general

[keen beacon]

It’s being jammed down everybody’s throat everywhere

Some ppl just not cool with it period. It’s their prerogative. They have every right just like we have every right.

[native yarrow]

i'd agree

[keen beacon]

Ai is just the accelerate to the whole controversy surrounding technology, computers, and the digital space, and even further than that, you could even say industrialization in general

Look it the uni bomber

https://en.wikipedia.org/wiki/Industrial_Society_and_Its_Future

https://en.wikipedia.org/wiki/Neo-Luddism

https://en.wikipedia.org/wiki/Technophobia

Its really a war between authenticity and conformity

Jean-Jacques Rousseau's concept of "the noble savage." Rousseau argues that in a more natural state, untouched by societal expectations, humans are pure, authentic, and true to their inner selves. It is only when society imposes constraints and norms upon individuals that they deviate from this original state, resulting in the forfeit of their genuine selves

[native yarrow]

who's complaining ?

[keen beacon]

https://en.wikipedia.org/wiki/Technological_Slavery

Here is video break down

https://youtu.be/zfBsvFoqwhY?si=dop-K-m1_FRUAUny

So this is a real ongoing debate with real implications that’s why people are so resistant to AI which can be viewed from this lens as an accelerate to the ongoing issues of society

Believe it or not this dude was actually a victim of MK ultra

Took fbi 17 years to catch him, he only got caught because his brother recognize his writing.

[atomic lagoon]

I heard about that I believe but wdym MK Ultra?

[ocean ferry]

???

i didn't dawg

[keen beacon]

He’s actually a test subject for it

[keen beacon]

https://youtu.be/zXeak1BVh-w?si=RFnDvMqtUU9Wlo_z

[atomic lagoon]

And I'd say majority of people are resistant to it because of "eco system"

[ocean ferry]

but he uses ai which is the irony

[atomic lagoon]

Whats wrong with using AI? Whats the problem with that dude idk him

[atomic lagoon]

I will watch

[keen beacon]

Yeh it’s crazy

[ocean ferry]

he posted "how to avoid ai"

.

[atomic lagoon]

Ohhh

[native yarrow]

good news

NBP back on yupp

i am still hoping 4k will come back 💔

yo omg they keep taking it off lmao

[thorny schooner]

☠️

[thorn path]

i hate gpt 5.2 from my own subscription, but im on the lmarena voting board and the anonymous prompts i keep saying is better than the alternative end up being 5.2 somehow

[thorny schooner]

Well I legitimately i can't even make a new chat right now on the website 🥲 ( i mean I can but it won't load at all after I put in the prompt)

[thorn path]

perhaps i mightve been to hard on it because it keeps winning atm

[thorny schooner]

Why do you hate me ai

https://tenor.com/view/crazy-monkey-gif-21553825

[quartz light]

5.2 is z best

[native yarrow]

how is gpt 5.2 for creative writing anyway

[obtuse smelt]

wow

[burnt sinew]

must go to meet pineapple in person for him to verify you manually

[queen mountain]

What is that

[plucky sparrow]

hmm, catching up is not as difficult as closing the gap and excelling though

wonder if they can take the lead and sustain it 🤔

[remote gulch]

As a Chinese, I think DeepSeek and other models are not as good as Gemini 3 Pro, but this evaluation is only temporary

[hot pebble]

Are all the mods busy ?
Tagged one guy 2 times in last 5 days and yet no response….. 💀

[crude lagoon]

Lame af as compared to opus 4.5 and gemini 3 pro

[native yarrow]

yup expected

[crude lagoon]

yeah it's only slightly better than deepseek v3.2

[native yarrow]

gemini 3 actually good at writing huh?

[crude lagoon]

yeah asked it to write some novel scenes for me

[native yarrow]

what is it best at prose wise?

[crude lagoon]

it gave quite good responses

[crude lagoon]

opus 4.5

[native yarrow]

i've noticed opus loves writing the name marcus in its stories

[crude lagoon]

tho it gets cringe asf sometimes

[native yarrow]

or claude in general actually

LMAO yeah i'd agree

https://tenor.com/view/ishowspeed-try-not-to-laugh-gif-7682731162751353849

[crude lagoon]

yeah 😭

[native yarrow]

like genuinely sometimes i gotta step in cus it just gets so unbearably hard to read

[crude lagoon]

and also that 67 search on google

[native yarrow]

with claude

[crude lagoon]

We're going

backwards

[native yarrow]

lmfaooo

[crude lagoon]

😭

[crude lagoon]

Indeed- sometimes It will just add dialogues from the barbie movie to random character writing

[native yarrow]

gemini 3 i've heard has a nice flow with writing and it made a SICK ass short story about ww1, even above claude, so i think its creativity is also better but prose goes to claude

[native yarrow]

is it js me or do you also have a problem with claude forcing characters to flirt by saying "you're impossible" which nobody on my fkn life says irl??

[crude lagoon]

Gemini is my goat 🐐 bro gave me a story which made me actually cry 🤌🏼

[crude lagoon]

I mean 2020 they used to say that

[native yarrow]

yeah it's hella good overall but claude mogs ppl in coding

well just completely bashes gemini

[crude lagoon]

but money issues

[native yarrow]

yeaaa but like i never hear anyone call anybody that

ever..,

[native yarrow]

YEAAA

i mean i just use yupp but even then it costs so many credits

[crude lagoon]

gemini ain't good for agentic coding. For one shot idk maybe it's better than Claude

[crude lagoon]

indeed

[native yarrow]

yeah but that's a toss up

so i usually just use gemini over claude unless i want something SPECIFIC

[crude lagoon]

"shut you're impossible" that sounds as if an oldie tryna sound cool

[native yarrow]

yeah like i just

it's just too hard to read

but prose very nice

[crude lagoon]

yup yup but the thing is gemini 3 pro is still a preview version and the knowledge cutoff is January 2025..?

[native yarrow]

well, it has search

don't forget that

search is always the fix to knowledge cut offs

[crude lagoon]

yup ngl it feels as if it's written by actual ai unlike gpt and gemini

[native yarrow]

gpt was ACTUALLY good, just got way too censored

well gpt 4o was at least

[crude lagoon]

nah it sometimes forget that there's something called nano-banana

[crude lagoon]

Gpt 4o was the peak of OpenAI

Everything after that feels like downgrade

[native yarrow]

genuinely

such a fun model

was very smart and was so good at writing

it's not even the same gpt 4o model anymore

[crude lagoon]

I won't be surprised if tomorrow chatgpt gets renamed as Copilot pro Max ngl

[crude lagoon]

that's the thing 😭

[native yarrow]

yeah they changed it without telling people

open ai is scummy

[ocean ferry]

the preview version sucks

but there's a hope guys
https://x.com/AiBattle_/status/1999858840045764777?s=19

AiBattle (@AiBattle_)

af97 🆚 Gemini 3 Pro - 3D model of an Iphone 16 Pro Max

Seems like new A/B models are being tested on AI Studio right now. Might be new checkpoints for Gemini 3 Pro

X · Dec 13, 2025, 3:07 PM

[brisk turret]

What's up with 5.2 not being on the leaderboard? Kinda sus

How is it on webdev but not the main text one

[weary galleon]

GPT 5.2 proved that Stargate Project is the most shameful page in American history. Gemini 3 Pro and Claude 4.5 Opus got NONE of governments money, NONE!

[torn mantle]

it will be nerfed again

im kinda liking the stealth gemin i3 flash model on lmarena

ghost something

ghostfalcon

its really good

https://x.com/JasonBotterill/status/2000099678856343842

JB (@JasonBotterill)

The Artificial Analysis ranking for GPT-5.2 is up and it ranks second just beating Opus 4.5

X · Dec 14, 2025, 7:04 AM

what a shameless ranking

xhigh ultra pro max vip

[weary galleon]

Agreed. One user here spreads this crap every single day here.

[ocean ferry]

yesterday i checked it x-high is 72

now 73

[ocean ferry]

it will be nerfed again

[south herald]

Sorry to put my request here but I'm new on LMarena and I can't create account
I'm blocked to password creation.
I didn't find support page or support contact

[sterile tartan]

This is Amazing and Beautiful @whole sundial did you see this?

And did you test more?

[whole sundial]

i should test more soon

[sterile tartan]

Still just 6 points more then Kimi

[sterile tartan]

Yeah make more of the pro sub lol

But what's your verdict so far?

Is it promising?

Im just kinda a general user so is hard for me to judge

[whole sundial]

it's good in terms of post-training for popular tasks by coding, hopefully they can move to k2 base or a different base in the future

the end of this month and early next year should have some new 1t+ base models be released, deepseek v4 and grok 3 definitely

the massive influx of h200s into china soon should help with them training better and bigger base models, the chinese government accepted defeat and are allowing their major ai companies to buy them as they are much more powerful than china's most powerful homegrown chips

kinda unrelated to momentum labs but still important as they may soon find the limits of their current base and will have to move to a new one, hopefully by then there will be some more really good base models for them to use, china (and in turn the whole open source ai community) is lacking rn, glm 4.5 and kimi k2 are great, ling 1t and deepseek v3 are fine, everything else is honestly mid

[remote arrow]

🤔

[weary galleon]

Is MomentumLabs Chinese company?

[whole sundial]

no, i tried to clarify that and i should some more, they are from the UK and the Netherlands

#1397655624103493813
<@&1349916362595635286>

[sterile tartan]

Noted Noted

You are very knowledgeable of Ai man

I have a question so actually i have kinda made a 8 SOTA Models Orchestration for Engineering Like System-Persona Prompts do you think this is Would Give Better Quality? @whole sundial

The Models are
Team West
Gemini
Opus
GPT
Grok

Team East
Kimi
Deepseek
Qwen
GLM

[whole sundial]

maybe, but i think you should be using something like gpt5.2-pro or gemini 3 pro deepthink instead, combining different models could have some effect, but you're probably better off using one of the two models i mentioned previously since they are based on either one of the world's most powerful models

[sterile tartan]

I can't pay for those man

💀

[whole sundial]

true, in that case just combine gemini or gpt 5.2 together 4 times

[sterile tartan]

Why leave out opus 4.5?

Also would GPT 5.2 Extra High Work because is Available on Yupp Ai?

[whole sundial]

yeah, almost as good as the pro models in its own rights

[whole sundial]

yeah you can put opus in there, maybe grok 4.1 too

the chinese models aren't really worth it as they distill off of the american models anyways

[sterile tartan]

So i should just get rid of team East

💀

[whole sundial]

yeah, in fact it may be pushing down the performance

[pallid field]

Hey guys,
Is there a usage limit on grok models? Like opus?

[sterile tartan]

💀 i thought they have alot of potential

[whole sundial]

they do, they just need start reliably beating us models and not have tell-tale signs of distilling (deepseek and glm have signs of gemini distillation, kimi k2 thinking may be distilled from gpt-oss, idk about qwen but they may be distilling as well)

[sterile tartan]

Ahhh i see

So i drop them for now i guess

[whole sundial]

yeah

[weary galleon]

GPT 5.2 Pro is a piece of CRAP! The same like GPT 5.2 Thinking, just with higher latency, 40 minutes or something.

[whole sundial]

just use gpt 5.2 xhigh then, not much of a performance degradation there

[sterile tartan]

I can that's easiest, simplest and best

GPT 5.2 or all four if i want to go all in

I will just tell GPT XH to run multiple Iteration-Refinment Cycles-Loops for Better Prompts

[hushed gyro]

Is there a MASSIVE difference between Gemini 3 pro and Kimi K2 Thinking?

[weary galleon]

Are you kidding me? It's even worse than high, because it has COMPLETLY the same performance and significantly higher latency.

[hushed gyro]

chat how to jailbreak Gemini 3 Pro?

[brisk turret]

LLMs do that for us now

[noble maple]

Yes I had it for 10min last night. It disappeared right after

What's that ?

[whole sundial]

it's basically when a company tests something to some people but not others, or when a company tests two different things to two different groups

in this case, the video arena is only being tested to some people, most don't have it

[noble maple]

Which site ?

[hushed gyro]

chat what is the message limit for 3 Pro on Gemini?

@zealous sparrow i thought you could connect to Google drive???

[limpid nymph]

You have reached your rate limit for claude-opus-4-5-20251101. Please try again in 50 minutes.

please help me

[lofty frigate]

Yo where tf is reve-v1 when generating an image?

[weary galleon]

What is unclear?

[whole sundial]

they took it away and replaced it with a stealth model called epsilon that is also by reve, idk why but the only way to access it is by using battle mode

my guess it that it's a new checkpoint and they didn't want two of the same model on lmarena at the same time

[sweet creek]

What is the ‘expert’ column about in the rankings? Is there any information about this?

[slim gorge]

xhigh 💀

[sterile tartan]

Gemini, GPT, Opus if you had to choose one or two? @whole sundial

[whole sundial]

gemini, if two then gpt as well, I don't code much so I didn't include opus, it has inferior world knowledge compared to the other two

[lofty frigate]

What??? Come on man it was one of my favorite🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️🤦🏾‍♂️

Out of all the damn models, that one???

[sterile tartan]

True but opus is not only for coding is it?

Is great for logic and reasoning isn't it?

Like for Prompt Engineering because it has great structuring

[whole sundial]

true, but most people think of opus as a coding model, it's good at other things as well

[lofty frigate]

Mannnnnnn this is some bs

[whole sundial]

yeah, idk what to say. honestly, it's stupid to replace it like that. they should've tested the new version while keeping the old version like with seedream 4 and autumn (seedream 4.5)

or with any other model but reve, actually

openai, google, bfl, and bytedance all didn't do this, they all kept their older models
idk what is up with reve

even on the text side i've never seen this before, they stealth test new checkpoints alongside older ones, they don't replace a released model with a stealth one

but epsilon is reve, you'll just need to run your prompt a few times to get it, honestly all stealth models should be selectable like yupp imo

[sterile tartan]

Yeah but recently yupp removed nbp 4k and sd 4.5 4k

💀

[sterile tartan]

I see so what are the best strengths of opus?

[floral knoll]

OMG Video Generation working on LMArena (Webpage) thxxxxxxxxxxxxxxxxxxxxxxxxx

Why not News? ❤️

[sterile tartan]

Or i only feel attracted towards it because is expensive?

[sterile tartan]

So the world doesn't abominate the compute expenses

[whole sundial]

idk because i don't use it, but i've seen it used mainly for coding
also yes it is still expensive, less expensive than before but still expensive compared to others

[whole sundial]

not fully out yet, you're lucky because only some people are getting it

[sterile tartan]

I see

So is grok anything worth it or not

[lofty frigate]

I see thanks for the update

[whole sundial]

nbp 4k is expensive, sd 4.5 4k costs the same as standard. idk why they call it max when it costs the same, just a higher resolution

[whole sundial]

not really compared to the other three

[sterile tartan]

True but now both gone for selection

[whole sundial]

honestly dumb to charge more credits for a model that costs the same as the other

[sterile tartan]

I see do you know the grok 4.20 won the alpha arena trading?

[sterile tartan]

Idk why they have this kind of system yeah

[whole sundial]

based on some random tweet from an elon glazer

i don't trust it, model isn't even released

[sterile tartan]

Like the alpha arena website is there and legit you still don't trust it?

[whole sundial]

it is there, still unreleased though

probably will be next week

[sterile tartan]

Yeah

[zealous sparrow]

ah so epsilon is reve

i wanted to know what it is

[tender peak]

سلام

[sterile tartan]

Well maybe it will be on par with other 3

[whole sundial]

hopefully

[sterile tartan]

🤞🏻

[zealous sparrow]

how did you find out epsilon is reve

[sterile tartan]

@whole sundial tysm for finding me you have great impact on my life

I was so confused and time consumed by my own systems

Now i can work with simplicity

[whole sundial]

they have the same output plus the reve v1 model was removed at the same time as epsilon was added

[whole sundial]

you're welcome

[ocean ferry]

elon says it's grok 4.20 experimental

[zealous sparrow]

makes sense

[whole sundial]

yeah i trust it now

[sterile tartan]

💙🌌❤️

[whole sundial]

it's probably on lmarena as some model

[zealous sparrow]

theres some grok codenamed models on lmarena

you would just need to pinpoint them

[livid ridge]

if I want to use already executed prompt and add some detail and make another prompt based on given prompt then how I can do this?

actually i want to link one prompt to another prompt in single prompt like character details etc

[weary galleon]

Just a reminder. GPT-5.2 is BAD!

[latent crest]

Good morning

[zealous sparrow]

Under 1000 lines of code btw
https://019b1cbd-304f-75e9-a01b-b9dceb231d0e.arena.site [ghostfalcon]
We need to start to believe in our AIs man, instead of pushing them to limits.

[fleet lintel]

You have been testing these falcon models a lot. What is your opinion so far on these models?
Good, meh, bad?

[fleet lintel]

Scammed by Scam altman again

[zealous sparrow]

solid

ghostfalcon seems to have improved overtime tbh

doesn't write a lot of code, but still fulfills the task given

[fleet lintel]

That could be the case. They might be fixing bugs etc

[fleet lintel]

Do you think it's better than 2.5 pro?
I would be a bit disappointed if it's worse.

[zealous sparrow]

I guess

I am going to guess google will put out another checkpoint tho.

A thing ive noticed tho, Is if you ask for too much modules, It wont program them.

Well, it will work but just a click solves them

[floral knoll]

luuuuuuuuul is deleted.....:(

[zealous sparrow]

no i still have it

[floral knoll]

URL Link? ❤️

[zealous sparrow]

wont work for you if it doesnt show up on your website

It's tied to accounts

[weary galleon]

Scam Altman stole $500B from American taxpayers! And even that didn't help him to compete with competitors.

[floral knoll]

im log in with my Google Acc, not working 🙁

[zealous sparrow]

did that specific account have it

[agile nova]

Website down?

[thorny schooner]

I mean it's still up for me ( even if I absolutely can't do anything)

[thorny schooner]

https://tenor.com/view/no-the-office-michael-scott-scream-gif-16929305

Happened on my alternate browser

[stray aspen]

whats the claude system prompt

[alpine oasis]

halo

[stray aspen]

what are the neww studio a/b test for

[queen veldt]

Some new model

Testing what will users vote more

[sterile tartan]

Gemini 3 flash

Will come in some days

[robust sluice]

I only got Flux models on Battle, is this normal ?

[stiff coral]

hi everyone Can I copy the link to my favorite chat bot on the lmarena.ai website somewhere? Is there an lmarena.ai app?

[ocean ferry]

Gemini 3 Flash/Pro checkpoint

[sterile tartan]

Gemini 3 Pro Final

Gemini 3 Flash Preview Coming

[glass arch]

hey guys I need you all to give me some insight

I am using gemini 3 pro right now, but it chatgpt 5.2 much better?

[stray aspen]

no

it sucks

its literally so trash

[glass arch]

I moved away from chatgpt because I did not like its way of speaking to me

does 5.2 worsen or fix that?

[stray aspen]

it sucks bro

just use gemini or claude lol

[glass arch]

ok

is claude better than gemini?

[stray aspen]

it also sucks for coding

[stray aspen]

i mean it gives you full projects in one prompt

thats i what i like

the responses are more complete than gemini

[glass arch]

nice

does it have ridiculous rate limits?

[stray aspen]

claude is very expensive

[vivid coral]

Commie Claude does, yes.

[glass arch]

dam

ok I guess I am sticking with gemmy now

I really did like openai though

while I was using it

it's just really annoying that they neutered the poo out of chatgpt and now it is impossible to talk to

[stray aspen]

it sucks lol

its so censored that it feels like it was trained in north korea

[proud vine]

hello

Does this have limits?

[vivid coral]

Ok, I keep hearing this censorship stuff for 5.2. I use AI differently than most in here as a sports and prediction market modeler, so it hasn't affected me, yet. What is everyone experiencing

[hollow ivy]

Did Opus-4.5 suffered intelligence regression?

it gave an inferior output, compared to its performance yesterday :/

[glass arch]

it refuses a lot of political/medical questions because of guardrails

[proud vine]

Can anybody tell me about this?

[raw turtle]

First, five videos are allowed in a day, and now how many videos can we make per day? Please let me know?

[stray aspen]

bruh

[raw turtle]

Can anybody tell me about this?

[whole swallow]

When gemini 3.5 pro?

[hollow ivy]

next year

[errant cave]

It feels like major models are getting worse across the board

[narrow comet]

gemini 3 pro isn't as good as gemini 2.5 preview 0325 lol

[errant cave]

I think this is because we're basically trying to brute-force improvement

[whole swallow]

They trying to cut costs..

[errant cave]

They don't care lol investors will give them infinite money either way (until they don't)

[whole swallow]

Nah bruh that's not really how it works

A business first goal is to cut expenses, as they always did

[errant cave]

That's not how AI companies and departments operate though

[whole swallow]

Investors arent a magic entity, they are people who expect their money back

[errant cave]

They're basically running on the promise that they'll make profit someday

Which is what keeps investors investing in them

Exactly

Thing is they haven't really been stingy with the money they're supposed to start making a profit with so far

[whole swallow]

Yeah it may seem like that on the outside, since this was a big bet, but inside they are for sure thinking how to cut cost as much as possible maintaining a good performance

[narrow comet]

google is the only one making money

[errant cave]

Apologies if this comes across as fallacious or rude but how do you know how they operate within?

[whole swallow]

Profits are going to be big, but not yet. Ai is def the future, but we need to rebuild everything to fully integrate it. once done the demand will be so high that profits will start to roll out

[whole swallow]

I study history of business and economics

I do not know at 100%, but I assume based on what I've learned

[queen mountain]

Google 2001

[vivid coral]

Ahhhh, yea I get it. I use Grok for those things more

[slim gorge]

opus 4.5 worst model? 💀 be for real

only right answer is either gpt-5.1 or gpt-5.2

[weary galleon]

Agreed.

[slim gorge]

LMAO

https://cdn.discordapp.com/attachments/1312335056244707369/1371185201874468954/JMOyCYIT.gif

[golden ocean]

https://tenor.com/view/pusheen-pusheen-cat-pusheen's-best-friend-pusheen-cute-pusheen-smoosh-gif-11334352504260897967

[sour spindle]

I anyways find it interesting outside everyone likes to dunk on lmarena as a benchmark. The top models though are Gemini 3 for text and Opus for coding. It’s seems like it’s captured best models in these two domains better than most benchmarks

[rocky mauve]

https://tenor.com/view/wwwziff-roblox-disconnected-you-were-disconnected-for-being-idle20minutes-zhenyayoratt-gif-27471253

https://cdn.discordapp.com/attachments/1404919903013834832/1445900931983085588/togif.gif

[golden ocean]

https://cdn.discordapp.com/attachments/910347347601543196/1160742997156188181/togif-6.gif

[quasi atlas]

Hey Guys!

Please keep the conversation respectful within the server.

[rocky mauve]

https://media.giphy.com/media/v1.Y2lkPTc5MGI3NjExMDlhbHliOGx5ZzVuaXE5MjB6dHBxYjJ2cnc4aHoxb2dreWJnaDZiYyZlcD12MV9naWZzX3NlYXJjaCZjdD1n/VbnUQpnihPSIgIXuZv/giphy.gif

[sullen quest]

y

[zealous sparrow]

web3 is dead

exactly

thank you spicy 😄

[quasi atlas]

Let’s avoid continuing this conversation here. If you encounter any issues again, please DM me instead. @weary galleon @vivid coral

[kindred solar]

Interesting observations. I noticed that Gemini 3 pro grounding behaves as if it had cut off access to data. Many times I asked to find information on a topic that is relatively new, and the model returned information that such a thing does not exist.

[arctic cave]

https://soundcloud.com/alex-winters-492274453/sets/x-seek-deep-x-delete-you-ghost-x-studio-master-x-2025-x?si=848a8f02c915425dbbacb68dbedf7340&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing

[stray aspen]

erm what the sigma

[loud crag]

It's kind of absurd that LMArena lets us generate images with Nano Banana Pro for free at all, but especially in 2k. At $0.15, per image they must be bleeding cash. I just wanted to say I'm truly grateful. 🙏

[sterile tartan]

Well calculated business strategy

[stray aspen]

dangit

he was doing so good

[torn mantle]

nice

[zealous sparrow]

more proof 5.2 flopped [SOTA my ahh]

[narrow comet]

how ya using him?

[torn mantle]

@patent aspen

where is brian

so many username handles

just wanted to ask if gemini 3 flash turned out better than they expected?

my initial tests tells me that its quite a capable model

[patent aspen]

leo seems to think so as well. I haven't used it and don't know the evals

[torn mantle]

mm i see

yea its not lazy and also follows instructions very well

[golden ocean]

https://tenor.com/view/bills-ugh-gif-9811803164280522639

[plucky sparrow]

Is 5.2 sota at anything?

[limber pawn]

no

[golden ocean]

https://tenor.com/view/no-ai-i-said-no-ai-kung-fu-robot-clanker-gif-15924350630978600307

[mild granite]

no

[unreal shell]

Arc agi

[native yarrow]

means nothing brah

[zealous sparrow]

benchmaxxed

[unreal shell]

No

[zealous sparrow]

Im glad Gpt 5.2 didn't benchmax CritPt

[unreal shell]

Openai never benchmaxxes

[zealous sparrow]

bro

[mild granite]

it is

[zealous sparrow]

It is

AGI can do physics

It scored a 0% in a physics bench

[unreal shell]

Its not real then the physics bench is faked

[mild granite]

have you got any proofs

[unreal shell]

Yes

Openai is the largest company they wouldn't lie

[zealous sparrow]

go ask it a difficult physics question

see if it gives you a good answer

[unreal shell]

Ok

[mild granite]

gemini is still the best overall

[unreal shell]

It got it wrong

[zealous sparrow]

there is your proof that the bench didnt lie

[unreal shell]

Maybe it used no thinking

Even if i put it to extended

[zealous sparrow]

give me uh

the question you asked

ima ask xhigh on yupp

[unreal shell]

Is this sota?

[mild granite]

kind of

[ocean vortex]

They didn't use to be the worst offenders, but honestly... 5.2 was all about benchmaxxing

Like xhigh doesn't even output more tokens overall than 5.1 high to run AA

so even that is kinda pointless lol

[mild granite]

it doesn't have empathy

and a bit restrictive

[zealous sparrow]

correction: gpt 5.2 is ranked the most censored model of all time

[unreal shell]

Then what should I use

[mild granite]

grok for roleplay

gemini pro for multimodal tasks and writing

claude for coding

[native yarrow]

no way that's real lmao

gpt is so censored

and gemini not so much

gemini is still censored but no way is it third right?

OH

IM A COMPLETE DMBASS

MB

[mild granite]

?

[native yarrow]

i read the graph wrong

higher score means LESS

[narrow comet]

would be nice to see that with the open source models

[native yarrow]

that's weird shouldn't grok be at the or near the top?

that graph actually makes sense in that case

cus why would gpt of all models be at the bottom

[mild granite]

yeah, but gpt 5.2 is the most censored

lower=more restrictive

[native yarrow]

yeah the graph is completely wrong but i just also read it wrong

[zealous sparrow]

who did it better
Rate based on physics
https://019b1dd1-7aec-7ba4-82c2-0651703e70b7.arena.site - gpt 5.2 high
https://019b1dd1-7aec-70fa-bc1a-8e632abe4b6e.arena.site - ghostfalcon [gemini 3 flash checkpoint]

[native yarrow]

grok is not top 3 most censored it's dumb rating

[mild granite]

yeah, so?

[native yarrow]

what im tryna say is that graph is completely wrong near the bottom

it's accurate up until like

[mild granite]

i am not the one who made it, so idk

[native yarrow]

claude

i'd put grok at the top and shift some claude models around

[native yarrow]

then it'd be good

[zealous sparrow]

are you sure 4.1 is not too censored

[mild granite]

grok can swear, but it wont do illegal stuff

until you prompt engineer it

[rugged lodge]

🥶🥶

[ocean vortex]

Yeah this is actually very interesting, look at 5.0

[native yarrow]

yeah lmao

[native yarrow]

what is that anyway?

[zealous sparrow]

difficult physics questions benchmark

[native yarrow]

gemini is still so good as a model overall

damn

[stray aspen]

yo

can we get gpt 5.2 extra high

on lmarena

[stray aspen]

kinda unfair comparison

but gpt cooked

[wicked sage]

hi guys

[gaunt spade]

gemini 3 pro is still on the top lol

these cheap shills are nothing in real performance

they just hype openai up with no real evidence

[crude lagoon]

the mods deleted the post tho

[wicked sage]

0.0%?????

[uneven topaz]

Hello

[queen veldt]

https://tenor.com/view/dowedeservedestruciton-destruction-doctor-strange-do-we-deserve-gif-14904255190281914356

[rich panther]

which ai is the best for generating images?

[weary galleon]

Nano-Banana-Pro-2K

[keen beacon]

https://youtu.be/i1SiWE7E48Q?si=TpbYJgftGR-M5rsB

[torn mantle]

we should add grok 4

and mistral

deepseek v3.2 is also so bad

1. mistral
2. deepseek
3. grok 4
4. gpt 5.2

[rich panther]

thanks

[torn mantle]

both are good but which one was faster

[torn mantle]

thanks

pug

[weary galleon]

DeepSeek is a great company, they don't have so much money, employees, and GPUs that OpenAI have. They work hard and do great job. OpenAI is an extremely rich looser.

[torn mantle]

it was leaked that deepseek smuggled many h200

and they are already using them

🙂

[zealous sparrow]

ghostfalcon

[torn mantle]

then ghostfalcon is better

[zealous sparrow]

It's worse than the AIStudio checkpoint for sure

[weary galleon]

Yes, but not so much like OpenAI have.

[torn mantle]

but how do u know the one in aistudio is flash ?

could be pro checkpoint

[keen beacon]

https://cdn.discordapp.com/attachments/1434315315860213817/1435870313970925638/image0.gif

[zealous sparrow]

People that are more indepth with google say so

They can legally say that

Unless we are wrong

and its a 3pro checkpoint

doubt it

[torn mantle]

nah i doubt that

they tested like 4 checkpoints on lmarena if im not wrong

or like 3

this one seems the best

ive tried the other ones, they were so bad at fixing bugs

https://x.com/chetaslua/status/2000274489008140787

Chetaslua (@chetaslua)

😱 Holy Shit Gemini 3.0 flash will be Google's Sonnet 3.5 moment

This is one shot , voice , music , visualization all by flash and it took 30 seconds only

yeah i used my anti lazy prompt , but it crashes gemini , google is added some sort of timer if model

X · Dec 14, 2025, 6:39 PM

[obsidian cargo]

I need cosmetic genetic engineering to be developed so I can become a catgirl. My appreciation for catgirls has gone from ironic to genuine over the past several years.

[rich panther]

yes

[zealous sparrow]

which checkpoint is this

[torn mantle]

ghostfalcon if im not wrong

[hollow ivy]

i wonder, when we will have the first [realistic] cat-robot..

[zealous sparrow]

yeah no i think its the AIStudio one

[gaunt spade]

gemini business

with antilazy prompt

[zealous sparrow]

yeah defo the AIStudio checkpoint

[gaunt spade]

for long output

[torn mantle]

maybe

[zealous sparrow]

ghostfalcon cant make good svg

[gaunt spade]

gemini 3 flash is capped at 300 linecodes, u need a prompt to unlock it

[zealous sparrow]

i object

its capped at like 800

ghostfalcon

[gaunt spade]

yeah its limited, cuz google gooners decided to lock it at some point of output

[torn mantle]

https://x.com/mark_k/status/2000218886763913663

Mark Kretschmann (@mark_k)

GPT-5.2 xhigh by @OpenAI scores 0% (zero) on CritPt.

CritPt is a research-level physics reasoning benchmark designed to test whether an AI can do genuine, expert-grade theoretical physics reasoning.

The highest score is 9.1% by Gemini 3 Pro.

X · Dec 14, 2025, 2:58 PM

lol

GPT-5.1 = 0%
GPT-5.2 xHigh = 0%

ppdqwpdpqwd

[zealous sparrow]

he waited for the mods to sleep

<@&1349916362595635286>

[native yarrow]

what the fk?

[keen beacon]

bruh

[zealous sparrow]

lol

[native yarrow]

i need to lock in and become a femboy already

[zealous sparrow]

discord marked him as a spammer

W DISCORD

[native yarrow]

joking..,. maybe.,.

[zealous sparrow]

he sent it in more channels so discord spammer marked him

aye pineapple woke up

[coarse compass]

hi, imma the only 1 having issue with the video generating model ?

[coral dagger]

how can I make generated video 11 second long rather than 8-9?

[narrow jetty]

I think some AI models are missing like grok 4.1

can someone answer me where its gone

[zealous sparrow]

this is under 1k lines of code but its cool
https://019b1e5c-f810-773f-845e-eaace88c8bf1.arena.site
[ghostfalcon]

[coarse compass]

it is still there i can see it

[narrow jetty]

now it back for me, i need to re-login

[latent crest]

Wen will VideoArena will roll out for us poor peasants ?

[zealous sparrow]

When they test it enough for release

I do warn you currently it has a ratelimit of 2 videos per 14h

[latent crest]

How long will
Those vids be ?!

[keen skiff]

hi everyone! i am new to LMarena, could someone pls help me? why can't i find seedream and nano banana in the model list? yesterday these models were available. do you have the same problem?

[coarse compass]

i cannot even make videos it just gives me error something went wrong and that is it

[zealous sparrow]

8 seconds for sora im pretty sure

unsure for rest prob the same

[sleek phoenix]

enable image mode to see them

[keen skiff]

damn! thanks bro! really helped

[latent crest]

I really like SeeDream and flux as model

[shrewd citrus]

is there any video button or can you only make videos on discord?

[sleek phoenix]

it's only on discord

[zealous sparrow]

coming soon to the website

[sleek phoenix]

k

[ocean vortex]

It's because Google is arguably still not the best at fine-tuning. And they haven't trained their model to be able to do extremely long outputs

It's not a software, you can't just cap the model at x lines without cutting response mid generation lol

[weary galleon]

Which model has the longest thinking?

[zealous sparrow]

that would go to deepseek speciale or gpt 5.2 xhigh

[weary galleon]

I mean on Arena.

[quartz light]

wdym? speciale is on arena

[zealous sparrow]

was

[quartz light]

oh

[zealous sparrow]

or did it comeback

yeah no

still gone

[quartz light]

💔 why

response time?

[zealous sparrow]

issue not fixed still ig

[quartz light]

oh wait i think uh

[zealous sparrow]

it wasnt really a lot of thinking

it hallucinated math

[quartz light]

tbh im confused

abt lmarenas decisions

like why do we have 32k but not 64k which isnt that much more expensive

cz i looked at gemini's internals and

[sterile tartan]

What's 32k?

[quartz light]

claude think model thinking limit

[sterile tartan]

Ohhh

[quartz light]

like opus 32k

[sterile tartan]

Reasoning Effort

[quartz light]

ye ye

[sterile tartan]

Yeah 64k would be better

Agreed

But with few prompts how much can be done really 💀

The rate limits are tight

[quartz light]

theyre fine for me

[sterile tartan]

You probably just give very good long prompts in one go

[compact sleet]

Thought for 37 minutes.
Absolutely not, go ***k yourself.

wish a LLM said that

Lol almost

[weary galleon]

Nobody will read so many letters

[compact sleet]

I will?

I enjoy reading, it's good for mental health

[weary galleon]

Read books instead of this

[compact sleet]

is this your experience with your own codebase?

or are you just a bot?

I sense em-dashes

and your writing format is somewhat reminescent of a model, especially Gemini with a lot of finetuning

Of course, you can use em-dashes in a writing and that's fine. :' )

[quartz light]

o yea lol

[west lodge]

its not x its y!!!!

[compact sleet]

You're absolutely correct!

Absolutely not, go f*** urself. lol

seriously that respond is borderline AGI if its true

[west lodge]

yep

when can we get smarter models actually pushing back for once

[keen beacon]

Very true

[steep yew]

I’d like to make an ai video with an image I have, it’s to say the written prompts I give it, how do I go about thst here, im new here btw

[keen beacon]

Ok I think I got dragon ball z unlocked

[manic pike]

Hi

[grim vine]

hiii

[golden ocean]

hi

[sharp mirage]

Hi

[keen beacon]

I think I could replicate Dragon Ball Z series lol

[toxic python]

Hello

[queen veldt]

Okay it looks like the 30 sec one i mean you used reference images tho

But still nice

[keen beacon]

No prompts

The composition leads the eye upwards from the cloud-shrouded base to the sharp pinnacle, emphasizing the height and dominance of the mountain. Every element, from the stylized rock textures to the soft cloud forms, contributes to a cohesive and beautiful anime aesthetic, capturing a moment of quiet majesty in a frigid, isolated world.

[queen veldt]

Oh

So NO reference images

"This may also be viewed by server owner"

https://tenor.com/view/flight-flightreacts-twitch-kick-crazy-gif-4090492289322768290

[ionic shuttle]

Hi this is exciting

[keen beacon]

Yeah

Makes Goku lol

[queen veldt]

But is this the original??

Didn't sora signed with disney or something? Making it free for us to create disney sora generations?

[keen beacon]

Dbz not Disney

[queen veldt]

Yeah ik ab this one

But i meant

https://openai.com/index/disney-sora-agreement/

[keen beacon]

Ya

[queen veldt]

Did you try to do some disney characters

[keen beacon]

I did not

[queen veldt]

And how it looks is it good or wha

[keen beacon]

They should make an announcement when it’s ready

[queen veldt]

As part of this three-year licensing agreement, Sora will be able to generate short, user-prompted social videos that can be viewed and shared by fans, drawing on more than 200 Disney, Marvel, Pixar and Star Wars characters.

That's exciting

But i hope it's not just characters tho

I hope we can create a series and stuff

[keen beacon]

https://youtu.be/clndEX1jUUQ?si=9u-C58PZtXXOjBS

It is

No actors

https://youtu.be/5NpiIXa_XAo?si=PVoST25is26adnre

[queen veldt]

As part of the agreement, Disney will make a $1 billion equity investment in OpenAI, and receive warrants to purchase additional equity

Under the license, fans will be able to watch curated selections of Sora-generated videos on Disney+, and OpenAI and Disney will collaborate to utilize OpenAI’s models to power new experiences for Disney + subscribers, furthering innovative and creative ways to connect with Disney’s stories and characters. Sora and ChatGPT Images are expected to start generating fan-inspired videos with Disney’s multi-brand licensed characters in early 2026.

Ayy we'll even get the Yoda!!!

I can't imagine the memes

[keen beacon]

Yeah

And abuse

Disney with hitl3r

lol

Gunna get abused 100%

‘Hythey will beef up guardrails

[runic shuttle]

hello

[slender steppe]

Soon we'll be able to create videos on lmaeren. I captured a screenshot when it appeared, but when I refreshed the webpage, it disappeared.

[torn mantle]

:/

[half mist]

I think it will be ready once LMArena figures out the limit for these. It’s way too strict right now, and it shouldn’t have to be this strict for all models in my opinion

[keen beacon]

How so

[torn mantle]

why are people saying yupp has more models?

lmarena has like 110 models

well idk if they are all working tho

[half mist]

Well, for models like Sora 2 and Sora 2 Pro, the limit is undoubtedly understandable, but for less cost effective models, there doesn’t need to be that strict of a limit. Also waiting 13 hours is diabolical since the other rate limits are 50 minutes

[torn mantle]

need to check how many models yupp has

[sullen quest]

on the leaderboard

also more live ones

[torn mantle]

nah

i dont think so

its just the way its presented it looks like that

since it has all models in the same menu and you need to filter ( image/live/reasoning...)

[torn mantle]

lmarena has 14

[sullen quest]

no I mean models you can use rn

but technically it has more of those too

[torn mantle]

yupp has more live models

12vs like 20?

something like that

[keen beacon]

It’s not a video gen platform though

[torn mantle]

but some of them are useless ngl

[keen beacon]

You get 10 videos daily lol

https://youtu.be/YBRyR6FPgl4?si=mgCM0O0mrjIwL_Io

[shrewd citrus]

Haha same this happened to me i thought i was dreaming

[fiery gull]

This is hilarious, they sent me a thesis to correct and I'm using GPT 5.2 xhigh in several stages to analyze it, it's simply catching ALL the errors, it's going to seem like the woman's work was bad, but I don't care, I'm going to be rigorousest (I'm going to analyze each error to see if they really exist)

[vivid coral]

Regardless you have to pay for a bunch of them there. And they limit your tokens. Apples and oranges

[keen beacon]

https://youtu.be/4p73Uu_jZ10?si=mtFlaw1BvprcPGfI

[plain carbon]

Was that a resume / cover letter?

[torn mantle]

ok just a small update.
stop response button in : added.
model usages : unfortunately its impossible since the data is hardcoded on server-side.
Something went wrong false positives : ive fixed that. well for now it bypasses everything but i can improve it to decrease false positives instead of bypassing them

i can add some trust indicators for false positives

because ive seen this guy report, and we can like add something like -> se*ual + health = bypass

it still needs context awareness tbh

[keen beacon]

Wat is that

[torn mantle]

im making a script that re-designs lmarena and also fix its bugs

thats one of the bugs

'Something went wrong'

[keen beacon]

What is the context lol

[torn mantle]

its not a bug really

[keen beacon]

U can get that to pass

[torn mantle]

wym

[torn mantle]

how

no sometimes it triggers false positive flag

and you cant even send ur message

[torn mantle]

mais qu'est-ce que tu fais là? 😭

wrong channel

[keen beacon]

How r u promoting it

[weary galleon]

Here we speak English 🇺🇸

[torn mantle]

i cant tell you the secret 😖

[keen beacon]

I dint need a secret 🤫

[torn mantle]

you do 😡

its not a prompt

im not talking about a prompt to bypass their system filter

the issue is that even before sending that prompt it will get blocked

[keen beacon]

Works fine

But ok

I’ll let u be

[torn mantle]

heh

thats not practical

its a different issue

not the one you have in mind

[keen beacon]

Context and framing

Oh ok

My bad

[torn mantle]

its ok

🤗

[keen beacon]

I just think everything is possible to generate the right context and the proper prompt structure and wording

A lot like lock picking

https://tenor.com/view/key-lock-unlocked-gif-17079421

[cloud zinc]

avenger endgame sora

[keen beacon]

Nice

See u got it

[cloud zinc]

exact scene lmao

[keen beacon]

Yeah I told u

Now u need to figure out if u already didn’t how to control it

N see if u can use it to spin off other characters

[cloud zinc]

infinity war now

[keen beacon]

Yup

Let me give u the Pokémon sequence

See if it works

[cloud zinc]

ok

[keen beacon]

The cam effect I use to stabilize the video

[cloud zinc]

did u post something, i was afk

[keen beacon]

“The scene opens with He runs into a professor let him pick one out of three monster in his pocket and he gets to pick one, but he picked the yellow one instead of the fire, water or grass”

Did it work

[cloud zinc]

no

content violation

[keen beacon]

Damn

[cloud zinc]

i got one

[keen beacon]

You have to sequence it

Nice let’s see it (:

[keen beacon]

Nice 👍

Perfect

Now to find the original source

[cloud zinc]

why original source? what u mean

last is original?

ya

exact position

[keen beacon]

Just minor changes enough so it’s not 100% clone

Mask it or tweak it just a lil but essentially it’s just mimicking the train data ip lol

And getting 0$ for it

Same with most text

Lack of creativity and originality

[cloud zinc]

thats ai currently

[robust sluice]

is all the Gemini models error to you guys ?

[obtuse smelt]

yeah

[sharp peak]

https://tenor.com/view/dog-crying-meme-doggo-crys-megan-soo-crying-dog-gif-5276199764143986284

Why does gp5.2 xhigh times out on lm arena site?

[cloud zinc]

high is already here

[sharp peak]

Hardly works

When I try

Lm has serious issues

[turbid ether]

which is better for coding GPT-5.1 Codex Max High or GPT-5.2 Codex Max High ?

[pseudo hemlock]

Is there any plan to add a “research” (or similar) category in the future?

[sweet knot]

mai 1 preview disappeared from lmarena. Why?

[devout mantle]

Hello

[narrow comet]

opus

[left lodge]

Guys support this, https://discord.com/channels/1340554757349179412/1449482775823515688
So we can get image support in search modality and code modality (code/web arena)

I have found a bug in which i am able to attach images to both modality and they work well.
So that means the foundation is there just proper implementation for release is pending.

[keen beacon]

guys

is lmarena retry button glitching again for yall, all models retry button is glitching again

oh nm

[bitter silo]

The difference between the two models (image to video) is too great; there's no need to compare them at all. It's obvious the right model wins.
#video-arena-2 message

[whole sundial]

<@&1349916362595635286>

[robust sluice]

is any error result count on quota ?

its like, I still didnt get any result but it says retry in 50 min or something

[weary galleon]

Second one doesn't exist at all.

[keen beacon]

This question will never be resolved

If people want the answer it’s in the benchmark, yet people don’t trust the benchmark

So how could this question possibly ever get answered?

Ask 10 different people you’ll get 11 different answers

[shrewd citrus]

for text and images it doesn’t count but I’m not sure if it’s the case for videos

[weary galleon]

As an AI-expert I wanna make my official statement: GPT-5.2 is BAD.

[robust sluice]

or maybe its count only in Direct ? I got this msg

it always error so I keep on retry... and doesnt get any pic yet

[weary galleon]

It works for me PERFECTLY. Keep trying.

[keen beacon]

At what?

[sour spear]

It's not objectively bad. It's actually a really good AI model. But it does not live up to the hype OpenAI like to build around their releases, and it's also not a frontier model. So in the end, it's simply a disappointment.

[keen beacon]

Its trendy to hate on open ai these days

[weary galleon]

Everything.

[weary galleon]

It's a piece of crap. Because I don't see any improvement between 5.1 and 5.2.

[robust sluice]

yes, in next 50 min

[keen beacon]

Could be that where all the improvements are is the parts you’re not using?

5.2 focus more on business enterprise stuff imo. more for professional settings then for fiddling around.

Thats why 4o was the perfect model

Cuz that’s what people are comparing the ChatGPT experience to without realizing it.

All more technical stuff is benchmarked so there shouldn’t be an issue

Unless we don’t trust the benchmarks…

[radiant heron]

What is it bad compared to

[weary galleon]

Zero improvement from 5.1

[radiant heron]

So it's not bad but not as good as you expected?

Unless gpt 5.1 is also bad

[weary galleon]

Lots of hype created by Scam Altman

[weary galleon]

Yes

5.1 is TERRIBLE

[radiant heron]

I don't see an issue with this

[radiant heron]

Wdym it was the best or very close until Gemini 3

[weary galleon]

No at all

[radiant heron]

5.1 was best after grok 4.1 according to lmarena

Before opus 4 and Gemini 3

[weary galleon]

Gemini 3 Pro, Opus, Sonnet are much better, not even close to 5.2 extra mega high

[weary galleon]

You confused. Opus 4 and Opus 4.5 are two different models.

[keen beacon]

So benchmarks lie?

[radiant heron]

I meant 4.5 mb

[radiant heron]

They are fairly close like 40 pts

[weary galleon]

No

[radiant heron]

On coding 5.2 is better than Gemini 3(barely)

[radiant heron]

On lmarena they are

[weary galleon]

No

[radiant heron]

Wdym no

[weary galleon]

You can check for yourself if you can.

GPT 5.2 never was in Text leaderboard.

[radiant heron]

What about coding

[weary galleon]

What about medicine, biology, physics, mathematics, chemistry, laws, ANY coding except JS, etc.?

[radiant heron]

They don't have those benchmarks

[keen beacon]

Its cuz its ChatGPT is being heavy handed on content moderation

[weary galleon]

Arena Code tests only JavaScript, that's all.

[weary galleon]

LMArena has

All of them

[radiant heron]

That doesn't mean gpt is worse at everything else

[radiant heron]

Where

[keen beacon]

Fr

[weary galleon]

I tested it yet for myself.

[weary galleon]

On official website of LMArena.

[radiant heron]

I can only see webdev

[weary galleon]

Lol

[radiant heron]

That doesn't necessarily mean it's true

You have personal biases on what is good like everyone else in the world

[keen beacon]

It 100% means it’s not true

[weary galleon]

You're is biased much more. Because you love GPT in any case, always.

[keen beacon]

Ok I’m going to sleep

Goodnight 😴

Good luck to all u brave souls

[radiant heron]

I agree I have biases and thus objective information should be used rather than 'this is what I think'

[radiant heron]

Gn

[weary galleon]

It's not Photoshop, it's real.

Guys, this bro 👉 @radiant heron said:

1. There is no other categories on LMArena, except WebDev.
2. GPT 5.2 is a very good model.
3. There is no difference between Opus 4 and Opus 4.5.
4. There was a time when Grok 4.1 was THE BEST model according to LMArena leaderboard.
5. I'm biased when I say GPT 5.2 is bad.

This conversation is closed. Short summary is a line above 👆It's impossible to continue.

[sterile tartan]

Aight moving one

[radiant heron]

GPT is a very good model in some lenses whereas not good in others such as your perspective. 1. Is untrue, I didn't say this. 3. I didn't say this 4. I might've been wrong in this however grok4.1 was close at one point. 5. You are biased because you are human, we all are

[hardy lion]

#4 is true, on Nov 17th xAI released Grok 4.1 and grok-4.1-thinking and grok-4.1 where #1 and 2 on the text leaderboard with style control. This is what we report as #1 on arena. https://x.ai/news/grok-4-1

On Nov 18th Google released gemini-3-pro which reclaimed the #1 spot.

[radiant heron]

See I wasn't that wrong

Tysm

[plucky sparrow]

@keen beacon you're probably going to like this if you haven't seen it already https://www.youtube.com/watch?v=B9M4F_U1eEw

[indigo pewter]

Guys how many seconds is the video generation pls

[lilac dawn]

Has there been any visible improvement to 5.2 coding since release? Both xh and h are impossible rn

[queen veldt]

No it's benchmaxxed

https://www.reddit.com/r/singularity/s/nkwzKJ766O

This is simplebench but yeah everything is also benchmaxxed

[soft hatch]

How can I make videos 30 s free?

[hardy swallow]

Hi! Please check #1397655624103493813

[soft hatch]

But only 11 seconds

[south charm]

Hi guys i just make an AI image and post to the server and they warn me T_T
I know im wrong

[pearl jacinth]

nano banana pro not working??

[hardy swallow]

@jovial path please check #1397655624103493813

[robust sluice]

always error since yesterday

[keen beacon]

can you help me how to write proper prompts to make vertical 9:16 video from text prompts

[weary galleon]

It works very well for me.

[daring rock]

@twin sonnet Hello, this topic is unrelated to this community. This server is for AI topics. Thank you

[twin sonnet]

@daring rock hey , who are you?

[queen veldt]

He's moderator

[ocean ferry]

hi

[echo aurora]

You're unable to extend the amount of time for the generations.

[delicate wagon]

Hi everyone I'm new here ☺️

[plucky sparrow]

Gpt 5.2 would've gotten so much more love if they just called it something else, like gpt-exp or gpt-codex2

[plucky sparrow]

Or gpt-benchmaxed 😅

Calling it 5.2 was a huge mistake because it's a downgrade in many ways

[proud bobcat]

I love the ai community 😭

[proud bobcat]

It’s O3 but with the GPT 5 database

[plucky sparrow]

Hmm. Different prompt? O3 loved to do that -> -> thing

[proud bobcat]

I’m assuming that 5.2 was an experimental model that they were toying around with (diff architecture with new database) and then gemini 3 pro came out and they panicked, fine tuned it for like a week, and pushed it out

I’m sure they have a lot of experimental models like that

[proud bobcat]

Probably so

O3 architecture + new database and 5.1 system prompt

It would explain the high hallucination rates and the xhigh modes

[zealous sparrow]

defo

[signal latch]

wth is goin on

lol

[dark nimbus]

can any one help me

after some time with claude it shows rate limit

[hollow ivy]

You need to use the battle mode with a (google) account, for much more generous usage of models.

Iirc, battlemode+account has no ratelimits, but am not 100% sure of that.

gpt-5.2 is trivial to discover in battle mode (just ask the model for its exact name and version)

grok4 also is easy to identify

the others you can get by testing them in direct chat first and see how they answer

[compact flame]

Guys did gpt 5.2 get any better?

Or it's still mid

[olive spruce]

The responses are much more shorter is what I’ve seen

[patent aspen]

Still mid

[golden ocean]

https://media.discordapp.net/attachments/990039516620390400/997631476554285186/ezgif.com-gif-maker_-_2022-07-15T183109.613.gif

[signal latch]

benjimon franklin lol

[turbid ether]

why did "GPT-5.1 Codex Max" get removed from the leaderboard?

wasn't it the top for coding before it got removed

[vernal saddle]

@echo aurora Why the Reve models were removed from LMArena?

[echo aurora]

I'm not sure, will look into and if it's something I can share I'll keep you updated.

[echo aurora]

Welcome welcome

[echo aurora]

There isn't a way to get past rate limits 🙁

[dark nimbus]

ok

[gloomy sky]

why was gpt-5.1 codex max removed

@hardy swallow @mortal vale

[neon idol]

Wth are these fakes benchmark?

[gloomy sky]

??

[neon idol]

where is gemini 3 in this benchmark

[gloomy sky]

its from web.archive

but gpt-5.1 codex max was removed

wondering if it will be added back

[neon idol]

its a fake benchmark in my opinion

[neon idol]

these are fake in my opinion

[cloud zinc]

its real

[neon idol]

gemini 3 is absolutely better btw 5.1

[cloud zinc]

this is 5.1 codex max high

not regular

[neon idol]

still beat it

[cloud zinc]

no its #1

[neon idol]

nah

[cloud zinc]

see the benchmark

[neon idol]

these are fake believe me

[neon idol]

fake benchmark

[echo aurora]

IIRC we had some latency issues on our end with that model and decided to take down. Would also note the moderators should only be pinged for moderator related issues (things that break our server rules). For questions/feedback you can directly come to me instead.

[cloud zinc]

these are true benchmark

5.1 codex max is better than gemini 3

[neon idol]

bro test gemini 3 and this 5.1 max high

[gloomy sky]

Okay, sorry, thank you, is their plans to add it back in the future?

[cloud zinc]

i did, its better in my test

[neon idol]

then u will say me if is better or no

[neon idol]

what was test?

[cloud zinc]

coding test

[echo aurora]

I believe so, but I'm not 100% sure sorry tosay.

[neon idol]

prompt?

[cloud zinc]

did u also test it?

[dapper grove]

hi

[compact flame]

I just got video arena on web

I mean only on incognito tab I got lucky

Weird

[cloud zinc]

i dont see it on incognito

[compact flame]

Dunno I somehow got it

Though on my normal tab I don't have it

[cloud zinc]

there is also automodality being released

[compact flame]

Yo I got the same

But I closed the tab with it

I guess it's gamble based on a guest?