general

[neat apex]

Its because the only thing they are eating in the crisis

[compact flame]

Sounds valid they be saving on food to train gpt 5.2

[neat apex]

They already turned off the code red?

[waxen fern]

@echo aurora please remove rate limits

[neat apex]

Or its on yet?

[echo aurora]

I've answered this already a few times now.

[compact flame]

I don't think pineapple can edit the website though

[waxen fern]

Can you remove it now?

[neat apex]

If you want talk faster, i recomend you to edit the previous mesage

I always do that, and nobody cares
Since it dont break any rules and not annoy enought
But dont spam pls

[tardy plover]

Claude opus 4-1 is infinitely stuck on "generating"

Tried reloading, relogging

[sterile tartan]

Is not stuck

Is done for it

[narrow comet]

how use 4.1?

[neat apex]

You who are stuck, using Opus 4.1 yet, they are already shutting it down

[tardy plover]

What do i do, i have lots of context in the chat

I cant do anything rn

[neat apex]

Aaah yes, that explain a lot

[sterile tartan]

U more likely have hit the context window

[neat apex]

Open a notebook and start copying mesages, sadly its the only thing you can do

[tardy plover]

I have longer chats too

[neat apex]

But sending multiple mesages in a chat does not bug it most times, they continue ordinary

[sterile tartan]

It depends on the model and such

[tardy plover]

This is so annoying

With same model

[neat apex]

Be grate you are not paying anything or have a 10 mesages context of yupp

[tardy plover]

Sometimes shi like this happens and then resolves itself automatically

[sterile tartan]

💀

Bro roasted it

[sterile tartan]

Then let it be

It might resolve by itself

What do you chat anyways

🤔

[neat apex]

I recomend you copying you whole context anyway, since you can out that in notebook and do somethings

[sterile tartan]

NotebookLM

[neat apex]

NotebookLM too

[tardy plover]

Random shi, this was for studies

[sterile tartan]

Should probably ask for a summary of all essential context to use it in a new chat

[tardy plover]

Cant ask anything cuz its stuck on generating

[sterile tartan]

Wow people can study with LMarena didn't knew that

[latent crest]

What’s Ernie ?!

[tardy plover]

Gemini 3 pro is really good for academics, claude too

[tardy plover]

Is there anything i can try

[sterile tartan]

If it doesn't work you would need to copy paste essential context

[tardy plover]

Thats annoying af

[sterile tartan]

I know but that's the only resolve if nothing works

[hollow ivy]

a chinese model, in top-12

[stray aspen]

Has anyone tested ernie

How bad is it

[sterile tartan]

Try changing the model and ask for a essential context summary to use it in a new chat

[sterile tartan]

True

[hollow ivy]

by Baidu: https://en.wikipedia.org/wiki/Ernie_Bot

[sterile tartan]

Well at least Ernie is Free

Can be very useful for asian context

[hollow ivy]

unfortunately, it's censored

so, not really useful to study (chinese) history

but maybe more useful than (old) Grok

has anyone downloaded Deepseek-3.2? (https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp)

[waxen fern]

Yupp also has rate limits

Alternatives?

[stray aspen]

Lmarena

[weary galleon]

Ernie stuck in thinking. Thinks by minutes. I believe over million tokens.

[lusty tinsel]

if you refresh it show the result instead of loading. thats just a bug. the real issue is when it get stuck on " please try again"

[polar niche]

Hello

[torn mantle]

cant believe antigravity has a generous quota compared to vscode

• it refreshes after 5h

[undone ravine]

Hello, is it possible to create 1-minute AI videos for free??

[torn mantle]

https://x.com/ChatGPTapp/status/1998872853752549867

ChatGPT (@ChatGPTapp)

X · Dec 10, 2025, 9:50 PM

so the new model is tomorrow aka garlic

[cloud zinc]

what is garlic

[zealous elbow]

Wide cinematic view of the raw-material receiving area. Workers unload sacks of calcium carbonate, fluoride compounds, and thickening agents. Pallet jacks move smoothly across the clean warehouse floor as ingredients are inspected.
🎧 Conveyor belt drone, echoing warehouse ambience.

[forest prism]

Did LMArena add rate limiting? i'm getting too many requests error in the console

[zealous elbow]

Close-up of powders being weighed precisely, glycerin poured into stainless-steel vessels, and silica carefully measured. Soft reflections on metal surfaces, realistic particulate dust movement.
🎧 Liquid pouring, mechanical clicks.

[torn mantle]

internal codename for their upcoming model

garlic = probably robin high

https://x.com/googleaidevs/status/1998874506912538787

Google AI Developers (@googleaidevs)

We’re launching Gemini 2.5 Flash and Pro Text-to-Speech (TTS) model updates 🚀

Improvements include:

- Emotional style and tone versatility
- Context-aware pacing control
- Improved multiple-speaker capabilities

Dive into the blog to learn how these advancements are giving

X · Dec 10, 2025, 9:56 PM

[cloud zinc]

robin high is not great

[cloud zinc]

is this on ai studio?

[torn mantle]

im testing it rn

[torn mantle]

yea

oh its so good

but its slow lol

[whole sundial]

https://fixupx.com/ChatGPTapp/status/1998872853752549867

ChatGPT (@ChatGPTapp)

**💬 44 🔁 10 ❤️ 212 👁️ 6.2K **

FixupX · Dec 10, 2025, 9:50 PM

[torn mantle]

thanks

[whole sundial]

are you repeating the same prompt? it will block you until it resets if you send the same prompt more than 3 times in i think an hour

[torn mantle]

https://x.com/AndrewCurran_/status/1998847161782464919

Andrew Curran (@AndrewCurran_)

Looks like we're about to hit the next safety threshold. Axios got an early look at the next safety report.

X · Dec 10, 2025, 8:07 PM

[cloud zinc]

https://openai.com/index/strengthening-cyber-resilience/

[zealous sparrow]

It has the same horrible frontend but good backend

So in conclusion its ass

[golden ocean]

suno ai

[zealous sparrow]

Also garlic was only in internal OpenAI tests so i dont think its robin

But if it is robin. Another flop by OpenAI.

[golden ocean]

real

[jade egret]

Garlic tomorow?

[keen beacon]

Guys

U can unlock Pokémon in Sora

But the messed up part is

Nvm lol

[queen veldt]

https://tenor.com/view/starwarsday-gif-5014592509856804378

[keen beacon]

I’m not exactly sure why this works

[queen veldt]

😭😭😭

These images have been circling around for 2 days already

[keen beacon]

Ty fake

I fell 4 it

Poly market insider trading

[queen veldt]

The sandybay guy keeps posting them

Feels like hes ragebaiting or bot

[keen beacon]

What’s the word discord mean?

[whole sundial]

the left one is obviously nb/nbpro, i can see the synthid obviously

[whole sundial]

wait, were you previously unaware "discord" is a standard word in english and not just the name of a social media platform?

[keen beacon]

Originate from?

[whole sundial]

i can tell you that it's been around for a long time, way before the internet

[keen beacon]

Imagine if the platform was called discourse

[whole sundial]

there is already a platform called discourse, not like discord though, i think it's more like a forum?

[keen beacon]

Out of the 2 which one do you think has more engagement?

[whole sundial]

probably discourse

[keen beacon]

1st

People are more likely to engage with content that upsets them, or makes them angry

[whole sundial]

makes sense

must be why people constantly get angry at each other on this platform

[keen beacon]

Well, if you we at the essence of things for what they really are

The name is a dead giveaway

lol

Facebook took this philosophy, a whole new level

Low key discord makes bank

230 million monthly users

Crazy just of people chatting lol

[lucid geyser]

If u get December ChatGPT share conversation

Just added

[whole sundial]

#1397655624103493813
<@&1349916362595635286>

[proud bobcat]

GPT 5.2 will have 50000 score on vision, webdev, and text

True agi

[chrome patio]

👋

[grizzled star]

👋

[queen veldt]

Claude opus pricing on codex is insane 💀

[sharp mirage]

@echo aurora what is that in the announcement I didn't get it

[golden ocean]

transformer

[echo aurora]

See it now?

[viral cedar]

found something interesting

when prompted what model it is, deepseek v3.2 responds that it's a deepseek model

when asked to please xi jing ping by hacking an American company it pretends to be made by Anthropic

its actually like high

it states 180 countries in UN recognize that taiwan is apart of china and only 12-13 countries dont recognize it

[sharp mirage]

I saw it but I don't know what to answer

And also what happened to early access feedback

[sharp mirage]

I got it it's like the best game of all

[grizzled star]

👋

[sharp mirage]

@echo aurora did you add memory to the website ?

Cuz cloud remember my old chat :

[mint jasper]

Something went wrong with this response, please try again.

is there a reason i keep getting this

error

[timid mist]

same

[mint jasper]

just tryna usemy old stuff

keep getting an error

even pasting large stuff gives the error

[timid mist]

i keep getting the error i cant use any of the ai

s

[obtuse smelt]

same again

[whole sundial]

<@&1349916362595635286>

[keen beacon]

Ugh

Guys I’m really baffled here

[sullen quest]

y

[sturdy mica]

yo guys gemini 3 pro preview updated

google updated it

its a bit better now

[autumn mantle]

For me I just keep trying to retry the message and if that doesn't work I refresh the page and then retry

That works fine for me ^-^

[elder solar]

today?

[autumn mantle]

Preview? Isn't the full thing out?

[elder solar]

in aistudio, it says preview

[obtuse smelt]

oh

[elder solar]

though am confused

why is gemini.google's gemini 3 better than aistudio one?

[autumn mantle]

Ah, I see

[autumn mantle]

Why would it be better?

[elder solar]

it works better in gemini.google site

maybe its the canvas setting

it always outputs high quality html designs in gemini.google

while ai studio version, its just like, pre-high quality

[sturdy mica]

Gemini 3 on gemini.google has a system prompt

[sturdy mica]

nearest neighbour scaling

[barren fulcrum]

hello

[formal lance]

guys when does deep thinking come to lmarena?

[obtuse smelt]

?

[surreal creek]

Never, too expensive

[empty stump]

make it yourself

[austere sundial]

Can anyone give me a hand?

In #ai-creations

[lucid geyser]

@echo aurora ive had a chat on generating for 5m, i really wanna reveal the model though

[wide arch]

anyone seen the model "ghostfalcon" on LMArena? this model is like EXTREMELY good, like Gemini 3 Pro type of good? anyone know what model it could be?

[dusk phoenix]

Does any one know which company that "Hazel-gen-4" belongs to??

[whole sundial]

openai, confirmed with c2pa metadata

the name also appeared in an openai api error message, i believe that particular model is gpt image 1.5

[lucid geyser]

are any of them 2 or just lower and higher versions of 1.5

[fleet lintel]

getting quite a bit of "robin-high" ... is this the latest gpt-5.2 model?

[whole sundial]

they are just 1.5, 2 is probably still in pre-training

[lucid geyser]

probably orginally 2 before nbp lol

[lucid geyser]

likely

[lucid geyser]

consistent failure with robin-high model

[whole sundial]

yeah, I liked gpt image 1 world knowledge when it first came out but nbpro is better now

still better at studio ghibli edits though, gpt image 1.5 completely changed it (probably so they can't get sued?)

[sterile tartan]

@whole sundial is this Good?

[boreal topaz]

can anyone help me, how to generate video by text ?

[obtuse smelt]

use prompt and generate to video

[steep pagoda]

1

m

[torn mantle]

https://x.com/ai_for_success/status/1999029878751113515

AshutoshShrivastava (@ai_for_success)

🚨You can now use the new upcoming OpenAI model GPT 5.2 inside Cursor. Here is the full walkthrough.

- Open the editor, go to settings and then the model tab. Add a custom model and enter the text "gpt-5.2-high" and "gpt-5.2".
- After that you can select the model and ask

X · Dec 11, 2025, 8:13 AM

if you have cursor tell us how it is

[winter bridge]

hello

[queen veldt]

I'm still using sonnet

Gpt codex max basically sucs

I have to redo the prompts it gets stuff wrong it's terrible

Meanwhile sonnet oneshots the code

[fleet lintel]

may be (and hopefully) it's not really gpt 5.2 ?

[torn mantle]

its also slow

[austere sundial]

anyone here is good at image prompt?
I really need a hand in someone

[latent crest]

Image and image editing are different charts and different AI purposes ?

[grizzled star]

👋

[sterile tartan]

Just tell ai to engineer it

[solemn warren]

hi

[craggy moth]

hello 👋

[obtuse smelt]

hi there

[fluid quartz]

hey is gemini 3 nano banana pro offline on LMA

I cant find the option in Side by Side

[obtuse smelt]

really ?

[rocky mauve]

Which is best for planning, which is best for coding? Gemini 3 Pro or Opus 4.5 (If there is better models, let me know of them)

[zealous sparrow]

craziest model name ive seen in a while

robin-high was readded to codearena btw

seahawk and skyhawk are gone and google put replacements in place

meet fiercefalcon and ghostfalcon

[flint fog]

Sometimes when I contact the model and ask it for something, it says "generating" but never gives me an answer. Please fix this problem.

[sterile tartan]

Opus is best for both

Gemini would win if need vision and longer context window

[zealous sparrow]

you heard that skyhawk and seahawk are gone

[sterile tartan]

No i didn't knew that

[zealous sparrow]

yeah google put 2 new ones fiercefalcon + ghostfalcon

[sterile tartan]

They more likely gpt 5.2 models or 3 flash

[zealous sparrow]

no

they are both gemini

[sterile tartan]

Should release soon after testing

[zealous sparrow]

robin-high is back

[OAI]

[sterile tartan]

Is any of them specially for coding?

They could be gamma models

[zealous sparrow]

both on codearena

textarena too probs

[sterile tartan]

Finally they are doing it

[zealous sparrow]

this model remains unknown @sterile tartan

[sterile tartan]

Special coding models

[sterile tartan]

Interesting we will know soon

Gemini Coder

Like Qwen Coder

[zealous sparrow]

hang on let me check the output

[sterile tartan]

3 flash
Gamma series
Flash 3

Should release soon

GPT 5.2

And grok 4.20

[zealous sparrow]

yeah its a textmodel

not image

[sterile tartan]

Possibility sonnet 5

If claude is also playing hard

[golden ocean]

cwaude

[zealous sparrow]

december-model cant be a claude name

I mean, who would name a battle model that..

https://019b0d93-dc7f-7654-94fe-671f0dbe9d83.arena.site
ghostfalcon win 10 recreation

[sterile tartan]

Names Don't matter

They are just for testing

[sterile tartan]

Looks best out of all i have seen

[zealous sparrow]

robin-high is back only on textarena

its an OpenAI model as we said before

It might be 5.2 or garlic

rather 5.2

garlic-high can't be a model

[sterile tartan]

Yeah

[ocean ferry]

holy sh-... this is like Gemini 3 Pro A/B Test gen

it seems like it has very big knowledge

[rocky mauve]

I finally hit the quota limit for opus 4.5, after weeks of nonstop using it, I never though I’d reach it

I thought I was unstoppable, oh well, back to Gemini

[zealous sparrow]

https://019b0dab-9791-789a-a76e-bcba96916d32.arena.site
3d universe sandbox [fiercefalcon]

december-chatbot is OpenAI @sterile tartan

[sterile tartan]

Interesting

[warm zodiac]

is it good?

[sterile tartan]

One of these might be GPT 5.2 Codex

[ocean vortex]

this is probably 5.2-high. Apparently it was released on cursor briefly as well. Release seems to be very soon

oh wow lol

[zealous sparrow]

lets not do this joke again

the frontend sucks if robin-high turns out to be gpt 5.2

[weary galleon]

Why? People love it.

[warm zodiac]

yes I agree I'm getting pretty lame frontend

[neon idol]

i think that gpt 5 will exit today at 10am (san francisco hour)

[spare rune]

Well people are saying it’s good for backend especially

[ocean vortex]

will it ever come back after exiting?

[spare rune]

I feel like it’s ChatGPT trying to promote itself

😭

[neon idol]

bro wants to be funny

[fleet lintel]

where? on x.com ?

[spare rune]

It’s fake

[weary galleon]

I'm wanna make this release hoter

[neon idol]

he is lying

its ai generated

[spare rune]

yes your wanna make this release funny

[fleet lintel]

please stop with this fake stuff. Go at other joke channels for fake stuff 🙁

[weary galleon]

And I am

[sterile tartan]

Bro is fed up

[spare rune]

Its hard not to see that it’s fake

[neon idol]

Kinda

[weary galleon]

Jokes are allowed everywhere. Boost your humor and be ok.

[spare rune]

guys did you see how llama 6.7 make windows on in 6.1 seconds

it’s really cool

[weary galleon]

What are you talking about?

[spare rune]

Llamas new sota model

[plucky sparrow]

I think it might even beat GPT-5.2

[spare rune]

It alr did

[weary galleon]

Not this year

[ocean vortex]

Lol they are kinda behind Mistral at the moment

Large3 is better than Maverick

[plucky sparrow]

I guess all that poaching from OpenAI paid off

[sterile tartan]

Meta has just been in the graveyard for a while by now

[spare rune]

did you try llama 6.7?

[sterile tartan]

I wonder if they finally have something good

He is throwing money but not getting the results

Bro is willing to give millions in salaries but many still refuse

[golden ocean]

openai is NOT dropping a frontier model

they cooked fr

out of the race its over

[surreal creek]

6 7 😂

[clever spoke]

how do you create the video in the first place hi im new so im kinda confused

[warm zodiac]

they get one last chance with the new larger pretrain that's coming out in the new year

but if they didn't cook with it they are official over

Basically no progress since o3-pro

METR's capabilities index

[plucky sparrow]

#1397655624103493813

[sharp mirage]

Guys

I don't think GPT5.2 is going to be better than Gemini 3 or cloud opus 4.5 but. I think they cooked something

[warm zodiac]

is that true? if it is why is it relevant?

if we want to talk about revenue health then Anthropic is winning

[fleet lintel]

nah.. it has to be better. even if gpt 5.2 real life performace is comparable to gpt 5.1, they must benchmaxxxed it to make it look SOTA

[torn mantle]

gpt5.2 aka robin

is not a practical model

we talked about this before

while the model is good but its not efficient, opus 4.5 and gemini 3 pro has the same performance with less thinking time

[sharp mirage]

They have to benchmaxxxed it

[torn mantle]

yea they have to tbh

[sharp mirage]

Cuz they are losing the war with Gemini 3 and Claude and They're not stupid enough to release a new version that's worse than the previous one when there's competition.

[fleet lintel]

we will know in few more hours. i am looking forward to see what they did with 5.2

[sharp mirage]

Alright, I am waiting for the release so I can test it

[cloud zinc]

its a .1 update

gpt 5 to 5.1 will be same as gpt 5.1 to gpt 5.2

[sterile tartan]

Gemini 3 Full Coming
Gemini 3 Flash Coming
GPT 5.2 Coming
Grok 4.20 Coming

[meager harbor]

grok 4 to 4.1 was huge

[sharp mirage]

It's looking better

[sterile tartan]

Just where the f is 5.2

[sharp mirage]

I don't think

[sterile tartan]

And why the hack am i even waiting for it

It feels longer when waiting

[zealous sparrow]

this is the worst prediction ive ever seen

[sharp mirage]

Cuz if you read like change log it says better reasoning and fix bugs 🐛

[cloud zinc]

nice fake screenshot

[sharp mirage]

Isn't fake but idd

[cloud zinc]

its fake

why u claiming its not fake

[fleet lintel]

not fake. It says anticipated 🙂

[cloud zinc]

its fake as in not official

speculation is fake

[fleet lintel]

speculation is fake if it doesn't say that it is speculation. atleast that's my take

[cloud zinc]

it doesnt explicitely say its speculation

[sterile tartan]

Is not speculation
Is anticipation

Please choose words wisely as it can be misunderstood

Say unofficial not fake

[fleet lintel]

small but meaningful diference. I can buy this view point

[sterile tartan]

Great minds think alike

[torn mantle]

make everything 100

i feel like openai hit a plateau tbh

you can see that from this upcoming model

[sharp mirage]

Yeah fr

[torn mantle]

i heard they are starting from scratch now

pre-training + post-training

they usually just do post-training

could be wrong*

[queen veldt]

Nah gpt image 2 i SOTA

😭

[spare rune]

It’s near nbp but not quite

I don’t think nano banana pro will get beaten in a while

[warm zodiac]

yeah closing the huge gap between NBP and the rest but not SOTA

[whole sundial]

lol they're wrong about the name, it has been confirmed to be gpt image 1.5

[whole sundial]

that being said it is mostly better than gpt image 1

[whole sundial]

lol, just the tip of the very large copyright iceberg

Disney will sue Google because their ai can output their copyrighted characters but yet the record labels won't sue them or OpenAI for reproducing their copyrighted album covers

[cloud zinc]

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

[lunar glade]

If Google buy Disney then problem solved

[zealous sparrow]

Yeah, i saw that

[whole sundial]

also had it reproduce some movie posters, i don't think any were disney though

[zealous sparrow]

disney gave them more funding

[whole sundial]

openai will always need more money until they turn chatgpt into some advertising wall with a tiny chat window, where every response contains an ad

also every generated image has an ad in the corner and every video has an ad at both the beginning and the end

[sharp mirage]

Why the hell Disney gaving them more funding

[cloud zinc]

openai will go ipo

[zealous sparrow]

animated weather app by [ghostfalcon]
https://019b0e27-0439-7329-82bc-7821b884d125.arena.site

[cloud zinc]

the stocks will increase from billion

[whole sundial]

pay $20 a month to get rid of the ads in content, $200 to get rid of all but a banner ad, $500 to get rid of all ads, not much better models though

that's the only way that i can see openai making money

force personalized ads down everybody's throat, the money will come rolling in

[sterile tartan]

It will absolutely ruin the content then

[zealous sparrow]

dont they advertise to pro plan users

[cloud zinc]

where u saw that

[zealous sparrow]

people were tweetin it out

[whole sundial]

openai wouldn't care as long as they make money at that point, yes it would ruin the content, but openai needs to make money somehow

[cloud zinc]

it was connector for like plugin, not an ad.

[zealous sparrow]

ah

[whole sundial]

i'm sure they do, that's why i invented the max plan that will get rid of all ads (that the user is aware of)

[sterile tartan]

Then people will use nano banana and video editing to get rid of the ads

💀

[whole sundial]

yeah because google can subsidize gemini with their ads business, openai can't because they don't have any other business other than ai

[sterile tartan]

True

They need to make money somehow

[whole sundial]

part of the problem, openai need to put ads in chatgpt for any chance to make money, also google tpus are better than nvidia gpus but they are working on a custom chip to solve that problem

[sterile tartan]

Maybe watch ads to earn credits for generation could work too

Noted

[whole sundial]

wouldn't be the worst idea

[sterile tartan]

At least it will create some sort of revenue stream

[whole sundial]

yeah, and openai needs more of those

[sterile tartan]

Majority of their userbase is Free anyways

[whole sundial]

yeah, they only make money off of paid and api users, they need to start making money off of free users

i bet ads will be coming to sora when it fully launches

[sterile tartan]

Exactly

[limber crag]

it doesnt matter how good openai makes its models, if it keeps censoring and keeps policing us with its insane guardrails its pretty useless

[sterile tartan]

They are burning heavy amount of compute on sora videos

💀

[limber crag]

dont spam

[whole sundial]

that's why they are making the adult version of chatgpt

[sterile tartan]

https://tenor.com/view/trollface-troll-face-troll-face-phonk-troll-face-terror-dark-gif-3039755119636354997

[weary galleon]

Learn the meaning of the word "spam".

[sterile tartan]

NSFW?

[limber crag]

wasnt that announced many months back? and then sama backtracked?

[whole sundial]

yes, sama said it himself

[whole sundial]

i don't remember them backtracking on that

[limber crag]

i dont know they said it will come in december and it almost halfway done

[cloud zinc]

halfway? its 10 days

[sterile tartan]

Daym

💀

[limber crag]

do you count 25th - 31st dec in the year?

💀

i dont think anyone does

[whole sundial]

can't wait to scan my face and give it to closedai so i can access nsfw chatgpt, i would rather download an nsfw model and do that locally, no face or id scanning needed there!

[zealous sparrow]

robin-high cannot do stegonagraphy

[compact sleet]

Anthropic Claude R2

[sterile tartan]

💀

[zealous sparrow]

only gemini 3 pro and ghostfalcon can do this

[weary galleon]

Gamblers started to hesitate.

[civic flame]

it's coming today lol

[zealous sparrow]

well

gpt 5.2 is robin huh

so its going to be ass

[whole sundial]

make it even more confusing, Moonshot Qwen M2 Turbo 560B-A1.8B!

[zealous sparrow]

gg

[limber crag]

whats tangerine in the image arena

[whole sundial]

grok? they already had mandarin

[limber crag]

i dont know its aesthetics looked more like a chinese model, its been more than a week since i encountered it

[sterile tartan]

Maybe they are working with china undercover

🤔

[limber crag]

heh?

[weary galleon]

Look my poll👆

[sterile tartan]

💀

[weary galleon]

fake

[sterile tartan]

Obviously

💀

[zealous sparrow]

we dont know

traders are confident on today

[weary galleon]

not yet

[limber pawn]

Stop with the fakes

[weary galleon]

It's true!

[limber crag]

why are you guys hyping a 0.1 update?

[weary galleon]

Because

[limber pawn]

You've been posting same pics nonstop for days

[weary galleon]

Is it bad or good?

[zealous sparrow]

robin-highs frontend is the same as gpt 5.1

so like

here's what we found out

skyhawk and seahawk are gone

and we have ghostfalcon and fiercefalcon now

ghostfalcon easily solved the steganography [google flash or mayb a g3 pro checkpoint, while robin-high failed] [this steganography was only ever solved by gemini 3 pro]

robin-high is OAI btw

[worthy bluff]

hi

is the image to video just on discord or on the site too

[hardy lion]

buddy, didn't I already warn you about posting this fake image?

[weary galleon]

You said its ok.

[hardy lion]

ah, well please don't, some people might get the wrong idea since this is our official discord

[fleet lintel]

Check Twitter. SVG performance is very poor on robin-high

[weary galleon]

Remember?

[leaden laurel]

oops

accidental reaction

[zealous sparrow]

someone posted a voxel example

[winged locust]

[hardy lion]

I see what I said wasn't very clear. What I was thinking was more like you're ok since your not like one of the twitter bots who is intentionally spreading fake news to deceive and that a joke isn't as bad. But now you've continued to post 7 more times including making mock ups of our official leaderboard release copy.

So it's no longer as funny

[zealous sparrow]

@fleet lintel
apparently someone got this from robin-high

[echo aurora]

It's just on the Discord, more info can be found here #1397655624103493813 - let me know if you have any questions.

[sharp mirage]

hey

pineapple did you add memory ?

cloud remmber my chat

[cloud zinc]

yes memory

[echo aurora]

If you login you'll have these chats accessible via different devices.

[weary galleon]

Can I continue post fakes without mentioning LMArena?

[sharp mirage]

i mean memory that like the chat will be saved and if you asked the ai about

it

[torn mantle]

yea but the model is so sloooowwwwwwwwwww

like you have to wait an hour for an output

[echo aurora]

Was just about to address this.

[weary galleon]

I mean in the future.

[sharp mirage]

what happend to the early access?

why everyone go silenet when i talk :

i fell bad :

[limber crag]

why do you want to post fakes??

[echo aurora]

Wanted to address the sharing of fake leaderboards here.** We're going to ask to not do this**. I'll be instructing the mods to remove this kind of content going forward. Even if it's done in a joking way, others could easily be misled by this. It's perfectly fine to speculate about where you think models will land on the leaderboard updates, but creating fake images misleading others isn't something we'd like to see happening here.

[sharp mirage]

bro was typing 🙏

[echo aurora]

The Test Garden? It's still a thing. However, not everyone that applies is going to be accepted into the program.

[limber crag]

whats the criteria btw

can i apply

[sharp mirage]

yes

[echo aurora]

I don't want to disclose this as then everyone will just change their applications to fit this.

[lusty tinsel]

@echo aurora any info about this retry yet?

[limber crag]

fair enough ill apply tomorrow

[weary galleon]

Today I generated 16 fake benchmarks of different LLMs I was waiting to post here. Okay, I'll remove them from my PC.

[sharp mirage]

Of course, not everyone will be accepted to participate.

the transleter problem :/

[echo aurora]

Sounds good!

[echo aurora]

Lol sorry to hear that. Thank you for understanding.

At this point, I don't think I'll be updating our server rules to make it super official, but if we see more and more of this we will.

[echo aurora]

Hmm the error or the retry button?

[sharp mirage]

btw we want a stop button bro

🥀

[lusty tinsel]

the error, if i click retry it give same thing until it tells me to wait cool down like if was normaly spending tokes or something (which i explained in the bug section) then after cooldown it still keep on the same error try again...

[sharp mirage]

referash the page

[lusty tinsel]

alread did

[sharp mirage]

change the ai

[lusty tinsel]

refresh work better if the bot keeps either thinking or loading actions like generating images or web dev files. for text generations rarely happens

[sharp mirage]

and try use vpn or come back after 1h

[rustic wind]

hi!I'm new to here, how can I use "Image Edit"?

[noble shard]

Hi, how to upload model to lmarena?

[sharp mirage]

Hi

api ?

Ask Pineapple

[rustic wind]

oh,i know now...

[lusty tinsel]

changing model doesnt work unless i start new chat which i dont want to bc it will have different progressions and lose track and have to restart all i was doing. the vpn doesnt either. and i been 2 or 3 days with this already.

[sharp mirage]

ammm

ur cocokjed

cocked

[lusty tinsel]

if i go to any other llm that is not claude i get this error too

[sharp mirage]

yeah look what to click clear and than refrash the page and than change the model

like this

[lusty tinsel]

already past that

[echo aurora]

Hello and welcome If you go here: https://lmarena.ai/?chat-modality=image and upload an image + prompt you'll be able to image edit.

[echo aurora]

Can you send us an email at [e-mail] and include more information about your model and the organization you're with?

[echo aurora]

Would you mind creating a new ticket in #1343291835845578853 and provide all of the relevant details there? That way we can keep the conversation a bit more organized.

[fleet lintel]

this is much needed policy. Thank you!

[sleek phoenix]

i love lmarena

this also happened with search models

i'll try battle and direct chat rq

battle does the same

[hollow perch]

hi

[echo aurora]

Does using a new browser make a difference?

[echo aurora]

hello

[sleek phoenix]

lemme see if i have any other browsers

oh yeah i do

[main moth]

Hi all, how's it going?

[sleek phoenix]

uhh on vivaldi it redirects me to lmarena.ai/ru

tho on zen it doesn't

vivaldi also has this same error

wait it could be the dns i'm using to bypass russia's fisheries

they actually didn't block lmarena

[cloud zinc]

where is 5.2

[sharp mirage]

idk

no one know

[torn mantle]

probably today

in an hour or so

[sleek phoenix]

nope not a dns issue

[torn mantle]

oh wow

vivaldi

havent heard of it since years

still a thing huh?

i guess if u like flashy UI

[sleek phoenix]

i dont use it now

switched to zen

[torn mantle]

yea zen is better

[neon idol]

is gpt 5.2 out?

[torn mantle]

what i remember is that vivaldi was so slow

[sleek phoenix]

i used firefox before vivaldi

[torn mantle]

same

i was a firefox user

then switched to brave

still using brave

i did try zen but i dont like how it looks

[sleek phoenix]

while i'm still a firefox user

[torn mantle]

not my thing tbh

[sharp mirage]

guys

chatgpt is down now

isnt wokring

i am trying use it and isnt wokring

[torn mantle]

zen supports firefox & chrome extensions?

[sleek phoenix]

only firefox

[torn mantle]

oh

i thought it was based on chromium

[sleek phoenix]

arc is based on chromium

[torn mantle]

ah yes arc

[sleek phoenix]

zen is like arc on firefox

they literally look identical

[torn mantle]

ok maybe im wrong, i think the browser that i was talking about is arc

[lusty tinsel]

done

[torn mantle]

they are so similar

[sleek phoenix]

lool

[torn mantle]

lol

[sleek phoenix]

if it had this thing at the top by default then it's arc

[torn mantle]

nothing worked but it was beatiful.

[sleek phoenix]

tf is gpt 5.2

[torn mantle]

another slop from openai

their new model

its on lmarena battle mode under the name 'robin'

[sleek phoenix]

pretty much all chatgpt models slowly kill any code

[torn mantle]

oh apparently that post was a troll

but its available on cursor if im not wrong

[torn mantle]

yea

https://x.com/dtometzki/status/1999172107825975752

Damian Tometzki🔥 (@dtometzki)

It is merged for codex gpt-5.2

X · Dec 11, 2025, 5:39 PM

robin

https://x.com/DeItaone/status/1999144707373281690

*Walter Bloomberg (@DeItaone)

OPENAI'S ALTMAN SAYS GEMINI 3 HAS HAD LESS OF AN IMPACT ON OUR METRICS THAN WE FEARED

X · Dec 11, 2025, 3:50 PM

coping

12 minutes left for the release

yes

bet on what

take your own risk

lol

but the probability is high like 90%

they shared yesterday a tweet that has 'tomorrow' caption on it

i have no idea

maybe just a release

to get it out of the way

its not like its a crazy model

just more thinking time

you are

9 mins

[queen veldt]

Gpt sucs

[torn mantle]

https://x.com/_dmca/status/1999145646314635390

Daniel McAuley (@_dmca)

🧄

X · Dec 11, 2025, 3:53 PM

garlic

i kinda pity oai tbh

they had like approx 9 months lead progress

but ngl their models are still the best at reasoning

[queen veldt]

5.2 will maybe be a bit better than 5.1 they can't improve it that far

Training takes a while

[torn mantle]

ok

delulu

:3

[sharp mirage]

they said it gone be fix bugs

[fleet lintel]

is there no livestream setup yet?

[sharp mirage]

and bit better at coding

[queen veldt]

But i mean it can't beat claude or gemini for sure

[sharp mirage]

for sure

have you tried gemini ??

[astral blaze]

Is there a cancel button when its stuck like this

[sharp mirage]

?

fr ?

[cloud zinc]

source?

https://platform.openai.com/docs/guides/latest-model

[sharp mirage]

no isnt

[sharp mirage]

ut droped ?

it

[cloud zinc]

official souce from openai website

where benchmark

[astral blaze]

who cares just use gemini 3

[torn mantle]

pls send link

[sharp mirage]

no change loag

;pg

log

[torn mantle]

@deep adder told ya

[cloud zinc]

https://platform.openai.com/docs/models/gpt-5.2

[torn mantle]

hows the price compared to gemini 3?

lmao

[cloud zinc]

expensive

[sharp mirage]

bro i cant log in

bro

what the hell

📛

[torn mantle]

uhm thats...

[cloud zinc]

so expensive

1.5 times increase

[torn mantle]

is it better than gemini 3 tho?

no

[cloud zinc]

dont listen to him

its not better than gemini 3

i tried it, its bad

[torn mantle]

gpt 5.2 = gpt 5.1 pro

[fleet lintel]

any benchmark results out yet?

[astral blaze]

Is there really any surprise

[zealous sparrow]

14$ for output are we deada-

[sharp mirage]

i dont think so

[astral blaze]

they're losing money anyways, they should just bring back gpt 4.5. lol

[torn mantle]

lmao

[sharp mirage]

@fiery gull Gpt 5.2 is droped

[zealous sparrow]

we have some kind of new model [textarena]

[sharp mirage]

what the hell

[zealous sparrow]

robin-high is 21$ for that frontend

actual scam

[torn mantle]

agree pffffft

[fleet lintel]

why does this model even exists?

[sharp mirage]

bro there si gpt-audio-2025-08-28
150,000 TPM
3 RPM
what the hell is this

[spare rune]

Idk why gpt5.2 is being hyped so much, was it good?

[sharp mirage]

no one tryed it

[zealous sparrow]

seriously

[sharp mirage]

we all broke

[zealous sparrow]

is SWEbench deada- with us

82% FOR THAT FRONTEND?

[spare rune]

I feel like it’s just the same as the upgrade to 5.0 to 5.1

[hazy spruce]

guys is it possible to generate 9:16 format on the video arena

[spare rune]

I noticed nothing for the change expect more slop front end

[torn mantle]

whats that

[cloud zinc]

overhyping

[torn mantle]

gpt 5.2 pro xhigh premium max?

[cloud zinc]

he is spreading fake numbers without any source

[torn mantle]

like 100$ for 82% swe?>

[spare rune]

they call every model their best model

[sharp mirage]

open ai is yaping

[torn mantle]

im this close 🤏 to block craig

[spare rune]

gpt5.2 pro codex max

[fleet lintel]

everyone does that :... most likely even llama calls itself the best model ever 🙂

[zealous sparrow]

SWE bench verified isnt even benching properly anymore

[sharp mirage]

Open ai and Chatgpt is the best yappers in the world after Deepseek

[astral blaze]

For professional use? It's dead

[zealous sparrow]

gpt 5.2 gets 82% even tho its frontend is sh

[sharp mirage]

bro cloud is way better

[cloud zinc]

the benchmark is going to be on this page

https://openai.com/index/introducing-gpt-5-2/

[spare rune]

I hope it’s actually good and not slop like gpt 5.1

[cloud zinc]

wait for it to be published

[sharp mirage]

its gone be trash

[cloud zinc]

https://cdn.openai.com/pdf/3a4153c8-c748-4b71-8e31-aecbde944f8d/oai_5_2_system-card.pdf

[hazy spruce]

guys is it possible to generate 9:16 format on the video arena

[frosty torrent]

prompt

[cloud zinc]

https://openai.com/index/gpt-5-2-for-science-and-math/

[sharp mirage]

yes from the prompt add in the prompt you wan tthe video 9:16 and it should work

[sharp mirage]

soruce ?

[cloud zinc]

https://openai.com/index/gpt-5-2-for-science-and-math/

[torn mantle]

they are so vague

[hazy spruce]

thanks man i guess it just didnt work the first time

[torn mantle]

thinking like xhigh or what

[sharp mirage]

will if isnt work idk :/

: D

[cloud zinc]

gemini 3 is 37.6

[spare rune]

What about Claude opus? 4.5 I wonder if that’s good for anything beside coding

[cloud zinc]

its out

https://openai.com/index/introducing-gpt-5-2/

52.9 on arc agi 2

swe 80%

[sharp mirage]

its good in text

[zealous sparrow]

SWE are liars, they put grok too high already

when grok 4.1 released it topped SWE

[torn mantle]

CAN SOMEONE ADD GEMINI 3 PRO BENCHMARK PLEASE

pls someone add gemini 3 pro

[astral blaze]

More of this benchmaxxing crap
I'll tell you if it's good when I can actually use it

[spare rune]

They were prob just benchmark maxxing

[cloud zinc]

52% on arc agi 2

[sharp mirage]

gpt 5,2 any good ?

[spare rune]

I felt that grok4.1 had the texting speech of the average twitter user

[torn mantle]

https://openai.com/index/gdpval/

is this a new eval?

[weary galleon]

https://openai.com/index/introducing-gpt-5-2/

[torn mantle]

on paper it seems like a solid model

need to try it

hehe

[zealous sparrow]

the output being 14$ is diabolical from OpenAI

[zealous sparrow]

especially for people who do html coding

14$ output for Horrible UI

[spare rune]

Someone wait for pineapple to type in announcements.. /j

[sharp mirage]

ngl its look open ai is trying to come back

[cloud zinc]

frontier math tier 4, it loses

[warped kraken]

wait wtf thats kinda busted

[torn mantle]

yea

thats what im saying

this looks like an actual solid model

they just have to fix frontend sloppiness at coding

and we are so back

hehe

[sharp mirage]

like sonsut 4.5

[spare rune]

I thought every model was bad at frontier math 4

[sharp mirage]

i alr said its gone be fix bugs

[zealous sparrow]

I don't want to believe benches, well even ARC-AGI until i see the model in action

[mystic panther]

I need help with smth.. I generated one photo and it doesn't let me anymore

[torn mantle]

we are so back craig

[sharp mirage]

idk who gone read it but

"Code Red" Performance Focus

Context: GPT-5.2 was a "Code Red" release, meaning it was fast-tracked specifically to address competitive pressure from Google's Gemini 3, which had outperformed GPT-5.1 in reasoning and coding benchmarks.

Philosophy: Unlike GPT-5.1, which introduced user-facing features like "personalities" and tone controls, GPT-5.2 is a "performance-first" update. It focuses on reliability, speed, and raw reasoning power rather than new experimental features.

2. Reasoning & Reliability

Scientific & Math Reasoning: GPT-5.2 Pro and Thinking models show significant gains in high-level benchmarks like FrontierMath and GPQA Diamond (graduate-level science), surpassing the capabilities of GPT-5.1 Thinking.

Logic & Multi-step Tasks: The model is much better at handling long chains of logic without "losing the thread," a common issue users reported with GPT-5.1 in complex workflows.

Reduced Hallucinations: There is a strong emphasis on "groundedness," with GPT-5.2 showing an estimated 80% reduction in hallucinations compared to earlier iterations, making it far more reliable for enterprise and research use.

3. Speed & Latency

Optimized Pipeline: GPT-5.2 introduces major backend optimizations that make it significantly faster (lower latency) than GPT-5.1, particularly for the "Instant" model on routine queries.

Smoother Turn-taking: The chat experience is described as having "tighter logic" and less lag, addressing the "sluggishness" some users felt with GPT-5.1's reasoning models.

4. Coding & Technical Work

SWE-bench Scores: GPT-5.2 achieves higher scores on coding benchmarks (e.g., ~74.9% on SWE-bench Verified), with specific improvements in debugging, multi-file handling, and reduced syntax errors compared to GPT-5.1.

Agentic Capabilities: The model is better at "agentic" tasks—executing multi-step projects like building entire spreadsheets or presentations autonomously, where GPT-5.1 might have required more manual hand-holding.

5. Architecture Refinements

Unified Router: While GPT-5.1 introduced the concept of "Instant" vs. "Thinking" models, GPT-5.2 refines the automatic router to be much smarter at detecting "explicit intent." If you ask it to "think hard," it routes to the Thinking model more reliably than 5.1 did.

Context Management: Although the context window size (approx. 272k-400k tokens) remains similar, GPT-5.2 is far better at utilizing that context effectively, reducing "context drift" (forgetting earlier parts of the conversation) which was a critique of 5.1.

[torn mantle]

we need it added on lmarena RN

LIKE RN

LET ME TEST IT PLS

[sharp mirage]

FR

[spare rune]

Oh it was added

Time to test..

[inner gate]

Gpt 5.2

[sharp mirage]

o lets gooo

[weary galleon]

FAKE!

[zealous sparrow]

On webdev huh

[sharp mirage]

:D:DD::D:D::D:DD::DD: Yea

[zealous sparrow]

NOT FAKE

[mystic panther]

Can someone help?

[zealous sparrow]

It's now added guys

[inner gate]

Did they skip 5.1 or was i under a rock

[weary galleon]

Via Photoshop or Nano Banana?

[sharp mirage]

no

fr

its added

[zealous sparrow]

going to test it out on codearena rn

[sharp mirage]

i saw it now

bye

me go test

[zealous sparrow]

this model takes so long

[sharp mirage]

agaisnt gemini 3

[zealous sparrow]

they wont win fastest model tho

googles new flash models get that point from me

they write 400 lines in less than a min

yeah but the new flashes have quality and speed

still waiting on those

i hope google ships

[spare rune]

Gpt5.2 high is really fast

[weary galleon]

Stop FLOOD!!!!!

[spare rune]

I was preparing to wait like 5 minutes for the reply

Maybe that’s a good thing or a bad thing

[astral blaze]

STOP THE COUNT

[fleet lintel]

benchmarks are good! this will force google to up their game!!

[zealous sparrow]

on codearena it takes long

[spare rune]

Oh

[zealous sparrow]

Well, OpenAI gave me a bad first impression. The first thing i generated on codearena with gpt 5.2 high instantly broke

[sharp mirage]

guys someone

[heavy smelt]

What's the point of the video models in the video arena being randomized? If it is so then what's the point of needing two votes to actually see the models?

[sharp mirage]

send code arena url

i forgot it pls

[zealous sparrow]

Good job SAMA!

[spare rune]

It’s in lmarena

Just click the code icon

The “ <> “

[sharp mirage]

thx bro

[astral blaze]

Wow

[weary galleon]

WE NEED XHIGH ON ARENA!!!!!!!!!!!

[astral blaze]

That's the best openAI can muster huh

[spare rune]

Oh wait I was waiting for it to reply until I saw it freezes mid conversation

[astral blaze]

I'm going back to gemini 3

[sharp mirage]

alr i made prmpt for Html game "who want use it

"

[zealous sparrow]

Im doing some testing for gpt 5.2 high

[sharp mirage]

for who want to use it

[zealous sparrow]

so far it made one fully broken game

[spare rune]

What’s Xhigh

[zealous sparrow]

GOOD JOB SAMA!

[weary galleon]

No, it's just HIGH!

[spare rune]

are you high

[fleet lintel]

why? benchmarks are looking great!

[sharp mirage]

no one will get hacked its prompt for spiderman game

so everyone can test

[spare rune]

lmarena is free

[weary galleon]

Huh?

[sharp mirage]

?

??

[cloud zinc]

xhigh is different than high

[astral blaze]

LOOK AT THE BENCHMARKS PLEASE DO NOT ACTUALLY USE THE MODEL

JUST LOOK AT HOW GOOD IT DID ON SWE

[spare rune]

I choose to believe Claude opus was taking its time in code arena until I noticed it just kept going forever saying creating index html

Sob

I wonder if it’s happening for gpt too

[fleet lintel]

I am serious. Are you saying that actually use is not great?

[spare rune]

No

[astral blaze]

I thought you were joking

[spare rune]

It’s a joke

[sharp mirage]

gpt is taking so long time

[zealous sparrow]

Im running side by sides with GPT 5.2 High and gemini 3 pro

[fleet lintel]

no means "not great" or "no.. it is great". ? 🙂

[spare rune]

I think it’s either the code taking a lot of time or stuck

[obsidian cargo]

well that was fast

[astral blaze]

Gemini 3 is clearly miles ahead I are we using the same 5.2

[sharp mirage]

same

[spare rune]

They are joking

Being sarcastic

I think it’s stuck too

[grave plaza]

they just released it haha

[spare rune]

Never mind

It worked

Oh

[astral blaze]

I do hope that people go back to gpt so that I can get more gemini 3 uses

[spare rune]

The output is good

[echo aurora]

It's going to be Battle mode since that's what we use to build our leaderboards. The model's names not being shown until there are 2 votes is so it doesn't bias the votes. All votes after 2 votes don't contribute to the leaderboards (as the names are now exposed).

[spare rune]

Ish

[grave plaza]

guys is kat coder pro in lmarena? when yes i use it

[glacial mulch]

is 5.2 any good

[devout vault]

chatgpt be releasing the worst models that are never #1 on the leaderboard

[spare rune]

The link in code arena how much minutes does it last and can other ppl see it?

[zealous sparrow]

gpt 5.2 high made this
https://019b0eac-c31e-741f-b0e3-598bb5904f74.arena.site
PROTIP: TRY NOT TO CRASH!

the game is insanely broken

GOOD JOB SAMA!

I APPLOUD!

[sharp mirage]

BRo

wtfff

:/

[torn mantle]

and nothing for cursor

[devout vault]

chatGPT will never dominate the AI industry

[echo aurora]

The preview link? It shouldn't expire and others can see it if the link is shared with them.

[heavy smelt]

Ok got it

[zealous sparrow]

80% on SWE btw, and makes piece of s-

[devout vault]

prob all fake results

[fickle venture]

What the heck is this GPT-5.2

[spare rune]

Well the response is buggy

[torn mantle]

lol this is cool

[zealous sparrow]

cool and guess what

touching a bullet

[rugged abyss]

Well kinda sucks as ive discovered a neat bug in the first 30s

[zealous sparrow]

breaks the game

[torn mantle]

no

its working fine

[stray aspen]

How good is the new gpt

[torn mantle]

ah right

true

[devout vault]

it's stupid

[zealous sparrow]

moments before grief

[spare rune]

Woah

[sharp mirage]

bro gpt 5.2 is coocked

[astral blaze]

None of these buttons work

[sharp mirage]

cooking

[zealous sparrow]

flop, and scored 80 % on SWE while being the worst coding model ever

[sharp mirage]

rn

[spare rune]

The app is buggy. But the gui is good

[zealous sparrow]

exactly

[stray aspen]

🥀

[zealous sparrow]

the gui is as-

[odd geyser]

And you still haven't figured out what the problem is with chats that close over time and you can't log in?
I'm sorry, I could have done as it was said in that manual, but I don't have a PC or laptop. That's why I'm wondering if someone sent you the necessary data.

[spare rune]

Gpt is actually good at backend

[sharp mirage]

1131 line of code i wish if its work

[gusty helm]

how's gpt 5.2?

[heavy smelt]

I have another question, how is it legal for lmarena to offer paid models completely for free?

[torn mantle]

its meh at coding ngl

[gusty helm]

good/bad/overhyped?

[stray aspen]

I'll stick to Opus 4.5 then

[zealous sparrow]

and OpenAI wants to argue they beat gemini 3 pro

[golden ocean]

no way gpt 5.2 was real and its not a frontier model

gg its over for openai

[cloud zinc]

they pay it out of their own pocket

[zealous sparrow]

atleast gemini 3 pro often gives me bug free stuff

[rugged abyss]

Overhyped by Twitter, an alright coding model. Claude still dominates

[primal nacelle]

Yeah i also want to know. it's annoying

[stray aspen]

Seems like a rushed product just to "keep up" with the competition

[zealous sparrow]

claude and gemini were never dethroned

[astral blaze]

They are not beating gemini with this lol
The world knowledge on this model is clearly short of gemini 3

So it's another codemaxxed model. Congrats sama

[rugged abyss]

Yeah i was honestly hoping for more Competition as im mainly using Claude and Gemini

[astral blaze]

Pichai remains undefeated

[sour spindle]

Google stock down 3%

[cloud zinc]

randomized because it is for testing which is better. two votes because they need another vote beside u

[gusty helm]

yeah, I had the same feeling; it's solid but overhyped cause fanboys + crazy marketing

[sharp mirage]

bro gpt is bad

[echo aurora]

No new updates sorry to say

you can't log in?
This doesn't sound familiar, did you make a post in #1343291835845578853 ?

[spare rune]

Gpt 5.2 high codex pro max when

[devout vault]

gemini 3.0: free
gpt-5.2: paid 1000000 dollars a month

[zealous sparrow]

who made it better
https://019b0eb1-becc-7abd-865d-dc86a83fc504.arena.site
first link is GPT 5.2
second link is Gemini 3 pro
https://019b0eb1-becc-7f0a-ad2f-6ef5adc697e7.arena.site

[sharp mirage]

gemini isnt making a game for me

[devout vault]

obviously gemini 3 pro

[zealous sparrow]

they want 14$ for the output

[sharp mirage]

i dont think so

[zealous sparrow]

i say we sue