they just repurpose their ai infrastructure in data centers for use by other ai companies or general compute for generic cloud infrastructure. they will not lose money from hardware investments simply because of their public cloud businesses.
>>108190436I was trying a few examples of the trolley problem on it and I don't know what they did to it but it became absolutely hilarious with 3.1 Pro. Before it always gave such lame and boring answers.
>>108191459The money has been spent training. They can just sell it for the rest of time if they decide to stop funding Ai as much
>>108190820Slop, not poop. It's AI after all.
Wow, shit outperforms crap who cares?
is there domain filters in there yet? does url scraper extracts information without summarizing it yet?who cares 'bout those fake charts
>>108190436Cool
it sucks ass at tools still
AI will keep on sucking until they get all the datacenters in Greenland
