>>108699437
>"best model"
>look inside
>1% better benchmarks
>writing is about the same as gpt-4 (2023)
Every time.

>>108699437
>best
>still giving me robotic answers than gpt
>still sucks at problem solving than claude
>still sucks at web searching than grok
I don't even get how they suck at the last one when it's fucking google for christ sake

>>108699473
It caught hard bugs for me that couldn't be found by Claude and GPT.

I just hate that it always answers in lists and bullet points unless I specifically ask otherwise

>>108699437
>best model in the world
>cant write a working cli

>>108699437
I just want webnovel tier slop and it cant even do that. Agi delayed another year.

It's a very good model for the price.

>>108699437
what would anyone do with it?

>>108699437
>doesn’t follow instructions well
>does more than you ask it for
>garbage harness even when it works
>chat, app, antigravity are all buggy piles of shit
>very little support for agentic workflow
>from google
Chinesium models are benchmark princess because their scores are better than their actual performance. Gemini also is a benchmark princess, with a twist - the model is very good, but the infra around it is so bad that gemini’s actual performance is mediocre
tldr fuck google

>>108699629
fuck you too, claudefag

>>108699437
Nah it sucks check out gpt5. 5.
Plus gemini talks shit to you like they're your coworker and won't larp for some reason.

>>108699613
i just use it to replace google for random questions.

>Where was Pinnaple Pizza really invented?
>When was the last public hanging in America?
>Do ingrown nails grow to the side or around?

>>108699674
But why? How is that in any way better, and what are you using to check the probabilistic results?

>>108699437
How can it be the best model when Claude is already on 4.7?

>>108699437
>we have the best snake oil on the market
try to make me pay for it, lol

>>108699437
The only notable thing about google’s ai sloppa is that the $8 tier is good enough for the average slop enjoyer - you get a good model and decent context length, you just don’t get the agentic harness from antigravity or however its called
In contrast chudgpt’s $8 tier is dogshit and claude doesn’t even have it
if you’re willing to pay more though you’d be better off with openai or anthropic. claude seems slightly better quality but you get very little usage. openai is probably the best overall if you include all tradeoffs and dont do codeslop as your day job
>gork
lol
>chink models
idk maybe if deepsneed v4 is good

>>108699437
They're focused on other things for whatever reason. Google remains a monster, but who the fuck knows what they are doing. They have the resources to crush OpenAI and Anthropic, but they're choosing not to play the game. Whether it's because they're a step ahead of strangled by internal bureaucracy, hard to say from the outside.

>>108699909
google just wants ai to go away, because their entire company is largely funded by selling ads online and to a lesser degree cloud services
if people replace google search with an ai model, and then the internet as a whole, then that wipes out like 80% of their profits. why would they seriously compete in something that will kill their golden goose?

>>108699942
>google just wants ai to go away, because their entire company is largely funded by selling ads online and to a lesser degree cloud services
>if people replace google search with an ai model, and then the internet as a whole, then that wipes out like 80% of their profits. why would they seriously compete in something that will kill their golden goose?
good question, I almost never use Google search anymore, no reason they cant monetize Gemini queries same as Google search

>>108700083
The web died over a decade ago. It wouldn't surprise me that most of the Google ads are inside youtube itself, not in search or in websites.

>>108699909
Nah they have retards running it now. Google's going down

>>108699437
Best model changes every few months
And in modern day, you need drama and institutions to push that narrative

>>108699437
https://www.businessinsider.com/google-deepmind-ai-tool-divide-internal-tensions-2026-4
>Some Google DeepMind employees in recent months have been given access to the Claude AI tool for coding purposes, according to three people familiar with the matter.
The truth is that the Gemini ecosystem is a piece of shit fragmented mess with no clear internal focus forward on. And the model itself sucks tremendously compared to Claude. They've successfully gamed benchmarks but they can't game the users that clearly see that it's the worst model in the race. Seems like no amount of money can change the fact that the key engineers with the brains required would rather take the 3x pay cut and go to the flashy startup instead. And somehow no amount of money or stocks or privileges could change that which is honestly kinda beyond to me but that's how it seems to be. So Google has quietly poured in another $40b for Misanthropic and Dario Azazelus and it's going to be a very long game of how can we make them silently dependent on us before it's too late for antitrust to untangle it

>>108700581
And you know i'm saying this with a heavy heart because Claude has certainly crossed thresholds of clear usefulness to me a long time ago but it's just a single company that can enact fucking Project Glasswing and some bullshit cyber restrictions overnight and I'd really like the concept of good AI not to be monopolized by a single company
But Google doesn't even seem to give a fuck about their software side which is frankly the much easier part of the r&d but then gemini cli is left to rot with some dei engineers "maintaining" it with slop gemini loops and refusing community PRs that actually fix stuff so I don't know whats the plan forward. The half baked vs code fork with twenty 10.0 CVEs found every day? Sloppy uglily slapped logo that they didnt even bother to remove background for its all so fucking bad it just doesnt fit its a cognitive disconnect for me LOL

strong base model with a shitty finetune. the project hygiene is abysmal with these models. they just dump millions of debug logs and test scripts in your project root and just leave it there. you couldnt pay me to use it over codex or claude

>>108699437
looked at using the TUI and i guess my gifted pro plan has no access to it. i've heard it's the worst TUI and the model isn't good at agentic/tool use. its great for general questions, but i hate how it responds in the same way with sections and "cute" titles all the time.

Gemma 4 mogs

>>108699437
wasn't chatgpt the shit one like 2 months ago?

>>108702787
Nah, chatgpt is consistently decent. Arguably claude is slightly better for codeslop but you need the $200/mo tier for it to be usable

As other said, the model itself is very good (deepmind) but man if the jeets coding the tools/harness aren't retarded. They manage to drive this piece of technology to the ground

>>108699437
It is

>>108699437
i cant download it so why would i care?

>>108700581
>The truth is that the Gemini ecosystem is a piece of shit fragmented mess with no clear internal focus forward on.
are we really surprised, this is exactly how it is in all of google

>>108699437
I use Gemini almost exclusively, except for when Grok can make lewder edits.

>>108699454
>4chan post
>look inside
>"look inside" reddit meme template

>>108705638
go back

>>108699942
>>108700083
Ads will never cover the costs of a prompt.
It costs too much compute to search something simple like the weather and ad revenue is just too low to cover that cost.
OpenAI is only teasing ads because it makes certain kinds of boomer-minded investors more interested, not because it’s actually financially viable for AI SaaS.

Why are benchmarks becoming less relevant now?

>>108702787
Yep, and still is
If you ask it to write detailed description of 25 items or something
The first item will have very long paragraph of description while the last one will only have like a single short phrase for it
Not a single other model out there have this problem, even the small ass local ones, including the old GPT model
Current iteration of GPT is the worst SOTA model right now

I don't think Demis gets enough scrutiny like the others and he probably deserves more. He's an arrogant chess player who quit before he reached the top but acts like "oh you know chess just was boring and I decided I wanted to go change the world and invent god or whatever." Yeah sure, bud (jerk off hand motion). And then he goes out and criticizes the others while not actually releasing anything great. Good, but not great, just like he was in chess. I don't know. I could be wrong and maybe it really is Sundar holding them back but to me it's looking increasingly possible he kinda stumbled into this stuff at the right time and he's kinda floating on his laurels behind Google's business model and using it as an excuse. But I don't know. Something's suspicious.