general

[surreal zephyr]

Claudes already cooked

[light sleet]

Yep

[surreal zephyr]

Stargate = datacenter
Terafab = manufacturing fab

Js merge already istg😭

If they merge we get asi in 2 years

[light sleet]

😭

[light sleet]

stargate in my country 😭

[surreal zephyr]

Groks hallucination rates on 10t model, but as efficient as gpt 5.5?

On cheaper hardware?

[light sleet]

Iran dont boom it too pls

[silent tree]

Same

Let's get @surreal zephyr a ticket

[surreal zephyr]

Iran cant boom musk

Lol

[silent tree]

when stargate is fully done

[light sleet]

Well I mean in April they did upload a video

Of Stargate

[surreal zephyr]

Thats openais

And still

Musk is the guy who sold flamethrower to civilians

[light sleet]

yea

[surreal zephyr]

And literally sells electric armored personel carrier with sloped armor, 360 vision, remote control, and weapon bed

Aka cybertruck. Totally not military vehicle

Not to mention most satellites are musks

[light sleet]

😭

[surreal zephyr]

Js wait for mass driver on moon that might be conveniently aimed towards earth

🤔

[pine meadow]

Can artificial intelligence create an AI-powered application? No Apis No powerful device

[light sleet]

Hm

[surreal zephyr]

64 fat mans 💀💀

One starship can wipe a country

If fully loaded

[light sleet]

💀

[surreal zephyr]

In summary, a single Starship carrying 64 Fat Man bombs could end Iran as a functioning modern state by targeting its people and economy

Yeah yall dont realise how much power musk has 💀

[light sleet]

Xd

[surreal zephyr]

Or 1400 standard w88 warheads in one starship.

(1 starship btw)

[light sleet]

Damn

[surreal zephyr]

This one is peak

This one is peak too

xD

[mellow nexus]

is there any models that can handle colour correction if given a photo for reference? that's without changing anything at all except the colour

[light sleet]

💀

Tuffship

[loud herald]

Ah yes the time when Scam interviewed Elon and Elon had to correct him when Scam called it a company, Elon said its a non profit organization

[loud herald]

Like an image model?

[mellow nexus]

Yeah, or even a text model that could send me a lut file

[loud herald]

If you mean just editing a color or somthing, I'd suggest GPT Image-2 or Gemini nano banana

[mellow nexus]

iv'e tried gemini nano banana and it slightly changes the image itself as well as the colour

[loud herald]

Then try Image-2, its more consistent

If you also say

At the end o the prompt

Something like

"Keep the image quality the same as the original" it will be more real to the original and if you ask it to change like say the color then be like "just change exactly what I told you, change absolutely nothing else." that should improve it too

Or get a better result

[mellow nexus]

Will do. And when I add two images, should they be titled something particular in the metadata? how will I explain to the model which one is the reference and which one I want changed?

[loud herald]

I normally just say something like "Image 1" and ofc the first image you put it then if you want it to have something to do with it just say like "Image 2" or "The second image I sent" and "The first image I sent"

[pine meadow]

looke at this

[loud herald]

Kimi 2.6 being up there in coding is so valid

Kimi 2.6 is SOTA and the best open source model out there

Its underrated

[silent tree]

Code Categories Correction: 5.5 being the top

dont trust designarena mostly

its not even meant for an arena lol

[loud herald]

??

Its the same as LMArena lmao just no direct or model selection

But has more options

Which means, peoples selections aren't biased

[silent tree]

source: trust me bro

[loud herald]

The source is on the site

[silent tree]

designarena is literally meant for website building 💀

it has full on database

and stuff

and it has way less members too

[loud herald]

Its meant for mutliple things not just website building

[loud herald]

That doesn't mean anything

[silent tree]

part of your language ig il stick with it

5.5 beats opus

its the truth

grok is coming to beat opus too

also a truth.

[loud herald]

Grok is coming to beat everything

[silent tree]

will you cry for your opus?

[loud herald]

The only thing opus has on everything is creativity

[silent tree]

(5.5 does too actually)

[loud herald]

You seem to have the idea that I like opus or something, quit assuming its cringe

[silent tree]

I support both Grok (x.ai) and ChatGPT (OpenAI)

[silent tree]

yeah because you disagree on gpt 5.5 being better

[loud herald]

If you are "choosing" a side then you really are just brainless

[silent tree]

side?

didnt i just say

[loud herald]

You can go through the chats, I've never in my life said that

[silent tree]

I support both

x.ai

and

openai

is that one side?

[loud herald]

Yes its one side against Google and Anthropic and basically everything else then 😂

[silent tree]

um I like Gemini too

[loud herald]

Then say that next time

[silent tree]

when did I say im against "Google"?

thats basically me saying you disagreed with gpt 5.5 being better than opus

say that next time too

-# (in your language btw xd)

[loud herald]

You didn't but thats the same thing you're doing to me, I've never stated that I think GPT is worse than claude

[loud herald]

No, that's assumption, what I said was just the truth

[silent tree]

same thing youre doing to me too lol

[loud herald]

If you say you choose a side then it means you just aren't on the side of anything else

[silent tree]

when exactly did I choose a "side"?

lol

[loud herald]

What I said wasn't assumption

[silent tree]

I said I support both openai and xai

[loud herald]

By supporting specifically "XAI & GPT"

[willow mirage]

hrllo

[silent tree]

i didnt explicitly

say

[loud herald]

Hello

[silent tree]

that I hate all other ais

[silent tree]

never said specifically

maybe its in your language too soo

[loud herald]

No but thats what you are insinuating

[loud herald]

Naming specifically those two out of everything else would mean specifically those two

[silent tree]

no wonder hes 38

[silent tree]

so u want me to name out every other companies out there

are you dumb lol

😂

[loud herald]

You said 37 and 40 yesterday, which one is it

[loud herald]

Sure

[silent tree]

Hmm

[loud herald]

or general sense

[silent tree]

Probably 37

[loud herald]

But you didn't need to say you support GPT thats for sure, you like to make it your entire identity

[silent tree]

you literally want me to name 90k+ companies

Lol

[loud herald]

Thats why I just said a general sense

[silent tree]

in your language too

[loud herald]

Because you wouldn't know 90k+ companies

[silent tree]

cool language bro

[silent tree]

yeah like you do

💀

[loud herald]

I never said I do or insinuated I do

I just said you wouldn't know them, nobody would

Who on this earth would genuinely remember every single company out of 90k?

[silent tree]

you with your language

[loud herald]

Sure

[silent tree]

🔥 🔥

[willow mirage]

where is video arena pls tell me

[loud herald]

At the end of the day, theres no reason to hate a model and no reason to love one, they're AI. They all are trained differently and all have something the other doesn't

[silent tree]

The Video Arena is currently accessible through: https://arena.ai/video. More information on how to use Video Arena can be found in this article

[loud herald]

Diversity with these models only increases the rate of growth

And thats good

[rigid holly]

Hey the giant text size still was not fixed

[azure cove]

anyone know what prompt can bypass claude restrictions

[silent tree]

Movie Ending 😭 im crying 😭 😭 😭

[loud herald]

Wrong place to ask

[silent tree]

Everyone supports each other now 🥹

[azure cove]

where to ask

[silent tree]

amazing

[rigid holly]

Depends whixh restrictions you are talking about

[light sleet]

I dont think it's really a good idea sharing that here.

Might wanna delete it though

[azure cove]

scripting

[loud herald]

I haven't really read through the rules to know or not so you're probably right

[rigid holly]

Yeah ok i have no idea about scripting

[loud herald]

If you try to use it on Arena, Arena will likely not even let it send, they put guardrails for specific JBs, but if you use it on actual claude theres a chance they suspend your account

[azure cove]

oh

[loud herald]

The easiest one I found it works on was Kimi 2.6 and Kimi is pretty good for coding, I'd say try it on the official kimi site though

Anything more will have to be in DMs

[azure cove]

yes

[silent tree]

If this happens, no other model is gonna be able to compete at all.

although sam would mostly be a btch on it

[light sleet]

True

[main iron]

When is Claude opus coming back to lmareana?

[loud herald]

Its in LMarena

Just battle mode

[main iron]

No it’s not

Really?

Why not aswell in direct chat?

Now how does one know it’s in battle if both lm are anonymous

[loud herald]

Elon wants sam to be gone and also for openAI to pay $134 Billion, the only way OpenAI doesn't go bankrupt is if they add elon to the royalties but Elon doesn't want that, its not looking good for them

I mean I'm hoping they stay up but theres so much proof against it

[loud herald]

because when you select which one was better it shows the names of both models

[loud herald]

It's too expensive

[silent tree]

elon as ceo would be better

[loud herald]

I 100% agree

[silent tree]

Finally

[sweet knot]

The Arena has been performing terribly over the past few days. Responses constantly get cut off, making it impossible to get a complete answer. The exact same LLMs that used to produce full, well-structured outputs now return incomplete, broken responses that just stop mid-way. This is not a minor glitch — the platform is essentially unusable right now. What is going on? Arena simply isn’t working.

[shy mason]

In a good way

Not in hate way guys please do not cook me 🙏🏻

[loud herald]

Fair, Craig though only yaps about how chinese models are and nothing he says is really correct

How did you even make that?

[shy mason]

Gpt image 2

[loud herald]

Dang, I'm impressed it kept the pfps so intact\

[light sleet]

ayo?

[shy mason]

Same

[loud herald]

Mannnn

I was describing my architecture all wrong before

[shy mason]

Idk why I added openai logo, but yeah the information here is correct, unless you and watermelon approve of the last one.

[loud herald]

Lmaoo

[shy mason]

defeated was actually more like him winning but then @light sleet said not to continue it soo

[silent tree]

Accuracy: 99.5%

[pine meadow]

hi

[inland ember]

Have people already got funnel results?

[shell sentinel]

https://x.com/kimmonismus/status/2051630990456660044?s=20

Chubby♨️ (@kimmonismus)

Gemini 3.2 spotted in Gemini!

If we already receive Gemini 3.2 Flash now, the major release will probably be reserved for I/O.

h/t @Waguri_Kaoruko8 for finding

X · May 5, 2026, 11:52 AM

[kindred sequoia]

Nano Banana 3 soon? 👀

[pine meadow]

no Nano Banana 2 Pro

Fasting

[supple zenith]

Hey, the recaptcha is taking too long to load images to select, is it only happening for me? Do anyone have any solution for this issue?

[kindred sequoia]

That model from alibaba really bad from what i've seen on a video from ai search

dont know why its there as the 1st

[pine meadow]

How did it get a 2 rating in just a few days, with a difference of only 16 points between it and the first Seedance 2?

[light sleet]

Yoooooooo

But it's inspect element

[pine meadow]

A new free website has emerged that generates AI-powered tools.

[supple zenith]

Hey, the recaptcha is taking too long to load images to select, is it only happening for me? Do anyone have any solution for this issue?

[pine meadow]

sometimes

[shell sentinel]

maybe veo new model will be come？

[shy isle]

anyone have experience in using battle mode compared to direct chat for school projects?

im wondering if its better since most of the premium models have now moved out of direct chat function

[median smelt]

Oh, hell yeah hopefully.

[kindred sequoia]

Have you used veo for video generation? after i tried seedance 2.0 it feels like half-year behind lol. looking at how visible the AI stuff on the veo results compared to seedance

but sure no hate for the google bros, they may releases a more powerful ones soon

[median smelt]

Yea, but hopefully something like veo 3.2, veo 3.5 or veo 4 comes out that can finally beat sora

Well, not beat sora but at least be close enough to it

[kindred sequoia]

what's missing from sora is it cant use image as refs, but the realism sure pretty great just like a real tiktok videos

[median smelt]

yea

I made this with sora a few hours ago

[kindred sequoia]

but yeah unfortunately they have to shuudown it lol

[median smelt]

Hopefully the next veo model can get to something similar like this

[kindred sequoia]

why does it shows a google search engine aat the start lol

[median smelt]

bypass prompt for copyrighted characters lmao

[kindred sequoia]

great idea, does we have the same restriction with veo or seedance?

have you tried them withthe same prompt tho

[median smelt]

sorry i dont understand what you mean by "we"

[kindred sequoia]

bruh

[median smelt]

I wish sora never shut down bro

or at least it went open source/weights

so another company can run

[kindred sequoia]

you just didnt understand how the company ( openai ) works

[median smelt]

Yea, they never make anything open source unless its some text model

[kindred sequoia]

you can understand it just by watching a 20min essay video on youtuube about it, and you realizes how terrible the company was

[median smelt]

I know i know

[kindred sequoia]

i thought they were conspiracies by peoples who are cons to AI, but it turns out the evidence are becoming too much

ah sorry, what do we talk about earlier? XD

[median smelt]

uh we were talking about next veo model

which, if its just another variant of veo 3 (like veo 3.1) im gonna be disappointed

[kindred sequoia]

yeah just like gemini flash and flash lite, they releases the ultra-cheap ver of veo 3.1 instead lol

[median smelt]

At least we have happyhorse 1.0, for now

But I'd rather keep using sora 2

[kindred sequoia]

that's with happyhorse? what prompt did you use

[median smelt]

I kinda forgot, but it was something along the lines of "Zootopia-style scene: Nick Wilde is cooking burger patties at McDonalds because he got fired from his previous job".

[kindred sequoia]

mine always got same bad results, like its the same wan 2.x models lol.. but i guess my prompt is more about american cartoons rather than chinese ( which where htey based on )

[median smelt]

This is also happyhorse

[shy isle]

anyone have experience in using battle mode compared to direct chat for school projects?
im wondering if its better since most of the premium models have now moved out of direct chat function

[median smelt]

Same for this, but its kinda bad at animation.

[kindred sequoia]

sure battle mode is better since you can compare which result is greater than the other.. but what kind of school projects is it actually?

[shy isle]

well I wouldn't just be using it for one, i normally use ai to help with english, coming up with ideas for design, coding apps etc

[shy isle]

yo this is rlly good ngl

[kindred sequoia]

Oh cool, so how's about it now, have you find models that weorks best for u

[kindred sequoia]

Yep, it got the characters right, but the motion are kind of still clunky

[shy isle]

well i was using gpt 5.4 high beforfe them removed it and now im using 4.2 high

in direct chat

[kindred sequoia]

that's grok right?

[shy isle]

but im considering swapping to battle mode

[shy isle]

?? chat gpt

[kindred sequoia]

i mean, the 4.2

oh are you typo

[shy isle]

o sorry 5.2 high

I usually like to make a chat thread for each subject/project im working on

and upload a few pdfs about the task and my current progress etc

and just use that same thread the whole time so that it knows what im talking about

which is why im using chatgpt cuz before when they had sonnet 4.6/7 in direct chat there were like ratelimits

and sometimes it made me open a new chat thread

[jaunty mist]

holy hell, this is pricey (2.5$ for a 15 sec vid)

[kindred sequoia]

do you use it from arena or the gpt platform?

[median smelt]

Nah, I use it via a discord bot

[shy isle]

arena

[kindred sequoia]

that's free right

[median smelt]

Yep

[kindred sequoia]

why didnt you use it from their platform

[jaunty mist]

the model itself is, i mean

[shy isle]

cuz i don't wanna pay

[kindred sequoia]

arena discord channel?

[median smelt]

fair

[shy isle]

arena is free gpt 5.2 high

[median smelt]

No, its websim bot

[jovial palm]

Has anyone found a way where after a refresh of the page if there's a model that's continuously generating to get that stop button back to cancel the prompt that's seemingly stuck?

[kindred sequoia]

ah i c

[kindred sequoia]

websim?

[shy isle]

nah u might have to just make a new thread

[median smelt]

Its a discord bot called websim bot

[kindred sequoia]

Can i integrate it to my server myself?

[median smelt]

yea, but you have to join their server to use it

[kindred sequoia]

or do you access it from another server

[median smelt]

but you can use it outside of the server after that

[kindred sequoia]

oo

ok cool

[shy isle]

yeah unless you have any free alternatives?

with maybe a better model than gpt 5.2 high

[kindred sequoia]

huh, its hard to find some freebies for AI nowadays

how heavy do you use AI tho

[light sleet]

5.3 Codex

[kindred sequoia]

also i see that you also on MJ discord, do you works with image genreations?

[shy isle]

i don't see it

[kindred sequoia]

yep its for coding

[shy isle]

every day quite a bit

[light sleet]

It's for code only

Codearena

[kindred sequoia]

try search it on the code tab

[shy isle]

I used to play around with it when it first came out like back in 2021 or 2022 lol

[shy isle]

ohh

is it better than 5.2 high though?

like for non coding tasks

cuz most of the time im not using them for coding

@light sleet

[whole mural]

very weird

gemini 3 flash in arena knowledge cut off is 2024Jan

[kindred sequoia]

they may trained it with the older informations to cut cost but still uses the same engine for same intelligent result

[kindred sequoia]

its a distilled ver just for a coding task, so expect it to performs worse with general tasks

[short sluice]

suggestion: i feel like sending images without text should be possible

especially if no text is needed

i dont like having to write "." in every prompt that doesnt need an image

[misty star]

@echo aurora when will xhigh 5.5 be benchmarked on arena 😭

[next trout]

what's the point xd it'll just be #1

[misty star]

😭

[echo aurora]

Sorry to say I don't have an ETA for you 😭

[covert totem]

free claude opus dem

and 10+ other model

[light sleet]

Code arena is only for code

[fickle hawk]

Site cannot be opened

[waxen bridge]

which website?

[slender ledge]

What discord

[echo aurora]

Can you try a different browser? The Brave browser tends to have some problems with Arena.

[wicked sage]

who the fuh is uni

https://tenor.com/view/uni-cute-cat-silly-tuxedo-cat-unico-gif-1696378561949539858

[echo aurora]

Debutting!

[wicked sage]

What does this mean.

[obsidian cargo]

is it flow-state?

[echo aurora]

from Luma Labs - https://lumalabs.ai/app

[wicked sage]

OH

Ok ty

[surreal zephyr]

ask in chinese

[stray aspen]

How

[wicked sage]

yo bro what is this discord bot 🙏

[light sleet]

Uni is good

[rigid crane]

Share

[median smelt]

Idk

[wicked sage]

can ihave it

pls

[echo aurora]

Yeah pretty good ranking

[rigid crane]

Which is the best ai for Roblox exploiting

[winter loom]

do we have Claude Opus 4.1 model in Arena?

[wicked sage]

are we fuhking deadass

[uneven cypress]

Uni is goated

[obsidian cargo]

@echo aurora uni not working for me

[stray aspen]

That looks like openrouter

[whole swallow]

Show generations

[wicked sage]

exploiting in the big 26

[echo aurora]

Hmm will look into in one moment.

[rigid crane]

Why not

[wicked sage]

why do

https://tenor.com/view/newgens-man-gif-11700273215215303658

[rigid crane]

It's easy way to get money and robux

[wicked sage]

i swear to god if you mean steal a brainrot ot whatever im actually going to kill you in gta

https://tenor.com/view/compedant-meme-descendant-companion-son-gif-229683454138273454

ok too far

sorry

ok but seriously i have no idea
probably the ablitirated models

[median smelt]

me

[wicked sage]

in hf

like uhh

Hold on let me check

give me like 5 minutes

[rigid crane]

It's steal a brainrot I made a duel script and make it paid but rn it is detected

[wicked sage]

anyways sorry that i was mad

[light sleet]

Wait

[feral elm]

same error here

[wicked sage]

ok this is good

https://huggingface.co/paperscarecrow/Gemma-4-31B-it-abliterated

[zinc pewter]

https://tenor.com/view/scuba-black-cat-cat-gif-16349549187699415106

[wicked sage]

@rigid crane if you can run this, this is prob the best choice

[winter loom]

do we have Claude Opus 4.1 model in Arena?

[rigid crane]

Ok thx

[wicked sage]

actually

theres more but idk im checking rn

theres so much bs on my screen

[uneven cypress]

Guys what is the best model in arena ai (direct chat) for writing and coding

[stray aspen]

Ass models

[stray aspen]

Gpt 5.5

[zinc pewter]

claude something something

[wicked sage]

they cant be THAT bad

[uneven cypress]

Gpt is not in direct

[wicked sage]

that one

[light sleet]

Gpt 5.5

[wicked sage]

bro WHAT are we doing

[stray aspen]

Deepseek or kimi

[wicked sage]

qwen3
0,6b
bruh

[uneven cypress]

It's not in direct?

[light sleet]

Gemini 3.2 flash

For frontend

[winter loom]

Hey guys do we have Claude Opus 4.1 model in Arena?

[wicked sage]

❌

[uneven cypress]

Yea battle mode I think?

[light sleet]

3.2 flash might be the model 3 flash is using rn

[stray aspen]

No thats obsolete

[light sleet]

People say

[wicked sage]

is yo ass in the future or smth
either i live under a rock or there is no g3.2f

[stray aspen]

There is

[uneven cypress]

He meant 3.1

[light sleet]

G 3.2 flash is rolling out

[stray aspen]

Its a new and better gemini flash

[echo aurora]

Yeah you can find it in Battle mode

[wicked sage]

https://cdn.discordapp.com/attachments/1365589433319030794/1483644383411699894/image.gif

[light sleet]

Oh no I meant 3.2

[wicked sage]

oh its still here???

[stray aspen]

Its ik battle made

[winter loom]

Many Content Creators are making false claims like we have Claude Opus in arena

[wicked sage]

probably outdated

[stray aspen]

Theres a smarter gemini flash

[uneven cypress]

I don't think they remove any old models?

[wicked sage]

but as pineapple said, its in battle mode

[median smelt]

Are we deadass

[wicked sage]

only there

sadly

[uneven cypress]

Bro there are secret models in arena ai

[wicked sage]

actually looking in the message youre replying

what the fuhck is a duel script

[median smelt]

does it look like I know

[zinc pewter]

people 1v1

[wicked sage]

FAIR ENOUGH

[zinc pewter]

and who wins, gets the brainrot

[median smelt]

i dont play steal a brainrot

[wicked sage]

https://giphy.com/gifs/brainrot-shitpost-ysn01-kb05kuOVqNb4xq6VIy

is there a gemma 1

[median smelt]

https://tenor.com/view/stop-scrolling-jesus-letter-pfp-gif-14541080913545839969

[uneven cypress]

What is that model basalt-0422-1-v2

[wicked sage]

or is there just gemma 1.1

[wicked sage]

#codename-discussion if it doesnt have like a logo or whatever

[uneven cypress]

Yea it doesn't have logo

[wicked sage]

ok so its an unknown model

[median smelt]

its actually codename for asphalt which is codename for potato-3423 which is codename for what im guessing is GPT-6o

[wicked sage]

try searching it on this server and see if its

what the hell when did they make another omni model

[median smelt]

tomorrow

[wicked sage]

my predictions are people are gonna use gpt-6o to roleplay like what they did with gpt-4o

[uneven cypress]

Gpt-6o is out?

[stray aspen]

Nah

[wicked sage]

kind of

[median smelt]

HappyHorse 1.0

[wicked sage]

i think

[stray aspen]

The spud will be used for coding

[stray aspen]

Your in treys server?

[wicked sage]

same here

lol

im actually a mod in their server

[stray aspen]

Its the guy from the sora 2 bot

[median smelt]

When am I getting another role

[wheat ore]

kling 2.0 free?

[median smelt]

Yes

[wicked sage]

What

offtopic but what even is the diffirence between obliterated and abliterated

shut UP

[median smelt]

No one gives a sh*t bro

[wicked sage]

no offense

[wicked sage]

^^ this

like bro this is deadass just max

[median smelt]

I wanna pick my model myself

[wicked sage]

but with extra steps

i think

no this is literally max if it was a website and not models

[median smelt]

Its a bot

#ask-here

They posted here too

[wicked sage]

oh

https://tenor.com/view/squid-game-oh-my-god-bruh-saleswoman-oh-my-god-bruh-gif-2398788864743270124

[median smelt]

At least I think its a bit

bot

[wicked sage]

yeah its def a bot

[median smelt]

feels the aura

[wicked sage]

W

what did you use for this btw

[median smelt]

Doubao app which uses seedance 2.0 fast

[wicked sage]

Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated

[wicked sage]

alr peak

[median smelt]

Me when ✨️le sora✨️ is shutting down

[wicked sage]

its not that serious

[median smelt]

What is the context bro

[wicked sage]

Oh yeah if you wanna know what im looking at uh

hold on

[median smelt]

Please tell em thats not the system prompt

Of some ai

[wicked sage]

https://huggingface.co/huihui-ai/Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliterated

SHUT UP

[median smelt]

SHUT UP

[stray aspen]

Stop spamming gang

[wicked sage]

@echo aurora idk if this is a bot but just in case im gonna ping you so you can like timeout this dude

[echo aurora]

Thanks! All set. 🙂

[median smelt]

YESS

[vast star]

what do y'all think?

[whole swallow]

I think therefore I am

[vast star]

I am therefore I think

[echo aurora]

Think I am therefore

[vast star]

I am

[stray aspen]

https://d.tmpfile.link/public/2026-05-05/856e547c-e845-480e-bd9a-576f07d8432b/sora_video_1777997348_1.mp4

[wicked sage]

peak

[polar horizon]

how do i use this on the web

[wicked sage]

wdym, like host the ai?

[polar horizon]

yes

[rocky geyser]

Hii

[silent tree]

Have you checked my modmail yet

it's been 4 hours.

[echo aurora]

I have not

[astral jetty]

How does max model work

Can it only pick between the options in direct chat or all the models in the database

[silent tree]

please do it's just a report to just let people know to not share those kind of stuff here.

[jagged fjord]

this looks so real

[echo aurora]

I'd encourage you to check out the blog post about Max: https://arena.ai/blog/introducing-max/

We don't have a list posted of all models (and varients) of what Max will pull form.

[jagged fjord]

cant wait for gemini3.2/3.5, gives me hype

[thorn coral]

....

[silent tree]

Pineapple missile

Incoming

[astral jetty]

This doesn’t help

[stray aspen]

@verbal kite

Bruh

[astral jetty]

My question was specially about whether or not max pulls from models that were recently removed from direct chat

[stray aspen]

Im stupid

[astral jetty]

The changes that happened after the post was made

[stray aspen]

<@&1349916362595635286>

[vagrant temple]

Has anyone else tried the new agent mode?

[stray aspen]

I dont have it

[vagrant temple]

why?

[echo aurora]

It's currently an experiment, meaning not everyone is going to have it.

[echo aurora]

We're really hungry for feedback about this feature! Please share what you think in #1498702173650030756

[echo aurora]

We don't have a list posted of all models (and varients) of what Max will pull form.

[rigid crane]

What that mode do

[shrewd citrus]

woah I’ve never heard of uni 1.1

looks like it got a crazy good benchmark result tho

[light sleet]

yay

[next trout]

yay

[errant sand]

MAX got image upload

https://tenor.com/view/crushed-it-gif-18087987023179949748

[echo aurora]

It's a multi-modal chat experience where you can work across different modalities within a single workflow

[echo aurora]

Shared with the team, I don't think is yet on API

[echo aurora]

Other modalities too!! Putting out an announcement in a bit.

[light sleet]

New Announcement Incoming 🔥 🔥

[fluid ruin]

Then give me the damn thing 😭

[fluid ruin]

Its claude with tools and different models

The only real change is different models

[toxic verge]

https://arxiv.org/html/2604.14807v1

[velvet forge]

fast question

claude opus 4.7 or 4.6

[fluid ruin]

4.7

[rocky geyser]

Hii

[rocky geyser]

4.6 id rather not use a tokenizer that uses up 30% more tokens

[desert fiber]

hi , is there a page limit for pdf in battle mode ? i have uploaded 35 page and i keep getting this error

[honest flower]

oh, it again in direct chat

[toxic verge]

The arena agent is 😎

[echo aurora]

Tell us more!!! #1498702173650030756

[toxic verge]

Is that your guys in-house agent?

[echo aurora]

Thanks for sharing the Trace ID, looking into.

[toxic verge]

I’m really digging the loading feature when it loads that’s pretty awesome pretty creative

[echo aurora]

When did this error take place? Was it recent?

[desert fiber]

i mean when i uploaded the pdf and the erorr image

im trying to split it into a half and send 2 pdf 18 page instead , now i am waiting to see the respond

[echo aurora]

Would note the chat session is going to have a 100 page limit. Regardless if it's 100 1-page pdf uploads, or 1 100-page pdf upload, if it goes past 100, it'll result in an error.

[light sleet]

Is this a bug

[rocky geyser]

Yayyyyyy finally :D

[echo aurora]

Nope, we added it back

[light sleet]

YOO

Tuff

[light sleet]

Now that's another hint on the new usage system coming 🔥

I'm excited

[rocky geyser]

Yayy tysmm!

[honest flower]

At first glance, I was actually a bit scared, lol

[echo aurora]

I wouldn't read that much into it. When we have more info to share about the new usage system, we will.

[light sleet]

Kk

[rocky geyser]

I hope the usage limit will be transparent like you see the exact usage and i hope its not too restricting 😭

[honest flower]

Well, I reckon it’s still more than the standard limits for the Gemini 3.1 Pro

[toxic verge]

The new Gemini model better not be trash

[echo aurora]

you see the exact usage
Yes, this is the plan. Current rate limit system not having a clear indicator on-site for where it's at can be a better user experience.

[light sleet]

3.2 Flash?

[toxic verge]

They got a new Omni video/img model allegedly

Or an update to their current one

[light sleet]

@echo aurora

[rigid crane]

Free?

[rocky geyser]

Yea but i have never managed to reach even a single usage limit but i believe that wontbe the case with the new one...

[manic shore]

do you people ever actually run into usage limits? i havent ever been blocked from chatting, only on some longer sessions of claude's thinking models, but even then i could still just change which model i use and everything owrked fine

[fluid ruin]

How tf did yalls get qwen 3.5 on here

[toxic verge]

It’s good that they got the top bottles out of there

[rocky geyser]

I have google ai pro 😅

[fluid ruin]

Also I

[toxic verge]

It makes it more fair for the other models to get used

[fluid ruin]

DONT REMOVE THE GEMINI 3.1 PRO DAMN IT

[toxic verge]

People adapted real quick

[fluid ruin]

DONTTTT

[light sleet]

They added it back relax

Pineapple said its added again

[fluid ruin]

Oh

YEWWWEW

[light sleet]

For real this time

[rigid crane]

Why I can't upload txt file

[fluid ruin]

Yeehaw lil bro

[light sleet]

Gemini 3 Flash taking 4 minutes 🤨

What is bro cooking

[echo aurora]

We don't yet support a txt file upload 😭 We do plan to expand these abilities.

[rigid crane]

Ok

[light sleet]

Nvm bro Gemini 3 Flash is NOT Gemini 3.2 Flash

[echo aurora]

Never hit a rate limit or a context limit?

[fluid ruin]

Pls dont tell me there is a 3 message ratelimit

Per day

[echo aurora]

They sure do. An easy way to see this is using Video Arena as that Arena has the easiest to hit rate limit (2 per 24/hr)

[fluid ruin]

I easily uploaded zip to claude but in arena it dont let you do that

[desert fiber]

can anyone send a pdf file and see if it does send an error back or it's only me

[fluid ruin]

What

2 per 2.4 minutes yes?

[manic shore]

ah, so the video generation and "build apps and websites" tabs just eat up rate limits?

[echo aurora]

The modalities are on their own rate limits. So Video Arena is going to be different from Code Arena, and so on.

[manic shore]

so people who just use text/search shouldnt be effected much by the changes?

[echo aurora]

Also worth noting individual models in Direct and Side by Side will have their own rate limits.

[manic shore]

so for example "ai 1" would have a seperate limit in Direct mode and Side by Side mode, so if you deplete the limit in direct mode you could still use the ai in side by side mode?

[rocky geyser]

Oh right i hit a context limit in one of my chats once 😭

[rocky geyser]

Oh yea i always think if i do one now i only have 1 left so i never do them 😭

[echo aurora]

Not really. "AI 1" will share the same rate limit from Side to Direct. But all modes (Battle, Direct, Side) will have their own overall rate limits ontop of those model rate limits. Although the specific model limits don't apply on Battle. It's a bit confusing, I'm very very excited for the new usage system to make this all a lot more clear.

[echo aurora]

Yeah it happens 😭

Very hopeful the new system is going to make this much better.

[rocky geyser]

Will we maybe get the new models back like opus since it doesnt matter how we spend our "tokens" or whatever we get?

[echo aurora]

That is the intent. I can't make promises for what this will look like when released, but the hope is that this new system will allow for the returning of those models to Direct and Side by Side.

[rocky geyser]

I hope so 😭

Im excited but also scared at the same time

Since newer models werent available on arena anymore i bought gpt plus, gemini pro and claude pro but that is so expensive 😭

I love the codex app tho its amazing

I hope there wont be weekly or mothly limits in the new usage system...

[echo aurora]

[still musk]

Hellooo

[light sleet]

never mess with the Great Pineapple Cave

[manic shore]

does anyone know i there are filters on the model select? so i could filter for models with vision/file upload/both

also what does vision even do?

[echo aurora]

My daily routine

[errant sand]

vision alone works for images like ocr things, if it has file upload and vision it will work with images and pdfs.

[echo aurora]

There isn't a filter on the drop down, but the little indicators next to the model's names will provide that information. The little glasses icon means it can have image upload to be viewed via Vision Arena. For the document icon, means it has PDF upload support, so similar to vision, but only for PDF instead of IMG.

[jovial palm]

hah which model?

[manic shore]

a filter could be useful due to the really large amount of ais, or you could group the ais into categories like gemini, grok, claude, etc to make it easier to browse

[loud herald]

API, how else lol

[errant sand]

hey @echo aurora would MAX work pdfs?

[fluid ruin]

They say they do inference

[loud herald]

Wdym

[fluid ruin]

Like hosting the ai on the backend is inference

[compact flame]

Yoo max finally updated

That nice

[loud herald]

Yea via api

[fluid ruin]

If you locally run AI models, you're running it with inference

[loud herald]

Then no

Lol

[echo aurora]

Images for now, PDF upload isn't yet there.

[wispy roost]

Man I love this website been using it for a few days now being able to hit the limit and only take a hour for it to let me use high end models again is insane unlike other paid for products that only give you say 200 pro model searches a week

[errant sand]

ok perfect thank you

[loud herald]

They have Claude and GPT and other closed source models, they can't it inference

[next trout]

is it good tho

[wispy roost]

@echo aurora Sorry for the @ but do you know how to donate and or who the donations go to and what they are used for Im planning on just donating 20 a month that’s what I paid for perplexity and that site is so bad compared to this free website

[limber hound]

ayo since when gemini 3.1 pro came back

[shadow jewel]

will lmarena ever have MCP server connections?

[next trout]

wai9t it's actually back

[echo aurora]

Since ~an hour ago

[main plover]

no way

[echo aurora]

Possibly! Nothing that I'm able to share at this point.

[echo aurora]

Way

[limber hound]

was new max pushed already?

learn more still redirects onto older post

[next trout]

probs need to update that

[echo aurora]

Lol I appreciate that! But we don't take donations.

[next trout]

why not

[echo aurora]

Yes, multimodal Max is now live.

Should see it now.

[echo aurora]

That just isn't our business strategy.

[sleek urchin]

Pineapple did you take my idea of #1497332625281384649

???

[stray epoch]

oh so multimodal max wasn't a bug, i noticed it appear around 20 hours ago but i thought it was some visual bug thing

[limber hound]

the horror

[fluid ruin]

Yeee

[echo aurora]

We were working on this before you made this post. But if you'd like to think this one was all you, I wouldn't discourage that.

[sleek urchin]

@echo aurora

[sleek urchin]

🙁

[echo aurora]

Oh no!!!

Hard refresh?

[paper island]

yeah where'd it all go

[sleek urchin]

I predicted

[limber hound]

twice

[sleek urchin]

Doesnt work

[echo aurora]

Heck

[next trout]

oh its gone for me too

[slow basin]

what is arena shampion?

[paper island]

@echo aurora same issue i think max stole all the modals 😭

[next trout]

something broke

[echo aurora]

What browser? Is this just mobile?

[limber hound]

google chrome

[next trout]

on desktop too (brave)

[paper island]

Mr.Altman... I don't feel so good...

[echo aurora]

Why can't I repro this

[stray epoch]

😔 😭

[echo aurora]

Thank you all for flagging! I've report to the team.

[signal robin]

Im confused, what does the new update mean

[queen veldt]

Same on android lol

I don't think it's an issue

It's the final update

[paper island]

This is the new system to save Arena money /j

[sleek urchin]

"The bugs, the bugs"

[limber hound]

calm before the 3.1 pro and max overusage

[sleek urchin]

👉 ❌ ❌ ❌

[paper island]

This is it, the frontier:
Your Brain

[paper island]

genuinely gonna use max for everything now that it's multimodal

[queen veldt]

Can't even see what model won lmao

[sleek urchin]

👉 ❌ 👉 💹

[limber hound]

• a lot of claude 😈

[paper island]

WE ARE BACK

[limber hound]

yep we back

[echo aurora]

Others seeing the same?

[upbeat finch]

does anyone know what's going on and why there are no bots?

[paper island]

great reaction time @echo aurora!

[errant sand]

GEMINI 3.1 PRO IS BACK

[queen veldt]

Works fine

[echo aurora]

What you mean?

[errant sand]

you the best pineapple

[next trout]

models are back now

[light sleet]

Sora 2 pro

[stoic shard]

Yup, Max on all options as well, thanks Pineapple!!!

[upbeat finch]

I don't have any choices for choosing a writing bot

[light sleet]

Xd

[echo aurora]

https://klipy.com/gifs/mochi-mochimons-6--k01KQWS90CQNMBYS9WAKWYXV448

[jovial palm]

Sora's going away though right?

[queen veldt]

Battle works too

[zealous berry]

please bring atleast one opus model at least opus 4.6 only

[queen veldt]

What is this tetra seems not smart at all lol

[light sleet]

Already did, API stays till September

[jovial palm]

ah

[sleek urchin]

@echo aurora I FOUND NEW BUG!
When you go to battle mode and select what model is better it doesnt display
[Model] vs [model] (So it doesnt display what model you got) sorry for ping

[queen veldt]

Now it does

Try again

[sleek urchin]

Opus 4.6/7 (Not funny) cost same

[light sleet]

bring 5.5 😎

[queen veldt]

Who uses light mode btw lmao

[light sleet]

5.5 normal

[zealous berry]

what about mythos then?

[light sleet]

Aero

[sleek urchin]

No it doesnt

[queen veldt]

Boomer

[topaz epoch]

What about AGI then?

[next trout]

u need to hard refresh (ctrl+r)

and new convo

[sleek urchin]

I am on android

[silent tree]

Fake model or unproved

[queen veldt]

It didn't work for me either

[errant sand]

what about terminator then?

[sleek urchin]

Tried

[queen veldt]

But once the models were back it works normally

[next trout]

clear ur cache

[silent tree]

The Spud is better than Mythos

[next trout]

true

[silent tree]

(5.5 xhigh is Mythos Level too but even beats it at some point)

[next trout]

yes

facts

this man is speaking only facts rn

[topaz epoch]

Nice jokes

[sleek urchin]

Now it works

[silent tree]

yes

[echo aurora]

Can you share the Eval ID of this chat session? Eval ID is the random set of numbers/letters in the URL of a chat session.

[silent tree]

prove mythos right now

I want ONE proof

one accurate one.

[next trout]

there's none

[sleek urchin]

Lemme get it (1min

[silent tree]

Lol

[main nexus]

5.5 > mythos

[topaz epoch]

I'm not anthropic

[silent tree]

🔥

[next trout]

imagine getting gatekept by anthropic

[silent tree]

so what if youre not anthropic

[zealous berry]

when i was sleeping it came in my dream that's the proof

[silent tree]

wheres my mythos proof :(((

[silent tree]

record your dreams

(pro tip)

[old ginkgo]

Nope

[silent tree]

I heard u a guy from studio lite

[silent tree]

Yep

Its just the truth

[old ginkgo]

Based on?

[sleek urchin]

019df993-afd4-7118-9ee8-510f0cbf6d3f
Now it displays the model 🙁

[old ginkgo]

Some cherrypicked benchmarks?

[silent tree]

isnt that what anthropic showed

fake benchmarks of

mythos

[old ginkgo]

It's rather that they showed

[topaz epoch]

5.5 can't even beat opus 4.7 thinking 😂

[silent tree]

mythos isnt even proved lol, atleast one screenshot of it doing something

[old ginkgo]

A bunch of vulnerabilities

[next trout]

it's so jover for misanthropic!!!

[silent tree]

ok wheres mythos

[topaz epoch]

Yeah you were sitting beside the CEO

[silent tree]

where is it?

[old ginkgo]

If 5.5 as strong as mythos, ehere government concern? Where 0 day vulnerability breaches?

It's hilarious how they gave mythos to a bunch of companies that are even partially competing with anthrophic like microsoft and not one perso went "hey guys, this mythos model actually isn't so good"

[silent tree]

ok now wait.

[next trout]

no because it wasn't framed as "it's going to be hacker stuff" as anthropic did with mythos

[old ginkgo]

I waited a week?

[silent tree]

why is anthropic scared to release mythos?

is it because they have worse security

why did openai release gpt 5.5

[old ginkgo]

Because it's dangerous in the wrong hands.

[silent tree]

I say this point everytime.

why is openai releasing the spud?

[next trout]

u do realize that even opus 4.5 still can discover 0 days

[old ginkgo]

Because its not dangerous enough in the wrong hands.

[silent tree]

because they have better security

If mythos is dangerous then why reveal the model LOL

[next trout]

true

[old ginkgo]

Yeah, but not in major frameworks that billions of users rely on.

[topaz epoch]

Tell to openai beat opus 4.7 first

[silent tree]

@topaz epoch isnt this u btw

[next trout]

oh actually it does

[silent tree]

you literally rely on arena giving u free models

get a job dude

[old ginkgo]

Source: trust me bro

[silent tree]

u dont even know what we are talking about @topaz epoch

[silent tree]

tell stun to stay out of this btw

[old ginkgo]

Im not his boss

[silent tree]

he wants free opus 😔 😔

from arena 😔

give him his free opus someone

[topaz epoch]

Yeah i want

[zealous berry]

there must be more secure millitary grade ai's which we don't know publically i guess

[next trout]

i discovered r/w ring -2 OOB in MSI, Gigabyte, Insyde, and almost all other firmware providers lol with gemini 3.1 pro only

so that should tell you enough

[silent tree]

so what do you know about 5.5 when you havent even touched it lol

[topaz epoch]

Hahaha but at least i don't want 5.5 pgt

[silent tree]

ok?

💀

[old ginkgo]

Source: trust me bro

I'd rather listen to the trillion dollar companies

[zealous berry]

i bet millitaries has better ai's that we don't evn know

[topaz epoch]

What ai military has used ? Open ai ? No!

But anthropic

[old ginkgo]

Seems more credible that none of the employees speak out versus a random guy on discord telling me 3.1 pro can do similar stuff as mythos without ever having touched mythos. Just seems like an unreliable approach .

[silent tree]

btw

[zealous berry]

there's deals must be secret models

[silent tree]

Thanks grok

[echo aurora]

After a refresh?

[silent tree]

@echo aurora

[old ginkgo]

What do you guys think about subQ

[silent tree]

Translated text

[silent tree]

@old ginkgo whats your opinion on this guy

[old ginkgo]

He should speak the english language like the rest of us

[zealous berry]

bring at least opus 4.5

[silent tree]

nvm that was vulgar

[topaz epoch]

Thats way old chat bruh i got already out of that matter

[echo aurora]

Lets avoid reposting it

[silent tree]

Alright

[echo aurora]

That's just brining more attention to it.

[silent tree]

you were still using your own language btw

to hide the words

nothings wrong with your language but the words you said are definitely wrong though

[zealous berry]

bring at least opus 4.5

[sleek urchin]

Bruh heres pricing for all opus model:

[silent tree]

he will bring 5.5

[silent tree]

💀

[echo aurora]

The hopes is with the new usage system we'll be able to bring back some models to Direct and Side by Side

[old ginkgo]

I like 5.5 honestly, but it can't be confirmed that it's better than mythos

[silent tree]

5.5 😔 😔

I like openai and xai

[sleek urchin]

Oh yeah and theres 4.6 fast on openrouter

[old ginkgo]

Anyways do you guys think subQ is legit?

[signal robin]

So the older models are more expensive?

[rugged abyss]

@echo aurora has Max improved since the last mention of it

[silent tree]

Make Elon Musk the CEO of Openai 😭 😭

Elon pls win

Elon win

No sam

[old ginkgo]

Elon has lost his edge tbh

[silent tree]

still though

elon is way better than sam

[next trout]

no, grok is behind

[silent tree]

umm

[echo aurora]

Now multimodal

[next trout]

elon is not doing anything even with his 700b idk

[silent tree]

I'd like to disagree with you saying grok being behind

cuz its actually not

[zealous berry]

elon will make gpt 5.6 like like grok lol worst ai

[rugged abyss]

Thats awesome but I meant in terms of score :^)

[silent tree]

Pineapple missile incoming

[old ginkgo]

His companies used to be far ahead, now everything keeps getting delayed, all the timelines are mega firetrucked and xAI isn't the exception to this at all.

[errant sand]

@echo aurora

[sleek urchin]

<@&1349916362595635286>

[signal robin]

What does that mean exactly

[silent tree]

grok is already good rn btw

[sleek urchin]

Sorry I wanted to ping the scam guy

[silent tree]

Grok 5 will finish off opus 4.6

[zealous berry]

which model do u mean

[echo aurora]

More modalities (image, code (front end), vision). Before yesterday, it was just text.

[silent tree]

latest grok

but 5 is gonna be powerful

[signal robin]

I see

[old ginkgo]

You could say the same thing about gpt 4 if you only use it for first grade math...

[zealous berry]

i didn't tested it yet is it better that gpt 5.4

[next trout]

no it's not

[silent tree]

well not yet

[next trout]

grok 4.3 is not even on the level of glm 5.1

[old ginkgo]

It's not good artificial analysis is a pretty good indicator of overall capabilities and grok is currently around deepseek level. A frontier lab competing with open source models is not a frontier lab.

[zealous berry]

elon just making hype this days only

[shy mason]

I request strongest gpt commander @surreal zephyr

[silent tree]

grok 5 will cook

but

the spud will cook harder

[old ginkgo]

They were already at the peak with grok 4 and then fell off. You pretend like they are just warming up to the game, no. They were already at the top and then fell off. Why would you think they have what it takes after falling behind months after once having had the best model for a brief period?

[old ginkgo]

5.5 is spud

[silent tree]

Its not

[old ginkgo]

Pretty much confirmed

It is.

[next trout]

no

[silent tree]

confirmed by?

[raven stone]

where the catalog code?

[old ginkgo]

Everyone

[pseudo hemlock]

@echo aurora #announcements message

Is this not how it always was?

[silent tree]

and is everyone an admin

no

[zealous berry]

guys what about minimax ai i heard about ut in most claude convo

[silent tree]

spud supports voice in and out

[old ginkgo]

Do you need everyone to always spell it out for you? Sam won't tweet "btw 5.5 is spud, the model we hyped for a month"

[silent tree]

its omni

but 5.5 doesnt

[silent tree]

Its actually not spud 💀

spud is omni dude

5.5 isnt

[old ginkgo]

Source?

[silent tree]

Spud is 6

[silent tree]

source: trust me bro

[old ginkgo]

Send a link, i actually want to read it if its credible

[zealous berry]

is minimax worth it?

[light sleet]

Yes, he is right 5.5 is not at all spud

[surreal zephyr]

beep boop

[old ginkgo]

Source?

[silent tree]

Ask ampro he has all the links

[old ginkgo]

Who's ampro?

[surreal zephyr]

5.5 just earnt me 10k

🔥

[silent tree]

cuz its tuff

also uhh schmebulog says 5.5 is spud

[surreal zephyr]

altman just teased spud

[silent tree]

so could you help him

[surreal zephyr]

realtime is coming soon, alongside voice mode

spud is gpt 6o-realtime

[silent tree]

yep

[next trout]

spud is just going to mog anthropic so hard

[old ginkgo]

Source?

[silent tree]

5.5 already does

5.5 is good at frontend too now

[surreal zephyr]

xitter

[next trout]

it does, but spud harder

[surreal zephyr]

spud sees videos

[echo aurora]

Nope, the change did go in yesterday, but prior it was Text only

[surreal zephyr]

"you havent seen nothing yet" - altman

[old ginkgo]

The word twitter is not a valid source bro

[rugged abyss]

Welp, lets be happy that Gemini releases tomorror

[surreal zephyr]

"prepare to be even more suprised"

[silent tree]

5.5 is crazy

(ampro)

[surreal zephyr]

thats like the main source for news in 2026

[silent tree]

Fr

[next trout]

and it'll be terrible

[rugged abyss]

https://tenor.com/view/spongebob-patrick-water-bottle-bottle-head-gif-27083229

[old ginkgo]

He's saying that literally every week about everything. Do you also use the 39th time logan kilpatrick tweets gemini gemini gemini as a source too?

[next trout]

no, sam always delivers

unlike dario

[silent tree]

@old ginkgo