general

[cloud zinc]

yes

[compact flame]

Got the auto again

[echo aurora]

Yeah this is an experiment, meaning a small percentage of people are going to be seeing it currently. cc @cloud zinc

Also same with auto modality.

[proud bobcat]

5.2 still get mogged in every category by gemini

LONG LIVE GEMINI

[compact flame]

Oh okay

Now I got some type of new ui

Instead of icons it's like a dropdown

[neon idol]

what is it?

[proud bobcat]

benchmark unequal to general intelligence

people wont use gpt 5.2 because its too pricy

if gemini 3 pro is basically still the king why pay more for gpt

[cloud zinc]

nobody is king

[plucky sparrow]

Gpt useless 🤣

[mild granite]

cherry picking

[hollow echo]

👋

[echo aurora]

[haughty jetty]

👋

[cinder gull]

hello everyone what's up? I want to know how to face swap a pic with her face and her into a video on Instagram from another lady, thanks

[zealous sparrow]

Speaking of removed models, will speciale ever make a comeback? It's been a long while since it got deleted.

[mild granite]

deepseek 3.2 speciale?

[zealous sparrow]

yea

[mild granite]

has it ever been on lmarena?

[zealous sparrow]

yes but it was deleted the same day it came to lmarena

[mild granite]

oh

[tardy plover]

grok removed?

[whole sundial]

I also heard some complaints about reve-v1 and reve-fast-edit being removed, it was replaced with the stealth models epsilon and epsilon-fast, some people would like to be able to select these models again

[zealous sparrow]

are they still reve v1 tho

they might be updated

[whole sundial]

it's an upgraded model, but I know its removal and replacement with an unselectable model upset some people

[zealous sparrow]

It was prob the company decision, not LMArenas

[mystic sluice]

is there a limits for gemeni-3-pro?

[plucky sparrow]

Haha this is beautiful

[echo aurora]

Yes, all models will have some kind of rate limit associated with them.

[zealous sparrow]

peak new model name @whole sundial

its a textarena model

i hope its not amazon

[agile bloom]

so best overall ai model is gemini 3 pro for December 2025?

[golden ocean]

erm sydney

artificial neural network vs biological neural network

[cloud zinc]

gpt 5.2

[agile bloom]

frr?

[cloud zinc]

yes

[agile bloom]

wow

is there any way i can upload a txt file to gpt 5.2 on lmarena?

[cloud zinc]

lmarena doesnt allow pdf upload. you have to upload txt manually

[empty stump]

i dont like 5.2

i will have to try it in api

[agile bloom]

my txt file has 40,000 words in it, i wanted to use it to analyse using gpt5.2 as a txt file instead of the entire thing as text input

[frosty lava]

yes and we're definitly improving everything your talking about ? each new month a new model with better capabilities came out

and it wont stop in 2026 for sure

less hallucination, follow task much better, better at coding

like everything we want its what they're working on

[lunar glade]

anyone know any site that have nanobanana 4K model?

[cloud zinc]

yes aistudio

no it is a scam

scam models

[empty stump]

sota in what

[weary galleon]

I agreed. They are not look too big.

[weary galleon]

Nobody likes it. Even Scam Altman.

[cloud zinc]

5.2 is #1 model

[empty stump]

in benchmarks

[zealous sparrow]

5.2 benchmaxxing

[weary galleon]

It is designed for benchmarks only, not for real tasks.

[cloud zinc]

yes for real task also

[weary galleon]

It's a banchmark.

[empty stump]

that openai made...

[weary galleon]

GDPval is a banchmark.

[cloud zinc]

for work tasks

[empty stump]

of course a model by openai will perform best on an openai made benchmark

[cloud zinc]

show where its bad

[weary galleon]

But still a banchmark.

[weary galleon]

Agreed. They design banchmarks to win in them other models.

I will laugh when it will get lower place in Text Arena than GPT 5.1

🤣

[cloud zinc]

so what are u basing ur opinion on

[weary galleon]

Truth.

Banchmark is a banchmark. Period.

[cloud zinc]

ur truth is a lie

[weary galleon]

Is banchmark a banchmark or not?

[weary galleon]

I won!

[cloud zinc]

https://tenor.com/view/tom-senya-gif-26024901

[weary galleon]

This clown is GPT 5.2, not me. It lied for my demand, is it a smart output?

[echo sinew]

Hello! Let's keep disagreements respectful and friendly.

[weary galleon]

Highly agreed👍

[cloud zinc]

yes, gpt 5.2 is best

[weary galleon]

OpenAI got 500 billion dollars from the US government, Anthropic got nothing, Google got nothing and Gemini 3 Pro and Opus 4.5 both outperform GPT 5.2. How is that possible? Scam Altman's mismanaging of the company is the reason.

[cloud zinc]

us didnt give openai money, u are wrong

us only approved the stargate project

[weary galleon]

By the way, right answer is yes, GPT 5.2 said wrong answer, like always.

[weary galleon]

This Stargate Project is a gift of the most powerful AI farm to OpenAI.

[cloud zinc]

not a "gift"

[weary galleon]

Gift

[cloud zinc]

its called investment

[weary galleon]

The same

[cloud zinc]

gift to google

[weary galleon]

Contract isn't a gift. Investment is.

[weary galleon]

Even if a gift(but it's not) 500B and 200M is 2500X difference.

[cloud zinc]

500 billion over several years

[cloud zinc]

google can't get that much gift cuz they are bad

[weary galleon]

Google to Anthropic and US government (taxpayers money) to OpenAI. Are you feeling the difference?

[hollow flicker]

is it just me or do you guys think that most people use LMArena to get all the paid ai's for free

[weary galleon]

And taxpayers are good 👍

[weary galleon]

Yes

[cloud zinc]

not us government, its softbank

[weary galleon]

But some like me, don't

[weary galleon]

Anyway Scam Altman will get those money. Anthropic and Google will not.

[cloud zinc]

cuz scam google are bad

[weary galleon]

Why?

[cloud zinc]

cuz gpt 5.2 is #1

[stray aspen]

Lmao

Scam altman

[weary galleon]

Let's go!

Vote, guys! Democracy will win!

[empty stump]

its bad because of the safety

[cloud zinc]

where is it? u are making a rumor, not actual reality

so far with my testing, it looks bad

[rapid merlin]

Gemini hallucinates the fact it has dall e 3? Interesting....

[echo aurora]

Is it just me, or are others experiencing a bug when trying to login? Essentially, after doing an email login, after entering email/password, clicking the login button doesn't do anything. Are others seeing the same?

[cloud zinc]

yes

[echo aurora]

Thank you. Will share with the team.

[weary galleon]

Maybe because this "test" is paid by Scam Altman?

[weary galleon]

60% say bad, 40% say good. GPT 5.2 is feeling so bad.

[cloud zinc]

how many votes?

[cloud zinc]

u voted with ur alt ok

[weary galleon]

@cloud zinc If GPT 5.2 is #1 as you said multiple times today, why members of LMArena hate it so much?

[cloud zinc]

no one is hating

[weary galleon]

Proofs!

[cloud zinc]

yes u used ur alt to vote

vote is not over yet

23 hours left

[weary galleon]

Give proofs instead of repeating false accusations!

[cloud zinc]

vote not over 23 hours left

[weary galleon]

See you after 23 hours! Bye! 👋

[weary galleon]

Look my previous poll
#general message

[cloud zinc]

ur poll is rigged

[weary galleon]

Give proofs, false accuser!

[echo aurora]

Hey going to ask we move on from this conversation.

This doesn't seem to be very productive and is just escalating a bit here and there.

[weary galleon]

He said I vote from my alts in my polls with zero evidence.

Just accuse, accuse, accuse, again and again! Without proofs.

[cloud zinc]

give proof u are not lying

[night kelp]

can anyone help me a bit?

[cloud zinc]

yes

[weary galleon]

Troll! 😡

[cloud zinc]

so u have no proof

[weary galleon]

I added you to block list. Bye! 👋

[cloud zinc]

thought so

[night kelp]

im working on something with python, been workin on it for a few days, then the chat decides to just stop, not being able to send emssages (i mean like i can send but it gives me that annyoing red prompt "something went wrong with this response, please try again") i refreshed, restarted, tried typing again, same thing

using ai ofc

also gemini 3.0

[cloud zinc]

u need to start new chat

[night kelp]

i have almost the whole stuff on that specific chat, any other way i can restore it?

[cloud zinc]

that chat is bugged

always backup ur files after couple of hours

[torn mantle]

@cloud zinc what did i miss

any new model

[night kelp]

how do i even back a chat up anws

[torn mantle]

any new model

kaiser

[night kelp]

uhh

[stray aspen]

Why can't gpt 5.2 extra high think of another way of designing stuff

It all looks the same

[night kelp]

gimini

[stray aspen]

Wheres creativity

[cloud zinc]

flash tomorrow

[torn mantle]

how do u know

[cloud zinc]

u will see

[stray aspen]

Yeah it cooked for me but I just don't like that its not that creative

[torn mantle]

bet

[cloud zinc]

watch for logan tweet tonight

[neat apex]

Thats weird since even Qwen3 high manages that

[torn mantle]

imma sleep early

_<

[neat apex]

Likely because the gpt is sooo formal

So it will rarely try outshine, besises it writes well when you ask something exactly

[ocean vortex]

they copied the "high" thing from OpenAI? 🗿

[neat apex]

😂

[cloud zinc]

yes, the word "high" is copyrighted by openai

[golden ocean]

gpt 6 latent space reasoning

[neat apex]

I think its general knowledge

[ocean vortex]

That's not what I meant. Don't pretend you aren't very smart. Although maybe you...

[neat apex]

I hope

[ocean vortex]

I mean it's reasonable to assume they have bigger release in the works

there was less than 1 month between 5.1 and 5.2

this was like a small incremental update

[neat apex]

Thats not what he meant. Dont pretend you arent very smart
Although maybe you...

[ocean vortex]

No clue why they tried to oversell it this hard lol

Like pushing for benchmarks this hard with 5.2 wasn't necessary I feel like

[cloud zinc]

true

[ocean vortex]

5.2 you mean?

I don't think it is tbh

have you seen SimpleBench and other stuff?

5.2 gets beaten by both 5.1 and even more so by 5.0 there lol

That's the thing, my experience was the same. 5.0 > 5.1 and 5.2

And then Gemini3 just somehow manages to score great everywhere they test it at

Not definitively the best in select things perhaps, but not really underperforming anywhere either

[neat apex]

5.1 is way better than 5.0

Almost same smartness, but it actually efforts to make better responses

[steel dune]

Hi everyone, a quick question “Which AI model will one recommend to aid in accounting work?”

[echo aurora]

Hello would encourage you to check out our Text Arena leaderboard using the Business, Management, and Financial Ops Occupational filter -> https://lmarena.ai/leaderboard/text/industry-business-and-management-and-financial-operations

[steel dune]

Thank you

[cloud zinc]

@torn mantle nice

[thorn path]

Wait the text leaderboard updated and 5.2 isn't even in the top 10 lmfao wth

[echo aurora]

gpt-5.2 isn't yet on the Text leaderboard

[thorn path]

That makes much more sense, between you and us where do you (personally) feel it's going to fall 👀 (if you're allowed to say, no worries if you're legally bonded from doing so)

[burnt sinew]

anyone notice gemini 3 is hallucinating thinking it has experiences?
"I personally have a database with over 800 entries."

[burnt sinew]

when will it be?

if you can share

[echo aurora]

I'm not legally bound to discourage from sharing my personal opinion, but I'd prefer not to as some may interpret that the wrong way.

[echo aurora]

TBD. Sorry to say I won't be able to give an estimated time.

[burnt sinew]

i think it will score highly, though it does terrible on some edge cases

fails seahorse test

[echo aurora]

Where do you think it'll land?

[lucid geyser]

Really flipping it around

[swift oyster]

LIVE BENCHMARK UPDATE
Model: Hawk (Launch 25th)
We're currently halfway through the official ARC-AGI-2 benchmark - one of the hardest AI reasoning tests in existence!

Current Stats (48% Complete)
Correct: 14
Incorrect: 44
Accuracy: 24.1%
Progress: 58/120 tasks

Key Takeaways:
Outperforming Claude Opus 4.5 by 10%
Currently ranked #9 globally

[surreal creek]

“legally bound” bro it’s an AI benchmark site not classified information 😂

[sullen quest]

Can we make LIVE BENCHMARK UPDATE a meme?

[cloud zinc]

3rd

[thorn path]

realistically 2nd, but it could really go either way depending on how varied opinions go imo

[burnt sinew]

i think only o3 and opus 4.5 thinking gets seahorse question right?

[jade egret]

gemini does too i think?

[burnt sinew]

no

not without search i dont think

[jade egret]

gemini 3?

o

lol

[brisk turret]

Where is 5.2 on the leaderboard?

Wtf is going on

[jade egret]

Not on it yet

Must be tomorrow?

[tired plaza]

guys, what are all the site's keyboard shortcuts? is there a shorcut for new chat?

[astral bloom]

@echo aurora when will the video arena be accessible, I've seen people here claim they had access to it.

[frosty lava]

when will we get an ai with like 1% hallucination

that's all im waiting for

ofc its good to have genius ai but what if they always do errors

give me actual genius ai if you want but that are capable of actually doing things without mistake

[cloud zinc]

2027

[frosty lava]

you think ?

Cause like in 2026 we're supposed to see alot of new ai like grok 5 and all

and they're supposed to be like much better

[cloud zinc]

overhyped products

[frosty lava]

we'll see

no one know actually

[delicate wagon]

Hi everyone

[obtuse smelt]

hello

[sullen quest]

?

@astral bloom video arena is right here in discord, is a discord exclusive feature, go to the channel how-to-video-bot for more info

[cloud zinc]

there is video and auto modality being experimented in the site

[obtuse smelt]

hmm is happening
Something went wrong with this response, please try again.

[left lodge]

Anyone wanna search with images?
I have a way.

Literally just go to lmarena.ai
Add images
Switch to search modality
Send Prompt.
Done.

I already reported it in #1449482775823515688
So act fast , :p

Also, Lmarena is working on 2 new possible updates;

A new modality picker

And

A new Video Modality in web
which will be available at
https://lmarena.ai/c/new?chat-modality=video
And
https://lmarena.ai/?chat-modality=video

Enjoy the information 💁

Currently these links are redirected to lmarena.ai
And video generation requires login.

[cloud zinc]

how do i make video on site?

[sour spear]

Hallucination is baked into Grok by design...

[torn mantle]

how did u know

so sus

[mild harness]

hi, how can I use gpt 5.2 xhigh on lmarena? I only see 5.2 high. thanks

[torn mantle]

not available

[vivid coral]

mother of God

[left lodge]

I got access randomly, but it is not currently available.
I tried but it required login and upon login it went away.

[keen beacon]

Does anyone know why it no longer displays images in 1920x1080?

[obtuse smelt]

is just 768 x1360

[keen beacon]

Yesterday, if you asked for them in 1920x1080, it would do that.

[frosty lava]

what do you mean its not only grok every llm

theres actually no llm without hallucination

[obtuse smelt]

oh i see

is gemini error ?

[weary galleon]

🏆Sonnet 4.7 is coming... until Christmas ⛄

https://tenor.com/view/cheers-cheers-to-that-wink-steve-carell-gif-23705252507487769

[sterile tartan]

I had thought it would be 5 but 4.7 works too

[weary galleon]

I too

[sterile tartan]

Expect quicker releases now from all companies

Thr competition is peak cutthroat now

[weary galleon]

AI winter is coming... My expectations are Anthropic and Google will release lots of great models under this competition.
P.S. Maybe xAI also, I am not sure.

Also, AMD and Nvidia will release new AI chips in January, which also accelerate this rush dramatically.

Googles and Amazons chips are also great 👍

My expectations are after Sonnet 4.7 release, it will get 🥇 first place in coding, Opus 4.5 🥈 second.

[compact flame]

I doubt google will release any models soon

They gonna be either focused on other stuff or just training the new models for long time

[weary galleon]

I'm sure no one tech giant is focused on any other things.

Grok 4.20 is coming until Christmas ⛄ also. Not sure how much good it will be.

[frosty lava]

Do you guys think ai will really replace like almost every jobs ?

many people is saying this is stupid and will never happen

but

why tho it can definitly happen

[compact flame]

Well it's based

Animators? Could be after I seen sora

[frosty lava]

then why majority of people still think it wont happen isn't it like crazy

[compact flame]

They just can't accept that ai is getting too good

[frosty lava]

how can we ignore it when we see the massive investment and the literal race

for ai

[compact flame]

Well some believe the big ai advancement will slowdown someday

Its kinda based how well they train ai

[frosty lava]

there's always a solution

[compact flame]

Like how do you think when ai will just reach its limit and stop advancing?

[frosty lava]

It will just be a problem to solve not something impossible to achieve, people be like its impossible yet every time human do new discoveries and what felt impossible is just a reality

there's always someone that find a new way

there will be some problem but saying we can't ever solve it feel stupid

in my opinion

[weary galleon]

Opus 4.5 Thinking 32K + prompt engineer can replace ANY programmer easily.

[frosty lava]

Not really due to hallucinations

we just want less hallucinations and yes

[weary galleon]

Prompt engineer is a human, they will give prompts to Opus until it would give an output without hallucinations.

[frosty lava]

your right but are we already at that point where it will be like as efficient as a team of programmer, i don't really know at all

i think definitly in 2026 what your saying will be doable

we're close for sure

[weary galleon]

I'm a programmer, and that's why I think LLMs are bad for us. To find a job is much harder for me even now, if to compare a few years ago.

[frosty lava]

its bad cause we only see what's happening right now, but in the future we won't just let people having no jobs and like die

it will just be very very different

it can definitly help us all if we think about medecine, new discoveries, and future

its just that theres this questions about jobs and what will happen but no one can answer already were not here yet

but for sure, we won't let people struggle forever

[weary galleon]

Only so groups of people, all people with intelligence work, people with physical work will feeling great.

[frosty lava]

Sorry, english is not my native language and im not really sure about what you try to say

Physical work might also get replaced

[weary galleon]

But not by LLMs.

[frosty lava]

No, but by robotics maybe ? which is also growing much

In china its already happening

[strong wren]

hi

[weary galleon]

Current LLMs are very good, current robots are not good. They will be able to do all physical works, but not in near future.

[frosty lava]

Already done in some factory in china, with this ai race its not only about LLM

there is also robotics going with it

and yes they are doing alot of progress

I would say we're really into automation these days

everythings is about it

we want more efficient, cheaper worker

and its doing alot of progress

You can look deeper into it

its happening.

[viscid cloak]

nanobanana pro DIRECT keeps giving“something wrong”, anyone as well?

[obtuse smelt]

yeah gemini looks error

[weary galleon]

Robots waiters are good.What about robots plumbers, electricians, carpenters, builders, etc.? It's impossible.

[frosty lava]

impossible ? no its not, its just not done yet, like i said we're doing alot of progress in it you really can look into it you'll see its not a dream or something impossible

im not saying its already done, but definitly it seems that it will happen

and not like in 50 years

that's definitly something else than LLM

but its also doing progress

[weary galleon]

I think even more. If robot (even futures one) would repair your electricity panel your house would burn.

[frosty lava]

there is a problem, and no problem is impossible to solve

like we definitly won't make something that kill you

or is dangerous

[weary galleon]

We need 50 years or more to solve this problem.

[frosty lava]

wow that's a big numbers, we're progressing faster than that

you know how the world were 50 years ago ?

its changing much faster than that

[weary galleon]

You're not an admin. How did you get insider information?

[frosty lava]

But we shouldn't see only the bad side of it honestly

nothing is all good or all bad

but from what we saw in history, when we progress its mostly good

[robust sluice]

yes and error is count on limit to me

[obtuse smelt]

need fix this

several hours is look error

[robust sluice]

mine its been 3 days

Gemini never appear in Battle only Flux models

[obtuse smelt]

hmm

[robust sluice]

and Direct keep error and said try again in 50 min

[obtuse smelt]

hmm

[plucky sparrow]

not 'forever' but don't be so sure the rich will just help people out

the question is whether you will end up living the majority of your life in hell, or abundance

[frosty lava]

it won't depend on one person but on the government, i can't say hey everythings will be good, but i can definitly say if its bad anyway it will also be bad for them in that case

[plucky sparrow]

the great depression was a pretty long period of time

[frosty lava]

its a win win

[plucky sparrow]

it will take a long time

[frosty lava]

if its bad for most people then its bad for them too

so it most likely won't be bad

[plucky sparrow]

at first they'll be like "oh, but, AI is creating new jobs. we'll just create more gov jobs."

they'll delay it as long as possible

we might even get a revolution before we getr change

historically, that's how it plays out

[frosty lava]

yes theres a time for it to be good actually they will delay it your right, but definitly the worst case will not happen

[plucky sparrow]

yeah i agree. to be honest, worst case i'm worried about is, the rich people are in charge of AI right now

[frosty lava]

Im not sure of what your trying to say cause ai are made by rich people already

[plucky sparrow]

and the rich people have access to the best AI models. not worried about an AI 'killing us all' as much as 'rich people having great influence over AI' and using it to manipulate people

imagine a 100x better claude opus 4.5, but only the rich have access to it, adn they use it to exploit you

[frosty lava]

that's the worst case, but it won't happen if it do you can already tell it will be a war

i don't believe we all just gonna accept something like that happening and they know it

[plucky sparrow]

sure, but, like all wars, it's usually won by the people with the most power

can you win against a robot army?

i think we see great famine before we see great abundance

[frosty lava]

why will they use a robot army, why will they want to be hated by everyone, and that mean like every country have to do it

[plucky sparrow]

historically, that's always happened. not sure this time is 'different'

[frosty lava]

if everyone like everyone do hate them they are loosing

they are human too

they don't want human to die

and nobody will let someone doing it

[plucky sparrow]

have you not seen how many humans were enslaved historically?

[frosty lava]

Yes, but like what your saying is something different actually its not the same weight

[plucky sparrow]

we're living in less than 0.01% of human history. we forget how bad things have been

[frosty lava]

if something like a robot army controlling everyone happen humanity is dead

would mean even them at some point will die

[plucky sparrow]

i don't think it will go on forever, i'm just saying, historically, we go through incredibly rough times before we get good times

and we forget, as a species, how good it is now, compared to how it was

and that cycle has repeated itself for as long as history was written

[shrewd citrus]

Anyone else having issues with nano banana pro

[plucky sparrow]

so either this time is different, or it's not

[frosty lava]

Why do you think its the same thing ? when we created pcs and phone and everythings no such thing happened

not that bad

atleast

[obtuse smelt]

yeah is seem error

[plucky sparrow]

the industrial revolution, the great depression, the last 2 wars..

[frosty lava]

But also people getting smarter with more right

you can't compare something that happened when like everythings was different from now to now

[plucky sparrow]

this is a very small % of human history. i'd argue we're still in the age of abundance, but we're on a downward curve into the age of famine

so you're saying. "this time is different"

it could be. but it rarely is

[frosty lava]

it always are different, cause something like that happening in the world of today doesn't seem achievable and if it is like everyone will loose

even them

[plucky sparrow]

yes, i don't think it will go on forever

i think history never repeats itself exactly but it often rhymes

i think we will get an age where a lot of people are in famine because of AI, and eventually a revolution of some sort, and eventually we will have abundance

[frosty lava]

they need to be popular too

[plucky sparrow]

the only question is whether we end up living in the majority of the famine and barely see the abundance

[frosty lava]

to be popular it also mean they have to do good things somehow

[plucky sparrow]

like many people who survived through WW2

we might end up telling our grand kids, "Back in my day, AIs made us all poor.."

[frosty lava]

WW2 didn't started cause suddenly we got new technologies

[plucky sparrow]

no, it started because of famine in Germany

[frosty lava]

so its completely different

not even the same reason

Ai are new technologies

[plucky sparrow]

we'll see. you're banking on the government being actually controlled by the poor majority instead of manipulated by the rich minority

[frosty lava]

yes cause like you said majority and minority, and if the majority start to defend themselves the minority wont win

and they don't want that anyway

it would in every case make them loose

[plucky sparrow]

i think what you're saying is right, i just think it'll take a lot longer than we'd like

[frosty lava]

your right too, people loosing job is not something avoidable now

but i dont know what will happen

i just know we can't go back anyway

[plucky sparrow]

yeah i agree

people somehow think protesting AI will get anywhere 🤣

[swift oyster]

Result are in for our upcoming model.

[plucky sparrow]

if anything, it's better the country you live in, develops AI before the country that may not share your same values

[sterile ore]

hello

[obtuse smelt]

hi

[frosty lava]

its like, when gun were created who will still use sword

its more efficient

and that's all we want

when people will realize they can't beat someone that use ai in the future

they will use ai

and you can't really blame them

[plucky sparrow]

imagine trying to protest the development of guns 😄

[frosty lava]

you just can't tell to everyone to stop

[plucky sparrow]

imagine if your country had no guns but every other country does 😄

[frosty lava]

and if someone still do it he will be more efficient

[plucky sparrow]

not even in the army

[frosty lava]

exactly that's why its not possible to go back when its proven its better

[plucky sparrow]

i do agree with you 💯 on one thing, protesting AI is stupid

it won't lead anywhere

except your own misfortune

[frosty lava]

we can't protest progress even if the progress is dangerous or will lead to big problem like mass unemployment

cause its somehow still a progress

all we can do is find solution but it doesn't even depend on us

but on few people :

all i can say is if its too bad the majority will defend themselves

so it probably won't be "too" bad

[plucky sparrow]

hmm

but things are 'too bad' in many countries

and the majority isn't able to defend themselves

[frosty lava]

its not equality everywhere, for the big countries it work like i said

[plucky sparrow]

i think that's why AI development is actually super important

it will determine whether your country is a 'big' country or not, and how long the famine will last

[frosty lava]

if ever a "bad" countries get the lead of the most powerfull progress then we're all at their mercy

that's why its a race

we're saying to know who will be the leader

of the next decades

but no one understand it atleast not yet

[plucky sparrow]

will be interesting to see who wins

might not be who we think at all

[frosty lava]

yes

we never know it can come from anyone

[plucky sparrow]

e.g. may not be US or China

[frosty lava]

it will be the smarter one

that's it

[plucky sparrow]

might not even be smarter, could be luck

[frosty lava]

yes your right

we just can't tell

but its a race and not for nothing

its much more important than people think

[plucky sparrow]

one very interesting video I find is the time lapse videos of the world borders: https://www.youtube.com/watch?v=-6Wu0Q7x5D0

[frosty lava]

when china and us are racing how can people still say hey its nothing

i don't understand sometime

[plucky sparrow]

put it on 2x or whatever. the borders historically change like crazy, yet somehow, we think war is over

[frosty lava]

how delusionnal we can be

[frosty lava]

Yeah i know already too many things happen but we see it as small and not important when it actually matter the most

[plucky sparrow]

I do hope you're right, that this time really is different

[frosty lava]

if its not big enough to see it through eyes people will still think its nothing

[plucky sparrow]

otherwise in the next 7-30 years we're likely going to go throuigh hell 😄

[frosty lava]

yes and it can even be faster than in 7 years

by seeing at the progress of it this year for example

its crazy

all of it is being speed up cause its a race

[plucky sparrow]

well i mean, people predicted self-driving cars everywhere 5 years ago 😄

[frosty lava]

yes but theres a difference, cause self driving car doesn't mean power

when its about power you'll be surprised

how fast it happen

seeing Trillion dollars of investment lol

i don't even know who have this much money but yeah

its still happening

[weary galleon]

It doesn't look real. In my experience really good models don't need hype, they get hype automatically from their existence. You post these pictures every single day maybe because after the release hype will over.

[frosty lava]

Yes gpt 5.2 what a scam in my opinion

they did this so they don't loose fame but its definitly not as good as expected

benchmark are impressive yet we dont see real result

like student getting 100 / 100 on a test but when it come to apply it in real life nothing happen

[plucky sparrow]

that is true

[frosty lava]

president themselves are involved in this

its not just a new fancy technologies

its power but people be underestimating this

while Trillion dollars being invested

[hollow mist]

Tell me what the problem is. I've been nerfing everything for a long time, but now I keep getting the error "Something went wrong with this response, please try again." (Gemini 3 pro image preview nano bsnana)

Help please

[compact flame]

It's something with nano banana itself

It's not available on yupp either

[robust sluice]

someone told me error result doesnt count on limit, but I try over and over when limit resets, I found it really count, cuz this hour I got only error msg and limits still runs out

[obtuse smelt]

yeah me too is sme happening again

[robust sluice]

how I test is: generate something random and Flux model appear (I remember the style it draw cuz I see it a lot) with other Error model I keep clicking on retry with that error model and it count on 11s then error again, I do this untill it doesnt count anymore, that means limit runs out already

[hollow mist]

This is an update, the server just doesn't work.

[magic ravine]

Nano Banana Pro is not working again

[obtuse smelt]

hmm, well me too

[exotic iris]

Everyone you love this image or not

[storm summit]

any cods invite for sora?

[exotic iris]

Everyone likes 👍???

[pastel bone]

You can now use lmarena to generate videos(Open Chrome, whatever you like)If u are lucky enough!

[storm summit]

i can onlyy generate image and web ersearch and code. How to generate video

[pastel bone]

Asking pineapple, probably release in a few weeks/minths

[storm summit]

but why u can generate video and i cant?

[ocean vortex]

"almost same"... For me the style feels kinda forced. And it isn't as reliable on some tasks.

[pastel bone]

But you can try your luck and use #video-arena-1 1, #video-arena-2 or #video-arena-3 and #video-arena-4

Maybe u can use Sora👍🏻Good luck

[true harbor]

Is image generation not working at all in lm arena since yesterday? At least for Nanobanana

[magic ravine]

I don't see a video option.

[pastel bone]

Yeah I do it several hours ago and it maybe a leak

I was amazed

And I saw many models like kling😆

[magic ravine]

Damn

[ocean vortex]

with GPT5 they went all in for genuine performance. With 5.1 they went for response style, and then for 5.2 they had this "oh sh'it" moment franctically responding to Gemini3

[pastel bone]

I didn't have time to try it out actually.BTW, it is a good try to add video arena

[obtuse smelt]

oh well is work again

[ocean vortex]

that turnaround of less than 1 month between model versions is not normal tbh

We had similar with 2.5Pro 05-06 vs 06-05, but that was marginal differences and was advertised as such

[magic ravine]

The problem with video arena here is the randomness. I almost never get Sora or Veo 3. It's always the bad models

[pastel bone]

Yeah, the mods tell me that they may add a function for a model at random and a model you pick option

[weary galleon]

They scammed Microsoft when Microsoft invested $500B into Stargate and now Microsoft wanna get their money back because models of OAI are terrible, but they can't.

[magic ravine]

I hope so, but yeah, I'd prefer it on the site itself. Way more convenient and less traffic than in a public chat where it may get drowned by other gens.

[frosty lava]

they wanted to release a new model fast due to them loosing people cause of gemini and claude

but it wasn't the best idea

[weary galleon]

There is a feature "Direct Chat" for text, visual, and image models. It would be a solid idea to add it for videos too.

[weary galleon]

And all people see your inputs/outputs.

[weary galleon]

It was the worst idea. Ever. GPT 5.1 outperforms GPT 5.2 in real tasks. GPT 5.2 is just a banchmarks scam.

[frosty lava]

i guess they just lost due to their idea can they get back on the race ? i don't know honestly it depends on so many things

people will always use the best

fake benchmark doesn't get them anything lol

if it can't do it in real life

[buoyant crypt]

Nano banana is not working today

[weary galleon]

Trying not to lose short-term race, but don't care at all about long-term race proofs Scam Altman has negative IQ level.

[viscid cloak]

yea, about only 1/5 generation was successful

[acoustic crater]

1/50

[magic ravine]

Yep. All we can do is wait.

And hope they fix it

[plucky sparrow]

Probably because they're trying to rush out flash 3

[magic ravine]

Flash 3? How's it different from Pro 3?

[plucky sparrow]

Faster and cheaper for those paying

(High use people and api people)

And some claim it's better in some ways. We'll see today

[vale vortex]

Quick question, which one is better for research,

Kimi, MiniMax Agent, or Deepseek?

[compact flame]

Why these specifically though? There multiple different ones but probably deepseek though

[plucky sparrow]

I haven't used deepseek or minimax but K2 is actually pretty good for research

surprisingly good

[pale torrent]

How to use Sor 3?

[plucky sparrow]

#1397655624103493813

[vale vortex]

Thinking or Instruct?

[plucky sparrow]

thinking, on their site (not api)

[onyx shore]

go with k2 much better then those

[hollow mist]

Is there any news when nano banana will work?

[obtuse smelt]

hmm still not respond

[sterile tartan]

Might as well Try Tongyi Deep Research in Qwen

[obtuse smelt]

still like this "Something went wrong with this response, please try again."

[plucky sparrow]

thanks for the suggestion, i've not tried that

[sterile tartan]

Yeah try all best and judge by personal experience

Is the research very important?

[plucky sparrow]

what is 'low price'

[sterile tartan]

Cheaper then original

[plucky sparrow]

so not very cheap then

[sterile tartan]

Probably well cheaper

The black internet market of ai yeah

I have a a seat in ChatGPT business myself

[compact flame]

Seems like advertisment

Should we call mods or something

[sterile tartan]

I feel like it yeah

💀

[compact flame]

Ping mods while replying to the message ig

[sterile tartan]

Right

The Stage is Yours

[compact flame]

Ig I'll wait can't test if we can ping mods or not though

[sterile tartan]

U can atleast ping pineapple afaik

[zealous sparrow]

gemini flash 3 is defo today

google put out another model onto battle LMArena

Updated Ghostfalcon and Fiercefalcon

[obtuse smelt]

what the

[sour spear]

I'm so curious to see how Flash 3 will perform.

[plucky sparrow]

preliminary reports are saying 'the sonnet to the opus'

we'll see soon enough if it is

[plucky sparrow]

don't think this is google. doesn't make sense for them to release a codename trial model on the same day as release

[compact sleet]

It felt like a Google model to me, could be wrong though

[sour spear]

If Ghostfalcon is Flash 3, then OpenAI is in even deeper s*** than they already are. I just got it in battle mode, and it did a fantastic job

[zealous sparrow]

google got the 3/3 on my questions after i changed them up a bit

btw

no AI could 3/3 this

i want to see if Xhigh can after i changed em up a bit

[compact sleet]

Yeah guess sundar wasn't lying it's their best model yet on that specific interview.

[zealous sparrow]

ima test xhigh rq

also deepseek r1 turbo got it right

so if xhigh doesnt

OpenAi is cooked

[fiery gull]

I hope it's good, I'll have to manipulate 300gb of documents in anti gravity, I need flash 3.0 to be good in agentic mode

[zealous sparrow]

gpt 5.2 xhigh officialy lost to deepseek r1 turbo and the new gemini model

2/3 because of the Macy question

[fiery gull]

I cant use gpt 5.2 xhigh, my works it need more that 500 sec and crash

[zealous sparrow]

if you argue fell behind means she fell off the hill it aint, it means that she is in the back of the hill

[zealous sparrow]

im usin on yupp

[acoustic bolt]

Guys work the server on lm?

[neat apex]

Gpt 5.2 trys a lot to not allucinate and ends not reasoning many things

[fiery gull]

Too

[zealous sparrow]

btw r1 turbo and ghostfalcon-20251215got it right

[fiery gull]

How so?

[acoustic bolt]

Gemini 3 pro

[neat apex]

When R1 was released it was the best model sometimes by far in everything, yes?

[zealous sparrow]

the discord designs it made had me impressed

it was then NERFED

a f- ton

It was able to recreate websites

[neat apex]

Yeah, i am not tweaking

He was a burfed 4o and near to level of sonnet 3.5 but with a reasonable reasoning

[compact sleet]

It made deepseek moment happened, if that is a thing lol

[dull mason]

5.2 on text leaderboard wen

[zealous sparrow]

eventually

[neat apex]

He takes too much time answering, it will take weeks lmao

[dull mason]

I wonder how "code red" will hold up against Gemini

[compact sleet]

Do lmarena have a predecent of having a model avaliable on the direct select chat, but that model is not on the leaderboards yet?

[fleet lintel]

ghostfalcon is 100% flash

[sour spear]

At this point, I wouldn't be at all surprised if Flash 3 would beat 5.2 High.

[neat apex]

The fact open ai makes a benchmaxxed model, to then make it actually smart is very true, it is very noticeable in 4o

[fleet lintel]

I was almost certain that Flash 3 is tomorrow .. umm

[compact sleet]

I feel like it can be used as boosting purposes, if the side by side and battle votes are not separated.

[zealous sparrow]

Year Summary VS
OpenAI Vs Gemini
OpenAI

• 2 new models that they said were good
• Both flopped
• Benchmaxxing king
  Gemini
• 2 new models
• Both good
  Gemini no diff

[pale torrent]

Veo 3 please

[zealous sparrow]

tuesday

[neat apex]

Since he does many works perfectly, and loses perfomance gradually when the task is more different

[zealous sparrow]

They always release on a tuesday

[neat apex]

Its sad that gpt 5.1 had a relative flop

Its gpt 5, but actually efforts to make a better answer, it is great

[zealous sparrow]

Google also gave us the best OCR this year

[fleet lintel]

yeah, Logan tweet also suggest that it will happen today. Good thing that I dont gamble, otherwise I would have place bet for Wednesday 🙂

[zealous sparrow]

It will take a LOOOOOOOOONG time till someone beats gemini pro 3 OCR

[dull mason]

Yeah it's teacher model is Gemini 3 Pro after all

[neat apex]

Since OCR existed, Google always had the best one, have the best one and likely will always have the best one

[fiery gull]

I think will be gemini 3.5 lol

[compact sleet]

This is a no brainer, they been doing this since Google lens.

[zealous sparrow]

someone except google

[compact sleet]

And probably no one can't beat them.

[zealous sparrow]

yeah google lens was good

I also used gemini Live from the app and told it to identify things in my room

it didnt get one wrong

[neat apex]

Well, Qwen3 Omni had a level close to Gemini 2.5 and runs in a phone

[compact sleet]

Hmm, ofc China had to have one lol

Perhaps you're right.

[neat apex]

There was a company the Prime was not in the 3rd model?

[weary galleon]

GPT 5.2 is very bad.

[sour spear]

True. Like Geralt and Ciri. 😉

[zealous sparrow]

I want to know Flash score on simplebench

I hope its high

[obtuse smelt]

well yeah need fix

[plucky sparrow]

i don't get how some people like gpt 5.2

are they all paid shills?

[neat apex]

Primes are in the 3rd model?:
Gpt 3/3.5 ✅ (Gpt 4 was not way better against Claude 2)
Grok 3 ✅ (Grok 4 fast was great, but it suprassed the 1400 points)
Gemini 3 ✅ (By far)
Claude 3 ✅
Qwen 3 ✅ (By far)
Mistral 3 ❌ (The first Mixtral 22bx8 were a beast, yet nowadays they barely manage to be competitive)

[tiny halo]

we need claude has image import

[plucky sparrow]

i really tried to like it, i tried various things on it, but, it feels like a model from 1.5 years ago

[neat apex]

I know some people can be frustated about gpt 5.2, but you are isane

What are you dooing with that?

[fleet lintel]

Are you using Plus tier or Pro tier?
Plus tier is terrible.. you get medium gpt 5.2, which is like gemini flash models. They are treating their paid customers very badly 🙁

You need to use Pro or API to get the best out of gpt 5.2

[compact sleet]

Ouch.

[neat apex]

Ah yes, gpt 5.2 medium

[plucky sparrow]

i'm using it from api, even xhigh. i haven't tried 'pro' but i'm guessing it's similar to xhigh

[neat apex]

Its only gpt 5.1 medium, but allucinates sigthly less (to the unoticeable level)

[fleet lintel]

xhigh should perform .. what is your usecase?

[plucky sparrow]

maybe 5.2 medium. but i find 5.1 way better

[neat apex]

Why non Gpt 5.2 xtra high even exists? Lmao

[plucky sparrow]

i tried it with coding, i tried it with logic, tried it with medical text, tried it with long context

[sour spear]

If you look at 5.2 isolated, it is a really good model. But it's not excellent. Plus It's behind the competition, and above all, it's excrutiatingly slow. So all in all, it's neither fun to use, nor impressive in any way.

[plucky sparrow]

good at what?!?!

like can someone give an actual example, prompt and output?

[neat apex]

If you use xtra high, it does very very good analises in documents

Sometimes better than opus 4.5 or Gemini 3

[plucky sparrow]

give an example please

because every time i've tried it, it's worse than all the other models

[neat apex]

My brother creates documents using Gemini 3, and when go fixing it with Opus 4.5 and Gpt 5.2 xtra high

[plucky sparrow]

i might be using 5.2 medium, if xhigh is 5.2 medium

what kind of documents? what did it fix that opus 4.5 couldn't?

[neat apex]

Most times 5.2 notices way more issues than opus 4.5

[plucky sparrow]

proofreading?

[neat apex]

I dont have it now xd

[plucky sparrow]

ok but what kind of documents and what kind of errors?

[neat apex]

But yes, you are right, besises xtra high, it looks be worse than gpt 5.1

[plucky sparrow]

i had xtra high try to create a game for me. it was terrible

not functional

or barely functional, rather

[neat apex]

Documents in general, he find subtle erros and mistakes, and even give good fixs plane

[plucky sparrow]

please be more specific than in general because every document and information i've fed it, it acts like it has low context

and it's all confused

and the few times it suggests things, it's wrong

my specifics are html/js code, transcripts, and medical text

it's failed at all three compared to other models and even gpt 5.1

i'm very confused what it excels at

[sour spear]

You have to use standardized benchmark tests, it's only optimized for that. 😉

[plucky sparrow]

🤣

ok that is fair, i've not thrown any standardized texts at it, just actual use cases

[tiny halo]

omg why all claude moldes does not have image import

[worn flume]

Hey

I'm giving away an apple 🍎

[lunar glade]

can we all agree that LMArena's banana pro is spoiled? just can't get any result beside other than "Something went wrong with this response, please try again."?

[sour spear]

But it is a little crash prone atm, needed two attemps to work. 😉

[tribal kernel]

Hi everyone, I'm using Arena, but it always gives me this error, how can I fix it? Thanks

[magic ravine]

Yep. It's been happening all day.

[sour spear]

They're probably busy implementing Gemini 3 Flash Image. 😉

[echo aurora]

Sorry to say we've been experiencing a higher than usual error rate with this model. It's also possible that you're hitting a rate limit.

If you haven't already would recommend to try: hard refresh of site, clear your cookies/cache, and if no luck starting a new chat. This may help.

[tribal kernel]

Thanks for the reply, no.... I use the service 3 times a day so I don't think I exceeded the limit, also because the model tells me to wait 50 minutes, but it didn't tell me anything.

[echo aurora]

the model tells me to wait 50 minutes, but it didn't tell me anything.
It's a bit confusing but both error messages can both be caused by rate limit.

[echo aurora]

Can you: open Developer Tools -> open Network tab -> run a new prompt throwing an error -> in the search bar in dev tools search for the word "Stream" -> open the file that has the Eval ID (random set of numbers/letters you see in the URL) -> and then look for a Status Code.

Does it say 🟢 Status Code 200, or do you see 🔴 Status Code 427, 400, etc?

[sour spear]

Works after a handful of retries. 🙂

[obtuse smelt]

is status red

[tribal kernel]

Hi, I checked DevTools in detail.

Network → Fetch/XHR is working and requests return 200, but no stream request is ever created.
Searching for stream or eval shows nothing.

This means the stream never starts at all, so there is no Status Code to inspect.
The error “something went wrong with this response” happens before the streaming endpoint is created.

Looks like a backend / model issue on LMArena, not browser or client-side.

[echo aurora]

Okay good to know, thank you for sharing. And you're sure you have the network tab open when running another prompt to through the error message agian, yeah?

[tribal kernel]

I had DevTools open on the Network tab the whole time, with Fetch/XHR enabled, before and during sending the prompt.

I retried multiple times and the error appears, but no stream (or eval) request is ever created.
Only regular fetch requests return 200, then the UI shows “something went wrong with this response”.

So the request seems to fail before the streaming endpoint is initialized.

[echo aurora]

Understood, thank you for the details. Yeah this sounds like it's on our end then for sure. Like I mentioned this model has been erroring out at a higher rate than usual. Our team has been looking to lower this as much as posssible, but I'll be sure to bump this again. Thanks for the info.

[tribal kernel]

Thank you, if I can help I am at your disposal

[viscid cloak]

No models 😢

[echo aurora]

Hmm

What browser are you on? Seeing the same for Side by Side?

Can't say I'm seeing the same on my end.

[viscid cloak]

i tried google ios and firefox ios, same results for Direct. lemme try sidebyside

Sidebyside worked well, and Direct turned normal. may just a minor error

[echo aurora]

Okay glad to hear it's working again. Keep me updated if things change.

[viscid cloak]

hazel revealed?

[torn mantle]

Last time i checked lmarena are adding 3 new things

Video models
Auto modality
New model selector

[atomic lagoon]

Lmaoooo

The image is silly

[molten cipher]

just go on gemini's web itself..

it free

[steel steeple]

Why pro banana doesn't work anymore

[echo aurora]

@quaint raptor you'll want to review the information in #1397655624103493813 for a better understanding on how to use the bot.

[echo aurora]

Sorry to say this model has been pretty high error rate. Our team is looking into a fix asap.

[steel steeple]

Hope this will work soon🙏🙏🙏

[echo aurora]

Same. Fingers crossed.

[steel steeple]

Love you service, thanks

[zealous sparrow]

i wonder if ghostfalcon is flash too, Maybe gem 3 pro?

[echo aurora]

Thank you! Very glad to hear it!

[torn mantle]

nah its flash

[grand flame]

gemini died

[zealous sparrow]

guess the leak only covers fiercefalcon for rn

[fleet lintel]

It's flash.

[winter rain]

Is lmarena a permanent video generator

[zealous sparrow]

yeah prob so

[fleet lintel]

Only difference between ghost and fierce is search on or off

My guess is ghost is with search

[echo aurora]

Hmm what do you mean by this?

[fleet lintel]

What is the link to the original source?

[winter rain]

@echo aurora free lifetime?? Or any chance of premium

[echo aurora]

chance of premium
Can't say I'm aware of these plans.

free lifetime
This is our intention.

[zealous sparrow]

twitter post

[grand flame]

i think gemini 3.0 pro died

since it gives me something went wrong everytime

[zealous sparrow]

works in AI studio

[winter rain]

@echo aurora we respect your hard work bro??

[muted timber]

hello guys

i have a problem

[echo aurora]

hello

[muted timber]

on the AI

[grand flame]

at lmarea

[muted timber]

not working anymore

[zealous sparrow]

um

[muted timber]

and i dont want to create a new chat

[zealous sparrow]

pineapple, is your endpoint down?

[echo aurora]

It seems to be working for me.

Can you: open Developer Tools -> open Network tab -> run a new prompt throwing an error -> in the search bar in dev tools search for the word "Stream" -> open the file that has the Eval ID (random set of numbers/letters you see in the URL) -> and then look for a Status Code.

[zealous sparrow]

nevermind huh

wait ill tes

[winter rain]

But how can u access those tool bro lmarena

[grand flame]

nvm fixed rn

[echo aurora]

It's most common that you're hitting a rate limit, can you try these steps @muted timber and let me know what that status code is?

[muted timber]

What do you mean?

[echo aurora]

Can you: open Developer Tools -> open Network tab -> run a new prompt throwing an error -> in the search bar in dev tools search for the word "Stream" -> open the file that has the Eval ID (random set of numbers/letters you see in the URL) -> and then look for a Status Code.

[muted timber]

open Developer Tools

where is it??

[grand flame]

F12

[echo aurora]

It's a browser setting/option

[muted timber]

i know now

but

" run a new prompt throwing an error"

[echo aurora]

Yeah so with it open, run a new prompt (in LMArena) that'll then result in an error

[echo aurora]

This is how we get better inforamtion to understand what's going wrong.

[muted timber]

so i need to chat another time ?

[cosmic salmon]

Oh yeah, same thing, but also with Nano Banana Pro sometimes. It seems like the fetch request returns nothing after a while, but still goes through with 200 code

[muted timber]

i need for scripting

[echo aurora]

Follow the rest of the steps: in the search bar in dev tools search for the word "Stream" -> open the file that has the Eval ID (random set of numbers/letters you see in the URL) -> and then look for a Status Code.

[zealous sparrow]

theres some 429 status showing in his network tab

[echo aurora]

Okay yeah that's rate limit.

[muted timber]

English, please

[echo aurora]

Need to wait to use again

[muted timber]

i waited one day

i text for fisrt time this day

15 minutes ago

[echo aurora]

I was directing that at pro pro, because they were able to provide a Status Code

I'm not sure whatyour Status Code is

[cosmic salmon]

In my case i get this as the response:

b3:"Error during image generation with google-genai for model endpoint gemini-3-pro-image-preview: Failed to fetch image: Too Many Requests"

for https://lmarena.ai/nextjs-api/stream/retry-evaluation-session-message/ PUT request

[zealous sparrow]

Well, It was actually at him. I read it off the image he scribbled on.

[fossil fable]

3pig broken smh

[echo aurora]

Ah, thank you.

[echo aurora]

Okay yeah it's rate limit for why you're hitting this error

[muted timber]

Yes, but when I chat with him, he lets me use it, and I don't want to because I have all my information on that chat, and he already has all the information from all the files, because I have over 20+.

[muted timber]

so i need only to wait? another one day?

[fossil fable]

can ppl stop calling chatbots he smh

[cosmic salmon]

Same. I've noticed that, when you come back to your chat with Opus after a while, it just stops working for some reason

[echo aurora]

There could be something else that's causing an error @cosmic salmon . However, if you're seeing a 429 Status Code that means it's rate limit. And yes @muted timber will need to wait.

[fossil fable]

where can i smooch off 3pig access whilst it's not working on lmarena

[muted timber]

That's why yesterday he told me to wait 49 minutes (last night), and now I have to work continuously, it will be much harder for me this way... but what can you do...

[cosmic salmon]

Nope, that's a 200 Status Code, so i assume that this is either related to the Gemini API or LMarena is losing my image somehow on the way to me

I'll try to log out and change my browser to see if this issue still persists or nah

[zealous sparrow]

The gemini API has to work, I just used on AIStudio

[echo aurora]

You're getting a 🟢 Status Code 200 when you get an error?

[fossil fable]

im getting a generic error

[cosmic salmon]

Yeah, for the PUT fetch request to https://lmarena.ai/nextjs-api/stream/retry-evaluation-session-message/ when retrying my Something went wrong response

[cosmic salmon]

It's always generic, dev tools shows this as the response to the request

[long jackal]

Is there any way to upload files other than images for the AIs in LMaren?

[zealous sparrow]

no

[half mist]

ChatGPT’s new image model just released!

[zealous sparrow]

images v2?

or well

gpt image 2

[half mist]

It allows for copyrighted characters, and political figures which were originally rejected in previous versions.

[cloud zinc]

nano banana pro is better

[half mist]

Also is way faster

than gpt image 1

[fossil fable]

right

[fossil fable]

What

wait did it

[half mist]

It did, just not with a huge announcement for some reason just a post on the OpenAI Discord

[zealous sparrow]

what degree of copyright

[cloud zinc]

where to use it

[half mist]

On the ChatGPT website or app

It’s out right now

[cosmic salmon]

Yeah, that was a preview, a sneak-peek for the announcement that's going to be soon.

[fossil fable]

holy sh t it's releasing

but it's not out right now is it

openai discord just says this though

[half mist]

Well, way less strict with copyrighted characters than Sora 2 for sure

[half mist]

It is

It’s out

I tested it myself

[fossil fable]

there's nothing on openai.com

[half mist]

It just released, so it’s probably not there yet

[grand flame]

bruh broke again

[zealous sparrow]

sure the whole disney uh scale
but if you can generate goku and stuff yeah OpenAI just wants a lawsuit

[half mist]

Anyways, here is an image made with gpt image 2

[zealous sparrow]

im pretty sure they dont have the rights to this

[fossil fable]

so openai launches it on the one day of the year i need image edit

...

[half mist]

Would say the quality degraded a bit in my opinion

[compact flame]

I still think nano banana looks better though

[zealous sparrow]

aren't they afraid to be sued by uh

who has spongebob rights

[fossil fable]

it's one image you need to check out more

[fossil fable]

paramount

[half mist]

They will probably patch this soon and not allow copyrighted characters in the future

[fossil fable]

at least now we know to rush before it's locked up

we didn't get to rush sora 2 because we thought it'd be normal

[cloud zinc]

just use nano banana pro

[zealous sparrow]

paramount is deada livid on copyright, they are cooked if they dont act

[magic ravine]

Can't. Keeps giving me the error

[half mist]

Here is an image of Donald Trump with Sam Altman using GPT Image 2

[zealous sparrow]

defo

paramount is already planning to sue if they see this

[compact flame]

Okay this one looks kinda good

[grand flame]

now try gemini 3.0 pro image nanobanana

[fossil fable]

Holy fuuuuuuuuuuuuuuuuuuuuuuuuuuuuck we got a gold rush inbound

[cloud zinc]

u can already do that in nano banana pro

[zealous sparrow]

this will be blocked quickly

trust

[half mist]

Alright

[fossil fable]

let me try and get it to expand an image like nbp can do

[grand flame]

looks nice

[fossil fable]

nbp lets me seamlessly expand images to 17:6 for use as a profile banner

i think i'm finally gonna update my banner after like 2-3 years

[magic ravine]

Image 2 isn't on LMArena

[fossil fable]

ik but they say that it's on chatgpt

[magic ravine]

Ah, makes sense

[half mist]

It is. Go test it out yourself

[fossil fable]

i dont use twitter so i have no reason to generate 500 political images but enjoy the gold rush people

[fossil fable]

have you tested it enough to tell whether it's better than 2.5fig

[half mist]

I would say GPT Image 2 better than Nano Banana, but not better than Nano Banana Pro

[fossil fable]

that's why i said 2.5fig

[zealous sparrow]

you cant judge yet

you know why

google is dropping nano banana 3 flash today

lol

[fossil fable]

ik

that's why i asked if he's tested it enough

[fossil fable]

are you sure

[void elm]

too late

[zealous sparrow]

it is a tuesday

[half mist]

Wait, Gemini 3 Flash? I thought it was still 2.5 Flash

[zealous sparrow]

tuesday is new day

[zealous sparrow]

not yet

it hasnt dropped yet

google always drops stuff on a tuesday

[torn mantle]

https://x.com/testingcatalog/status/2000965497110573307

TestingCatalog News 🗞 (@testingcatalog)

BREAKING 🚨: It looks like the Image 2 model rollout has started already for ChatGPT.

It would also come along with a new UI for the Image tab with h style selector and a prompt bar.

Did you get it too? 👀

X · Dec 16, 2025, 4:25 PM

image 2 model is already out for some people

also gemini 3 flash prob tomorrow?

[zealous sparrow]

google releaes stuff on a tuesday usually

and there is already an entry on vertex for 3 flash/fiercefalcon

[half mist]

Will Gemini 3 Flash be better at coding than 2.5 Flash?

[void elm]

???

how is that a question

[zealous sparrow]

2.5 pro even

[neat apex]

the Flash 2.5 09 have near 2.5 Pro level (since always when they release a new flash model)

and looks nothing changed, 3.0 Flash must have a very near level to 3.0 Pro, but without overreasoning issue now

[magic ravine]

Oh, image 2 has just released?

[zealous sparrow]

yea

[magic ravine]

How is it? Still nerfed with fictional characters?

[neat apex]

nope, they fixed that issue

[zealous sparrow]

barely any 3rd party guardrails rn

[magic ravine]

Nice, and it's available on the free tier?

[zealous sparrow]

for a few gens i guess

this is neat too

[torn mantle]

flash wen 😖

[zealous sparrow]

this could be images v1 im not sure

i dont know if i have v2 yet

[grand flame]

hell nah

status 200 btw

[zealous sparrow]

also flowith tweeted this

so flash might be today

[grand flame]

@echo aurora

[cloud zinc]

they dont have insider info

[torn mantle]

do u?

[echo aurora]

Thank you for sharing. I'm not surprising the infinite generation bug wouldn't have an error status code associated with it

[zealous sparrow]

sure not but

they are already prepping

its this week sure

but when

[neat apex]

i am hoping they launch Gemma 4 (what is very likely soon) and someone finetune it to be responsive like the new generation at all

it would have same level of Opus 4.5 lmao

[zealous sparrow]

them asking us to stalk their huggingface page most likely says its this week

[half mist]

Gemini 3 flash is what I am more excited for since I don’t really care about the flash image generation

[zealous sparrow]

I give em 4h to release it

if it aint released

oh well another day

[autumn snow]

hwo to use promt htere

[empty stump]

8% chance it releases today

[zealous sparrow]

nah strong 50

[neat apex]

i would say 50%

[zealous sparrow]

logan tweeted 3 thunders today

[neat apex]

they only delayed Gemini 3 pro a lot to make it 100% consistent

[grand flame]

my chat died

sad

[grand flame]

i tried to make new message and it went instantly to something went wrong