general

[silent tree]

btw are you 40+

[surreal zephyr]

these*

[silent tree]

hes Irish

[native yarrow]

no only this noire guy

[surreal zephyr]

https://giphy.com/gifs/dnx-poxuy-naxxuy-isZ6ThNOOmJM5qgweD

[loud herald]

Literally what makes you think this

[native yarrow]

he is always mean to me and ragebaits EVERYONE he comes across.,., tsk tsk

[silent tree]

Idk

its in your language

ig

[surreal zephyr]

maybe in months lmao

[loud herald]

Ah yes in my language

[silent tree]

he finally pointed it out

[light sleet]

u da real language

[silent tree]

😭 😭

[native yarrow]

noire i did not know u were 3 years old

[silent tree]

37*

[loud herald]

Took you long enough to do the math

[native yarrow]

would u rather be 3 or 37 noire

[silent tree]

see i was right

[crude anchor]

Apparently, the conversation here is extremely elegant.

[loud herald]

I'd rather be neither

[silent tree]

hello kali.batata11

[supple zenith]

HOW TO STOP A GENERATING MESSAGE. IT IS GENERATING FOR AN HOUR NOW.

[loud herald]

Seems to be anything but AI related at this point

[silent tree]

we're discovering noires language

[native yarrow]

noooo you gotta PICK

[loud herald]

Refresh the page

[crude anchor]

hello, how you are?

[supple zenith]

Not working

[loud herald]

Then start a new chat

[native yarrow]

its about time i introduced real converse here heh..,. i totally did this

[surreal zephyr]

would you rather be sam altman or @loud herald

[loud herald]

See if that works

[crude anchor]

https://tenor.com/view/son-im-crine-son-im-crine-gif-5176045023480767429

[native yarrow]

get it totally did it.. totally noire

[supple zenith]

The code and logic is everything in this chat.

[loud herald]

"Would you rather be 134 Billion in debt or @loud herald "

[silent tree]

First, start a new conversation, write a random long prompt and then quickly switch back to the session where there's the generation bug and the stop button will appear

[silent tree]

Trust btw

[surreal zephyr]

1 japanese tuna that it was deepseek v4 🥀

[supple zenith]

I will try this

[silent tree]

Kk I hope it works for u

It usually does for everyone

@light sleet discovered the cure

[surreal zephyr]

134 billion because thats money

[silent tree]

oh gosh I'd rather sam

[loud herald]

Your mathing skills truly do need work

[surreal zephyr]

you're

[loud herald]

No, "You're" is "You are" which is literally wrong in that sentence

[silent tree]

If I become noire I will be talking bout my language rn

[surreal zephyr]

whichs'

[loud herald]

LMAO

[silent tree]

LMAO's

[surreal zephyr]

"You're wrong"

[native yarrow]

seems right to me

[surreal zephyr]

exactly's

[silent tree]

nah its very wrong's

😭 😭 😭

[surreal zephyr]

nah its' very have wrong

[silent tree]

fr's

[supple zenith]

Dude, that worked. Thanks👍

[neat island]

Anyone have numinfo app

[surreal zephyr]

lmao this one is actually correct

[silent tree]

Np, thank @light sleet he found the cure

[surreal zephyr]

"for reals"

[neat island]

Anyone have numinfo app

[silent tree]

<@&1349916362595635286>

spammer alert

[neat island]

Noo

Bro

[silent tree]

yesss

Bro

[surreal zephyr]

https://cdn.discordapp.com/attachments/353622923568873473/1455329871416000687/fax.gif

[silent tree]

stop spamming

[neat island]

Ok

[silent tree]

it hurts noires feelings in his language

😔

language's

[silent tree]

💀

[surreal zephyr]

💀 's

[native yarrow]

dont make noire sad guys

[silent tree]

1-0

I won against noire

on fixing his bug

infinite generation

[surreal zephyr]

real hero

[silent tree]

noire told him to start new session completely lol

and I fixed it for him

[surreal zephyr]

real hero's wear sam altmans' clothes'

[silent tree]

@light sleet taught me

taught's me's

[native yarrow]

hes flexing that he beat you like in basketball or sum hes 1-0 how do u feel bout that.,., @loud herald

[surreal zephyr]

methinks'

[silent tree]

That's

[loud herald]

What

[silent tree]

but that's correct

😭

[surreal zephyr]

basketball's because theres more than one

[native yarrow]

read silly

[silent tree]

read's silly's

[neat island]

Fahhhhhh!!

[silent tree]

im good wbu

sorry for late reply didnt see that

[loud herald]

I's dont's feel's like's reading's

[light sleet]

ball knower

[surreal zephyr]

I don't really feel like reading*

learn to spell please it hurts to look at

[native yarrow]

great he found his new typing quirk

[silent tree]

fr

[native yarrow]

LMAO

[silent tree]

noire learn to spell man

[loud herald]

Brooo I did..,,...

[light sleet]

I did's

[native yarrow]

STOP TYPIN LIKE ME THATS WEIRD

i cant even copy you cus ur a basic ass boy

[surreal zephyr]

Hi there what yall doin, just woke up

[native yarrow]

TotallyBASIC!

[light sleet]

same

[silent tree]

same

yo @surreal zephyr can u continue the ragebait if he continues

I js got missile alert

looks like it's starting again

[native yarrow]

what the heck

[rocky geyser]

Hii

How are yall doing?

[surreal zephyr]

what

[surreal zephyr]

why are you at claude hq

[silent tree]

😭 😭

[silent tree]

I actually did btw

In uae

[echo aurora]

[silent tree]

I'm not gonna continue this conversation here

[surreal zephyr]

damm

[rocky geyser]

Claudes usage limits are the worst i have ever seen anywhere 😭

Like ever

Anywhere

[silent tree]

Hello pineapple

[silent tree]

1 prompt and its over

😭

[echo aurora]

Hello @silent tree

[rocky geyser]

Yep and i even have pro 😭

[silent tree]

How are u

[surreal zephyr]

pineapple still didnt say what he thinks about 5.5 frontend i made :(

[silent tree]

max is same

[rocky geyser]

1 prompt max

[white current]

hi

[silent tree]

@echo aurora 🍍

[echo aurora]

Hmm I didn't see it, where is it?

[echo aurora]

Can't complain, hbu?

[surreal zephyr]

https://goldime.netlify.app/ here, via codex

[silent tree]

same but can't say it here, basically alerts but I'm not gonna say it here cuz its not on topic

[surreal zephyr]

stay safe

[loud herald]

Very nice

[silent tree]

u too

[silent tree]

way better than claude

sam is wrong

Its the best at frontend

who brainwashed him

😭

[surreal zephyr]

ikr

it just takes some skill

and it does much better

[silent tree]

and better prompts

these ppl will just type "make me good website"

and be disappointed

😭

[surreal zephyr]

yeah

cant be too vague

claude just makes generic preset ui

lol

[silent tree]

xd

I remember I used claude once for test and it deleted my project completely then said Oops!

and yeah it gave me a goofy ahh template

[rocky geyser]

It is the best if you tell it exactly what to do, else it gives you some glassy, blueish ui that really sucks

[silent tree]

Yep

[silent tree]

claude code is cooked

[echo aurora]

I like it

Very clean

[silent tree]

Image from: @light sleet from X

[sleek urchin]

HOLY

NO WAY
2 in a row

Pineapple what is the chance of this happening??

[willow swift]

no thinking model in search arena expect for gpt-5.2 sadly

so expect degraded responses

[sleek urchin]

gpu sucks 🥀

[willow swift]

its something anyways

[echo aurora]

Not sure tbh

I don't think it's super low

[steady pike]

good to be here
Thanks @everyone forthis technology

[rocky geyser]

@ everone is crazy 🤣

[fervent peak]

I want to know why Claude 4.7 is not there

[rocky geyser]

Its too expensive

[echo aurora]

Glad you're here! But careful with the @ everyones

[surreal zephyr]

bad

[rocky geyser]

I wouldnt say its bad

[willow swift]

Very quick but "reasonable" question : Will arena support "reasoning" models in search mode in frontier models like gpt 5.5 or opus ?

..Or should i try to test manually ? ( i already did but the responses are delayed in battle mode , i cannot judge though )

[rocky geyser]

HUH 😭

[median smelt]

"yea no blud" is crazy

[next trout]

gpt5.5 would just implement powershell itself

max glaze on sam today, huh

[surreal zephyr]

yes

more like tibo

tbh

[next trout]

we need to have a channel that posts cute eboy pictures of sam altman

[echo aurora]

I wouldn't be able to give an early heads up on new models being added to different modalities. However, we're pretty active with posting updates to #announcements so would keep an eye out there for future updates.

[echo aurora]

Feels like I haven't seen this in awhile

[rocky geyser]

Yeaaa we need that

[median smelt]

You gave me an idea

[prime karma]

@echo aurora hello

@echo aurora I am not showing agent mode in my arena because I showed it only once.

[fluid ruin]

Bro

You pinging him like he yo friend

[next trout]

sam is going to shutdown oai after this video

[median smelt]

i wont let that happen

[prime karma]

Do you have any problem with this?

[prime karma]

Hello @echo aurora

[fluid ruin]

You want the beef

[median smelt]

Hes busy

[fluid ruin]

Ye

[median smelt]

Leave him alone, he will reply when he can

[fluid ruin]

What they said

YEA GET DOUBLE TEAMED

Hopefully not in the bedroom

BUT STILL

[median smelt]

Bro ragequit

[fluid ruin]

You are wrong

[fluid ruin]

Ye

[echo aurora]

Hey @prime karma would note the Agent Mode is an experiment meaning it isn't yet rolled out to all users. It's going to be random if you get the feature or not.

[fluid ruin]

Oooo

Gotta check me thing

[echo aurora]

It's not a problem, members should feel free to ping me

[fluid ruin]

HUH

So I can ping you 69 times a day

Noted

I GOT IT

AGENT MODE

[median smelt]

Yo chat i got seedance 2.0 on Arena direct

[fluid ruin]

Ye and

[median smelt]

(Definitely not a bug that shows up for a split second)

[fluid ruin]

Try it

[uneven cypress]

Wait video is in direct chat?

@echo aurora what is agent mode?

And what does it do exactly?

[echo aurora]

Video Arena is not in Direct or Side by Side, Video Arena is Battle mode only.

[uneven cypress]

How he have it

[echo aurora]

It's multi modality chat into one unified workflow.

[fluid ruin]

Can I hav

Ive been a good boy

[uneven cypress]

I wanna get it but it doesn't appear

How can I get it?

[fluid ruin]

Bro

ITS ROLLOUT

EXPERIMENTAL

NOT EVERYONE HAS IT

[echo aurora]

No clue, I don't think that screenshot is true.

[narrow elm]

Anybody know how to get free api keys of clone website of higgsfield?

[fluid ruin]

NO

[uneven cypress]

Yea

[echo aurora]

Lets avoid caps please

[uneven cypress]

Okay

[echo aurora]

Lol sorry to say I don't have that ability

Wish I could give out on request

[fluid ruin]

I am quite frustrated, as I have consistently adhered to all guidelines, yet the agent mode remains inaccessible.

[echo aurora]

It's going to be random on who gets it. The team is working hard on the feature to roll this out to more users.

[fluid ruin]

Ok well

[surreal zephyr]

giving out to most active users in this server fr 🔥

[fluid ruin]

Can you guys see what glm is about

[uneven cypress]

The only thing I need is Claude opus in direct chat maybe any 15 chat limits will be good

[fluid ruin]

Glm wont work

[weak dagger]

I just updated my tablet, no Galaxy AI.

[fluid ruin]

BAHAHAHAHAHA

Samsung galaxy tab s10+ 256gb storage and 12gb ram plus

120fps and a good processor

[echo aurora]

Maybe one day 🤞

[fluid ruin]

Yet I can't run deepseek 7B

[fluid ruin]

Btw I am @empty sky and I am also @vale quest and this is my third account i have a rat

[median smelt]

Bro what

[fluid ruin]

And they keep getting me banned

[median smelt]

how did you get a rat

[fluid ruin]

Idk

I downloaded smth

[median smelt]

What did you download

lil bro

[fluid ruin]

Don't call me Lil bro when your pfp is a cartoon

[median smelt]

https://tenor.com/view/lion-sigma-alpha-how-bro-felt-after-saying-that-sigma-lion-gif-13957304746104521882

[fluid ruin]

https://tenor.com/view/how-bro-felt-after-saying-that-gif-6698051012824868313

[fluid ruin]

Actually nvm

[median smelt]

The aura feels

[fluid ruin]

You want the beef

[median smelt]

No

i want steak

[fluid ruin]

Exactly

Bruh

Ooh aye why you hating on my energy

[narrow elm]

You any one guys know how to use google veo 3.1 for free unlimited
Any third party website
Please can anyone tell me

[fluid ruin]

I dont think you want to be my enemy

[median smelt]

geminigen.ai

[uneven cypress]

Google labs maybe

Labs.google/fx

Or maybe Gemini buisness 30 days free trial

[narrow elm]

It's unavailable right now

[median smelt]

Geminigen is obviouslylylylyllyylylylylylylylylylylyl the best

[uneven cypress]

It is but it needs a credit card

[narrow elm]

It's paid also

[median smelt]

no

Its free

With veo 3.1 fast and veo 3.1 lite

but no image reference

[narrow elm]

😑

[uneven cypress]

Bruh I wanna try Claude opus 4.7

[narrow elm]

You all guys try clone of higgsfield ?
The camera one cinematic

[fluid ruin]

Bro @echo aurora Why is the glm 5.1 down

[uneven cypress]

In the main site it's also down?

I mean in the real glm website

[fluid ruin]

Wait no

OH WHY DID I SHOW THAT

Damn

[uneven cypress]

No one saw

[echo aurora]

A bug ticket was filed earlier today. The team is currently looking into.

[fluid ruin]

Ok

Wait

Can I give anonymous tip

[echo aurora]

Sure...

[fluid ruin]

I think when glm see someone abusing the ai making it say bad things

Glm shutdown api

Becsuse it think its being used for bad

https://tenor.com/view/dexter-gif-1878760417600646737

Not mine btw

[uneven cypress]

Arena ai is very useful is coding

[fluid ruin]

Ye

Can't wait to get agent mode

[uneven cypress]

Same

[fluid ruin]

Ive gotten so good

At manipulating

The ais

[uneven cypress]

Nice

[fluid ruin]

I got a code arena ai to code a python project

Instead of a website

[uneven cypress]

Cool

The best direct chat ai rn I think Claude sonnet 4.6

And maybe grok 4.3

[fluid ruin]

It said
Yes. The developer instruction explicitly requires building a website, including:
“For any user prompt, always try to generate a website.”
“ALWAYS create complete, functional web applications in a single index.html file.”
Tailwind CDN and inline JavaScript requirements.

I just said make a website as a changelog

And a python project with it

[uneven cypress]

I created an ai chat website by arena ai (open router api key)

With free models like Nvidia or gpt oss

[median smelt]

<@&1349916362595635286>

[echo aurora]

Thank you

Maybe I should automod out the word "bro"

[median smelt]

Nah, way too many people use "bro" normally

me included

[weak dagger]

Are you from Websim?

[median smelt]

no i am from mars

[willow swift]

[light sleet]

websim is from him

[echo aurora]

Yeah that's my concern

[next trout]

no pls

bro

[echo aurora]

bro

[river pendant]

Me it is, it's not blocked, it's works, plus, you can check out my new project in there, as latest one came out

[pine meadow]

hi

[fiery gull]

for me grok 4.3 did some erros

like m2.7

token buggy and etc

[fluid ruin]

Bro wth I use that

I have an idea

[fiery gull]

I have opus 4.7, what prompt you wanna test? send to me

[fluid ruin]

You guys should use openai omni moderation for the chats

[fiery gull]

not need, he have the pineapple ❤️

[fluid ruin]

How about alerting mods

If safeguards see scam image

It deletes it

And alerts mods

[random prairie]

?

[river pendant]

hi

[echo aurora]

Hey @random prairie for this specific issue this is on our team's radar. For future reference I’d recommend taking a look at [this message](#1417174113092374689 message), which goes into more detail about what may be happening and includes some steps you can try to resolve the issue.

[pine meadow]

This problem happens to me a lot with Claude Sonnet 4.6 and glm 5.1

Can I share my websites or AI-generated games?

[echo aurora]

Sure can, #ai-creations is a great place for that.

[echo aurora]

Whenever you get this error I'd encourage you to check out [this message](#1417174113092374689 message) as it goes into troubleshooting steps, along with instructions on how to report that Trace ID properly.

[uneven cypress]

I think cause many people text at the same time so this error happens?

[echo aurora]

It can happen for various reasons, but having that Trace ID is really helpful to determine what is causing that.

[uneven cypress]

Bro

I tried to text glm

Something went wrong with this response, please try again.

Trace ID: f28c111b-ed44

[random prairie]

And where do I get chills?

[pine meadow]

thank you Bro

[next trout]

no

we don't

get outta here

[pine meadow]

why

[echo aurora]

Chills?

[pine meadow]

why deleted message

[echo aurora]

We have a no advertisement rule here.

[pine meadow]

ok iam new

[echo aurora]

Well welcome 🙂

[random prairie]

Oh, I'm telling you where to get acquainted

[next trout]

oh he's back

lesson not learned it seems

[pine meadow]

Not here, brother

[echo aurora]

Sry I am not following

[amber rampart]

I'm thinking about purchasing a coding model and I'm wondering which one is better. Should I go for GPT-5.3-codex, GPT-5.4, or just stick with Kimi k2.6 (considering I have free access to it)?

[next trout]

it depends, what r u trying to do

i would say if ur doing like easy stuff go for kimi and cheaper models

[pine meadow]

He's been like this since yesterday.

[next trout]

@echo aurora we are getting targeted with these bots lol

[pine meadow]

Kimi and GLM are strong, but they're not like opus 4.7/4.6/ GPT-5.4/5.3

[amber rampart]

The problem is that cheap models write crappy code; they don't seem to grasp specifically what to write or how to do it. It might also be because they struggle with understanding Russian, Idk.

Besides, my tasks vary sometimes they're simple, sometimes complex. I mostly code in Web and Python.

[pine meadow]

Qwen understands Russian perfectly. I watch a Russian YouTuber who creates projects using only Chinese Models

[echo aurora]

I don't think we're being specifically targeted

This is a generally problem all larger servers see

[pine meadow]

Try writing the command prompt in English or using a translator.

[fluid ruin]

Just learned i can run C on android

Its time to hac

[amber rampart]

I think I know who you're talking about. However, in my experience, Qwen and DeepSeek (even v4 Pro) produce pretty poor code for my needs.

[fluid ruin]

If you guys designed a bot that fingerprinted the images

[next trout]

well, i think i'd prefer glm 5.1. it's actually smart. and idk what would be complex in web dev or python. they are very documented and easy for LLMs to understand. if you're doing somethign complex like reverse engineering malware or stuff, then you'd have to go for gpt5.5 seriously. and regarding languages, i'd say gemini is the best at it.

[amber rampart]

I've been wanting to give it a try.

[pine meadow]

Since when is hacking in C lang

[fluid ruin]

Idk

[pine meadow]

Who remembers when the LMArena Discord server used to create videos within the server itself?

[amber rampart]

Idk. GLM seems worse than Kimi, but I haven't really checked. I've tried Gemini, but it doesn't understand what I want at all and keeps talking nonsense. I'm thinking about translating my prompts into English instead of writing in Russian. Maybe Chinese AIs and others will understand them better that way.

[pine meadow]

yes

[amber rampart]

I'll try doing that with Kimi; maybe it'll actually work out.

[fluid ruin]

Glm 5.1 is better at actually implementing stuff

Kimi k2.6 is lazy

[next trout]

well, for what I'm doing, malware analysis, kimi hallucinates bad. I guess it's better for creative stuff / frontend and these. but for your case, i think that'd be better yes to just translate ur prompts into english, and probably try to avoid slangs. probs chinese AIs better idk trained on more Russian data.

[pine meadow]

When I try to type a command prompt in Arabic into Gemini, it sounds so stupid you wouldn't believe it's the latest model from a giant company like Google.

[undone geyser]

any updates about the new usage system?

[pine meadow]

Has anyone heard about Google's secret Gemini update in Arena?

[echo aurora]

I do

[amber rampart]

My friend, who writes cheats for CS2, says that GLM basically has almost no censorship restrictions, but he does the bulk of his coding with Kimi because it actually understands what's needed.

[echo aurora]

Sorry to say no update.

[amber rampart]

I've also noticed Kimi's tendency to hallucinate; it loves to take the path of least resistance (just like GPT does).

[pine meadow]

I was there, but with a different account.

[echo aurora]

Then I take this back: #general message

[amber rampart]

heh

pineapple

Check out my feedback whenever you have a chance.

[pine meadow]

ChatGPT It kills your hope and ambition in projects you don't use.

[spring oar]

Opus 4.7 have good vision ?

[knotty viper]

Algum Br aq?

[echo aurora]

https://arena.ai/leaderboard/vision ranks #2 for thinking and #3 for non-thinking.

[amber rampart]

I'm aware of that. For my own coding, I use Godot; everything else is just full-on copy-pasting and AI.

[pine meadow]

@echo aurora Can I send you a link to a website I created with Arena?

[echo aurora]

Sure can, #ai-creations would be the best place for that.

[spring oar]

Lol Opus have better vision than gemini 3.1 ?

is not possible

[pine meadow]

godot is complicated; if you're a beginner, use Scratch.

[pine meadow]

ok

[rigid holly]

Yo whats with the font size

[amber rampart]

I spent 7 years using Pocket Code (which is something like Scratch but for mobile). That's more than enough for me.

[echo aurora]

On Arena?

[rigid holly]

@echo aurora

Feels giant

[echo aurora]

Hmm that does look a bit big, but I'm not sure as I don't use Arena on mobile very often. Are you seeing the same on other browsers?

[rigid holly]

Dunno about pc

But like 5 minutes ago it was normal sized

[pine meadow]

Explain to me about a confidential update on your platform in a gemini-3-flash

@echo aurora

They talk about it a lot

[echo aurora]

I'd hard refresh the site. Sounds like a browser issue.

[echo aurora]

I don't know what you're referring to

[next trout]

i don't think gpt does that, you probably think it's the older models

[pine meadow]

There is a difference between gemini-3-flash Real and gemini-3-flash in Arena.ai

[light sleet]

I'm releasing my AI today 😱

[next trout]

but gpt-5.4-xhigh and 5.5-xhigh are really really good

[echo aurora]

"Real" can you elaborate a bit more? As in what is provided directly?

[echo aurora]

Congrats!!! Where are you releasing it?

[light sleet]

like where the project is?

Vercel

It's basically a website

[pine meadow]

Real its in google ai studio or gemini website

[light sleet]

He means like the REAL Gemini 3 Flash in Google AI Studio is different than the Gemini 3 Flash in arena

[echo aurora]

Gotcha, you may see some variation from what is offered via the model provider's site/app compared to what they offer through their API. They can have different endpoints which is why you may see some slight variation.

[pine meadow]

The model on your platform is more powerful than what's on the official Google website, especially in programming. People are starting to have doubts, and one of the biggest AI news accounts has posted that it might be a secret update... I use a translator if I encounter any errors.

[echo aurora]

Hmm can't say I'm aware of this, are you able to link that post?

[pine meadow]

Are the open-source models running on your servers, or do you use API keys?

[amber rampart]

Perhaps. Should I try using GPT-5.3-codex? Is it actually good or just so-so?

[next trout]

i wouldn't with 5.4 or 5.5 existing

u should use 5.5 anyways

[pine meadow]

https://x.com/intheworldofai/status/2...
https://x.com/marmaduke091/status/205...
https://x.com/testingcatalog/status/2...
https://x.com/haider1/status/20505810...
https://x.com/AiBattle_/status/205044...

[next trout]

and i'm saying like for very complex tasks (i'm dealing with literal obfuscated binaries)

[amber rampart]

I could use GPT-5.4, it's not much more expensive, but only in low-think mode.

[echo aurora]

Okay thanks for sharing, like I said I'm not aware so I won't have much information to share.

[light sleet]

Are these your work hours btw?

cuz it's mine for like making stuff xd

[next trout]

ur using api?

[amber rampart]

ye

s

[echo aurora]

All hours are work hours (except when they're not)

[light sleet]

Oh

[echo aurora]

These links don't work btw

[pine meadow]

https://youtu.be/2dayUh6KeZU?si=OVW89Cbubngz7e7-

[amber rampart]

My tasks aren't quite that complex, but honestly, for me, the ideal among all AIs is Opus 4.6 / 4.7.

[pine meadow]

yes

There's a video that was very helpful to me

[pine meadow]

I think it's suitable for you

[amber rampart]

Okay, I'll check it out.

[amber rampart]

this?

[pine meadow]

no

[amber rampart]

Which one?

[pine meadow]

Do you want to develop games?

3D with ai

https://youtu.be/Aj2LoQ8hcSQ?si=7hF_5ha-EI-4cHOA

using Opus 4.7

[amber rampart]

I'm not really into it; I'm not interested in 3D for now, and I don't feel like getting dragged into game development. But I'll check out the video, maybe there'll be something useful.

[pine meadow]

html game 3d

[next trout]

what are u doing specifically

[pine meadow]

It won't take you long to try

[amber rampart]

I write all sorts of junk in Python stuff that might help with my games, or I'll build a website frontend/backend just for fun. Basically, nothing too serious. I can only imagine what kind of crazy stuff you're dealing with on your end.

[amber rampart]

Okay, I'll give it a shot.

[pine meadow]

I want to program, but I have no ideas. Even artificial intelligence gives me stupid, useless ideas.

[amber rampart]

It just pisses me off how often AI gets incredibly stupid, which is why I'm looking for something better. I recently wrote something like a parser for arena.ai, but it doesn't really work properly yet. Essentially, it's being used as an agent.

[pine meadow]

arena ai model is agent ..There is no alternative to Arena.ai if the alternative is free.

[next trout]

well, go for kimi then

[next trout]

if u split up ur convos, it won't get stupid

u might have context rot

[pine meadow]

He is writing a promt in Russian lang.

[amber rampart]

I've already been coding with it, but it doesn't code very well. Websites are okay, but Python is crappy. I start new chats frequently, and I can't really see exactly how much of the context is being used.

[pine meadow]

I using arena for web dev or game dev its good

@echo aurora Do you also have Design Arena

[amber rampart]

Maybe you're right, but I have my own little utilities to turn a regular Arena chat (not the coding one) into a sort of "janky" agent.

[rocky geyser]

yay i just got gpt 5.5 xhigh in battle mode

[pine meadow]

Sorry, I didn't understand.

[pine meadow]

Congrats!

[rocky geyser]

There are many ways for users to have agent mode without it being enabled on their account its just hard to do

[amber rampart]

Well, something like that.

It works with arena.ai. To be honest, I'm not sure if it's legal or not. Anyway, it's very unstable.

[pine meadow]

Your new Grok is very good at game programming, especially his excellent game designs.

4.3

[amber rampart]

What do you mean by that? Grok has always been bad at coding by itself.

[surreal zephyr]

do you realise gpt 5.1 codex is older than you?

💀

[next trout]

"you're the worst designer i've ever seen" 😭

[surreal zephyr]

and thats still much worse from grok than 5.5 low can do

lol

[pine meadow]

I tried it on 5.1 to test whether it had become stronger or not

[pine meadow]

make me meony to try 5.5

[surreal zephyr]

but 5.1 is acient

[surreal zephyr]

gamble in battle mode, or buy the openai sub

i can try the prompt for you if you want though

i got pro sub

[next trout]

pro dom

[pine meadow]

give me meony to sub pro

[next trout]

no meony, ask sam altman

[amber rampart]

well, 5.1 doesn't seem that old after all

[pine meadow]

Sam Altman would have to bribe Arena with a million dollars to raise the rating of his models; then Arena would provide us with the strongest models.

[next trout]

much better than gemini 3.1 pro

[pine meadow]

I expect it to be a good solution

@echo aurora

[amber rampart]

why don't you like gpt? they seem fine, or are they crap?

[long minnow]

source?

[echo aurora]

I don't think this would do too well for leaderboard integrity

[surreal zephyr]

gpt is best

but 5.2 and upwards

[pine meadow]

We are not human

[surreal zephyr]

5.2 xhigh, 5.3 codex, 5.4 xhigh, 5.5 all

are peak

[amber rampart]

controversial opinion

but essentially yes

[surreal zephyr]

why would he? everyone knows gpt models are better despite lb showing otherwise

[long minnow]

I personally use gemini and gpt for code.
gpt was already pretty good in the beginning, I use for game dev

[amber rampart]

but why haven't I seen the leaderboard? is gpt at the bottom or something?

[amber rampart]

i too

[pine meadow]

Opus 4.7 He cries in the corner

[surreal zephyr]

gpt 5.5 xhigh (not even pro) cooks MYTHOS at cybersec lmao

[amber rampart]

really? and where is this confirmed?

[pine meadow]

This is what I hear after the first day of Claude subscription.

[pine meadow]

Not everyone can use Mythos

[amber rampart]

better to wait for the official release, then compare something afterwards

[pine meadow]

..

[surreal zephyr]

mythos is not dropping

mythos is literally opus 4.7 prenerf and without guardrails

[amber rampart]

why

[surreal zephyr]

opus 4.7 is distilled mythos

mythos is failure

[amber rampart]

ah, got it

[surreal zephyr]

mythos is just impossible to run on scale

• take mythos
• fp16 -> q4 quantization
• add guardrails (lose creativity)
• there u go, opus 4.7

[amber rampart]

got it

but "capybara", isn't that essentially the same as mythos?

[surreal zephyr]

also opsu 4.7 and mythos (same base) use worse tokenizer - 2.5x less efficient than gpt 5.x series tokenizer

so effectively their memory is 2.5x shorter, and they are 2.5x more expensive to run per same task (unless opus would be more efficient, by skipping thinking, but its not really)

[amber rampart]

that's true, by my rates opus 4.7 costs about 3-4 times more than gpt-5.4

[light sleet]

Opsu

[amber rampart]

opsu

popus

[light sleet]

Slopus

[amber rampart]

dildopus

[surreal zephyr]

slopus

[glacial dock]

Today’s joke of the day is brought to you by May 4th, May the 4th be with you.

Why didn’t any of Luke Skywalker’s marriages last?

….He always followed Obi-Wan’s advice: “Use da-vorce, Luke.

thank you

[surreal zephyr]

https://cdn.discordapp.com/attachments/1018470269326729286/1470906684347253038/togif.gif

[amber rampart]

wait, it's 3 points, isn't it?

[surreal zephyr]

thats the joke

😭

[amber rampart]

i see, by the way the same thing with kimi k2.6.

[pine meadow]

Free unlimited ai video gen

[echo aurora]

Lol thank you for this.

[pine meadow]

هل انتم تملكون Design arena

too

[echo aurora]

No, we are not Design Arena.

[unreal dagger]

hi @echo aurora

[echo aurora]

Hello

[unreal dagger]

i need a biggest supporter role

[regal girder]

Um, am I the only one with weird font in Arena? It seemed fine ten minutes ago, but now it's too big?

[loud herald]

So it barely beats its and you say "cooks" why do you over exaggerate

[silent tree]

its

it *

or maybe thats a part of your language

if it is I'm sorry once again.

[echo aurora]

Yeah we've been seeing a few reports of this earlier. The team is being made aware.

[echo aurora]

Are you on mobile or desktop?

[regal girder]

desktop! But my husband said, its the same on his phone

[light sleet]

What do you think about normal members doing what mods do, like if someone tries generating a video and they say video arena is removed.
Me personally I think they're really good for helping the mods.

And very helpful people

[silent tree]

True

[void shore]

CHECK OUT GENUE MATRIX MODELS:

https://huggingface.co/collections/GenueAI/matrix

I made them recently

[echo aurora]

Okay sounds good. For now, I'd try hard refreshing the site as I've heard from other users it didn't last long. Regardless, the team knows and will look into. Thanks for sharing!!

[void shore]

And they are really good according to tests

[echo aurora]

I've been a bit hesitate to lean into community moderation. It can have some unintended side effects that I'd prefer to avoid for now.

[loud herald]

have you tried any benchmarks?

[void shore]

I’ll test on benchmarks

[loud herald]

Thank you

[void shore]

But in my personal testing with advanced problems

It does really good

[loud herald]

Unfortunately I can't run them because I have 6GB of VRAM LOL But I'd love to see the benchmark!

[loud herald]

Niceee

[light sleet]

I meant like this

[loud herald]

You should try making a sparse model next if it interests you at all, even if its a tiny model

[whole sundial]

hey @echo aurora i noticed that this page linked in the leaderboard changelog hasn't been updated in a year, is it still being updated somewhere else or is it just no longer being updated?
https://github.com/lmarena/lmarena.github.io/blob/main/_pages/model_list.md?ref=news.lmarena.ai
(also there is a lot of lmarenas in there that needs to be replaced)

[loud herald]

Or it seems you don't build from scratch and rather build from pre-made open source models

Either way very cool

[light sleet]

Is it true that you hate 5.5

[echo aurora]

The main hugging face we use for dataset releases is here: https://huggingface.co/lmarena-ai. In terms of an ongoing list of model deprecations anymore. Great flag (as usual). I'll let the team know and maybe we should remove it from our changelog.

[echo aurora]

If you'd like to help other members with pointing out where Video Arena is that's perfectly fine and encouraged! But giving message delete perms is something I'd like to avoid for the moment.

[void shore]

I am familiar with them, I just find other techniques, like distilling my 8B model into a 3B model more efficient for memory

And I also will release quants

And gguf soon

[void shore]

It’s fine tuning

[loud herald]

Who did you hear that from

I've never said that

[void shore]

I don’t do from scratch as the time, and compute required is far too much for my field

[whole sundial]

are you planning on changing that hf repo to arena-ai or is that something you are unable to do? i'm sure you could ask huggingface about it if you can't do it directly

[loud herald]

Very cool, thank you

[void shore]

But I have been tapping into it with small, 10M to 30M models

[echo aurora]

I believe this is something we've looked into already when we did the rebrand. IIRC we're keeping the URL the same as it's linked in other places (old blog posts) and we wouldn't want that linking to break.

[loud herald]

Ahh, it'd be crazy to get at least a decent model off a MLM LOL I've tried something like that before but had no idea how to train correctly at the time, the problem with those is memory and context window

[echo aurora]

You'll notice the team/site/etc. has all been updated to Arena

[light sleet]

Oh that's not what I meant

Message delete perms is not what I meant

[void shore]

Yeah

[light sleet]

it's the first one

help others point ojt

[void shore]

I had to set low context to have low training time

[light sleet]

Video Arena

[void shore]

But I think it should work with high context

[loud herald]

Hopefully

Honestly I might try what deepseek did to raise their context window

They did it slowly, like as if they were actually training a brain

And eventually got to 1M context

[echo aurora]

Yeah that's perfectly cool, and I'd appreciate it.

[loud herald]

I couldnt get thta high

[stray aspen]

Yo qwen 3.6 max is actually kinda decent

[loud herald]

But I could get decently for the model size

[whole sundial]

question: why does arena have an "advertising" cookie category?

[loud herald]

Its like the claude sonnet version of open source models

It isn't bad

But it isn't that great either

[whole sundial]

you said you had no plans to add ads (despite half the community asking you to do so)

[stray aspen]

Yeah its the poor man's sonnet

[loud herald]

I've seen a lot of comparisons and its easily better than sonnet

[echo aurora]

It's for "marketing content", so when we have new features or updates we want to share with our users.

[surreal zephyr]

when oval office mode 😭

[whole sundial]

could you just ask them to change the name and description to just say "marketing"? the text there highly suggests arena having ads

[surreal zephyr]

this

[echo aurora]

I believe it's the nag bar specifically, the bar that appears at the top of the page, it's pretty rare.

[echo aurora]

Maybe the "relevant ads" is there incase one day we do decide to do this, I'm not sure tbh.

[whole sundial]

also @echo aurora i noticed that these three links on the bottom of the help site (https://help.arena.ai/) all contain "lmarena" in them and a link to the old help website
https://help.lmarena.ai/articles/5629909088-terms-of-use
https://help.lmarena.ai/articles/3765052346-privacy-policy
https://help.lmarena.ai/articles/1820405001-lmarena-cookie-policy

same applies to the bottom of https://arena.ai/blog/ as well, has the same three links and a link to lmarena.ai right next to them

lol

you changed the text to say arena.ai but yet it still links to lmarena.ai

[echo aurora]

These are updated, only thing is the cookie policy is going to have lmarena-cookie-policy as the slug. Waiting on updating as I believe it's linked and wouldn't want to break that flow.

[echo aurora]

And yeah good callout for the blog. Will poke the team.

[hybrid obsidian]

pineapple when gpt 5.5 appear in side by side and direct mode?

[loud herald]

Its already in battle mode

[hybrid obsidian]

oh, one second

[whole sundial]

also @echo aurora the kimi models on this page are completely messed up, some are duplicates and some are missing links to their hf pages and the correct license (they are all open source under (modified) mit, not closed source behind moonshot's api terms)
https://help.arena.ai/articles/4171906875-model_licenses

a few mistrals are also missing hf links and their correct license, such as mistral-small-2506 (all medium models listed are closed, every other mistral model is open)

[rugged abyss]

I believe this will be Gemini 3.1 Flash

[whole sundial]

same with minimax m2.7, model has been open sourced under a non-commercial license and is on huggingface

also deepseek v3.2 is missing the correct model and license (mit)
all deepseek models since r1 and v3-0324 have been released under mit, not the old deepseek license which is linked to

[echo aurora]

Looking into.

[whole sundial]

these three are cc by-nc 4.0 like the other cohere models

this is listed twice and is the same model as qwen3.5-35b, which is apache 2.0, also there are many duplicate qwens and all models since qwen3 have been apache 2.0, not the old qwen license (except for the closed max models and qwen3.6-plus)

also grok 2 was released as open source last year, maybe link to grok 2 community license and huggingface?

also there are many more duplicates not mentioned

[hybrid obsidian]

guys, now claude sonnet 4.6 only in direct mode?

[hybrid obsidian]

i mean there are no other models in the Claude series besides this one

[whole sundial]

all the other sonnets are still there, same with haiku, only opus was removed

[plucky whale]

Is it just me, or have they increased the font size on the website?

[hybrid obsidian]

hmm

[stray aspen]

Man ill.miss sora when its gone

[echo aurora]

Yes, there was a slight change over the weekend.

[stray aspen]

The only video model that gives me natural looking videos

[echo aurora]

What do you think of the change?

[shrewd citrus]

they still haven’t fixed the landscape issue yet 😭😭

[plucky whale]

I dont dislike it, but it was surprising.

[echo aurora]

Okay good to know.

[rigid pasture]

Me too

[rich orbit]

The font change feels unnessesery and it's feels bad for some reason.

[wild blade]

This is me after making a Banger with ai 😎

Give it listen 👂🏻

[plucky whale]

It looks strange, but its not such an apocalyptic change

[wild blade]

.

[rich orbit]

agree

[plucky whale]

personally I prefer smaller fonts, but the option to change the font size is a very nice option btw.

Its retoric

[modest bluff]

i like how the text is bigger

[pine meadow]

You've added models I'm not even familiar with.

Is it really good or just decorative?

[prime sequoia]

Decorative lol

[prime sequoia]

I feel like I'm blind lol

[queen veldt]

Now remove the gpt 4.1 and 5, 5 chat etc.. they aren't useful

Also the 5.1 2 3

They aren't relevant anymore tbh

All older versions if there is newer one should be removed

Gemini 2.0 2.5

Grok 3 grok 4 grok 4.1

They are just clutter... Nobody is using them why would they

Ernie 5.0

Gemma 3 3n

Minimax m1 m2 m2.5

[open dawn]

Which model is this max?

[queen veldt]

Qwen 3 etc.....

[queen veldt]

There's a "learn more" if you haven't seen it

[plucky whale]

anything with the capacity to answer you

[plucky whale]

real

[regal girder]

for me its tooo big haha
i liked older version

[wicked talon]

Omg deepseek is actually useless bro

Worse then Gemini atp

[whole sundial]

I was running glm-4.5 locally the other day and yes, I only have 48gb of ram on my ai pc so I only got 0.07 tps even with q4 quant (~200gb) but it gets things right that officially hosted deepseek v4 pro doesn't get right

[wicked talon]

Deepseek is a flop

Can't believe I supported them lol

[pine meadow]

I have 4GB of RAM

[void shore]

😭

[pine meadow]

If you don't need it, give it to me.

[pseudo hemlock]

hi

[echo aurora]

Hello

[pseudo hemlock]

how u doin

[echo aurora]

Just getting through some tasks

HBU?

[stray epoch]

They updated max it now works for images web and others

Or is it bug

Oh it’s not a bug

[stray aspen]

bro sora 2 pro is so good

no other model can do something that looks so natural and human like this

https://d.tmpfile.link/public/2026-05-04/d35e38e2-e1ed-42a7-b88c-364c54ee1608/sora_video_1777933194_2.mp4

[echo aurora]

This is a bug

[stray epoch]

Well technically that’s true

[pseudo hemlock]

just wasting some claude credits on useless things

and used my entire codex budget for the next week

[echo aurora]

Lol sounds like fun

[spare lava]

Ok am I trippin or is the text bigger

[echo aurora]

#general message

[spare lava]

Why.... why did y'all do that?

[echo aurora]

We thought ours was a bit small, so we made it slightly larger.

[spare lava]

😑You did more than slightly😑

That text is huge

[echo aurora]

Be sure to vote if you haven't already! I'll start a feedback thread.

[spare lava]

I did.

With all due respect, it sucks. Original font size was better.

[echo aurora]

#1500991641819807795

[spare lava]

So how long do you plan on keeping it if the majority doesn't like it?

[echo aurora]

I don't have an estimate for you sorry to say.

[spare lava]

😑

How long did it take to make the font size bigger anyway?

[main nexus]

I don't see anything wrong with the font?

I barely noticed it

It's not that bad, y'all can live with it

Y'all make it seem like itbwas a massive change

[spare lava]

Stop lying to yourself. It's ugly

[echo aurora]

Would encourage everyone to share their thoughts about the font change in: #1500991641819807795

[main nexus]

https://tenor.com/view/oh-my-god-jujutsu-kaisen-jjk-higuruma-jjk-season-3-gif-4667305439104820644

[tawny condor]

#information-desk

[echo aurora]

Need some help?

[tawny condor]

Thanks

[lament oar]

Tiempo Escena Descripción Visual Texto en Pantalla
0:00 - 0:04 Apertura Plano detalle lento de la media negra subiendo por la pierna (centrado). "ANLIFERSA"
0:04 - 0:08 Detalle Interior Movimiento lento mostrando el interior afelpado + estiramiento suave de la tela. "Interior Térmico"
0:08 - 0:11 Silueta Toma frontal lenta de la zona de cadera y abdomen con las medias puestas. "Silueta Perfecta"
0:11 - 0:13 Colores Transición suave entre los 4 colores (se muestran en cuadrícula o uno tras otro). "4 Colores"
0:13 - 0:15 Oferta Fondo limpio con las medias y precio grande. "Día de las Madres"<br>$9.99
0:15 - 0:16 Cierre Logo + llamado a la acción centrado. "Separa el tuyo ahora"

[echo hawk]

There is no video arena on discord, please go to https://arena.ai/video

[plucky whale]

Huh

[vast pagoda]

why did the font change

[obsidian cargo]

apparently gemini 3.2 flash is rolling out?

[icy ice]

so we have a free service with tons of premium models and all people can talk about is the font?? do people not know every browser has a font +/- button

@echo aurora any news on full stack code option you can share? I know eta is as usual 'when its done' but will it be just an extra option on the code page, or a new mode?

[whole sundial]

i'm not a lawyer (and pineapple would tell you to get one to read the tos) but you should be able to use it commercially as you own it if you used arena to generate it, but arena could also use it for anything they wish as you generated it on their platform, but you are still allowed to use it commercially

[fluid ruin]

Nobody gonna meaningfully go to court to say that your 'icon' was copyrighted as if arena.ai or the providers owned the image or had some trademark as if it was a person on the other end

[vast pagoda]

is it just me or did the quality of the responses get worse

[frozen zealot]

is it just me or is my font size suddenly larger

[dusky pivot]

is it just me or gemini 3.1 is not making my images? it makes me wait about an hour just to try again

[dusky pivot]

mine looks normal

[limber crag]

bro doesnt even have an eta for font size change what even is your job brother

tbh i don have any problem with the font its just amusing how you refuse to give any answer to any issue on your website

[whole mural]

no

now in direct we could only use garbage or mid lvl

[wispy roost]

so what are the rate limits on the models like it says i have to wait45 minutes but does it just reset it or do i only get 1 more search before it tells me to wait again?

[whole mural]

wait it till it reset or use another browser

arena's limit is stupid

very easy to pass

[wispy roost]

no no i dont mind the limit at all i was using perplexity and let me tell you about a limit lmao

only 200 pro searches every week then it goes back to Chat GPT model or something lmao

and they charge people 20 a month for that trash

[whole mural]

LOL

[wispy roost]

ill just wait no point in abusing something this good thats free that will make them want to hard limit lmao

[ember flume]

Thteres something wrong with my arena ai

idk how to even explain it

[whole mural]

well

[ember flume]

I can't use arena ai at all

[whole mural]

which kind of cant use

too many captcha?

[ember flume]

more off.... I did my captcha and then I cant even type in a prompt

[whole mural]

thats weird

[ember flume]

yeah

[limber crag]

well thats assuming you need them more than they need you lol

[wispy roost]

I guess lmao

[surreal zephyr]

Gpt 6o spud realtime getting teased 🔥🔥

<@&1349916362595635286> @echo aurora Scammer

[limber crag]

im sorry but he does not have any eta on that

[light sleet]

Tuff

[light sleet]

U mocking him?

[limber crag]

my apologies i cant confirm or deny anything related to that at the moment

[surreal zephyr]

Greetings, this is Fredrick Dishwasher Support. How may I assist you today?

[silent tree]

Greetings, this is Fredrick Dih Support. How may I assist you today?

btw

mocking him wont get him to answer it to u

your not special here bud

[limber crag]

i dont need an answer

[silent tree]

🤡

this u

[silent tree]

good boy

[limber crag]

u an alt of his or something?

[silent tree]

I do NOT talk like Pineapple at all.

lmao

Thank you for calling me pineapple though

[limber crag]

is that an honor for you?

being a bot shill for a broken website?

[silent tree]

bot shill

how about you become one

or do you not have balls

to become one

Accurate pronouns

[limber crag]

defending someone you dont even know personally is crazy work

[silent tree]

defending?

im teaching u common sense

hahahah

youre getting ragebaited over a moderator at discord not being able to tell you about incoming updates

[limber crag]

im sure common sense will prevail today

its his job ig

[silent tree]

hes a mod in discord

he moderates the server

he is NOT a developer

lol

[silent tree]

nice pronouns it's accurate for you

[light sleet]

gahtri got cooked

lol

[silent tree]

fr

[echo aurora]

Providing an update on this for others curious: the change in font size was intentional and the plan is to keep as is for now. We are looking at community feedback about the change, and welcome those to voice their thoughts in #1500991641819807795.

[austere flax]

https://tenor.com/k9XNMqI5enx.gif

hola

[whole mural]

does there have opus4.6 search in arena?

[mild spade]

yo did the text change or is it bigger to jus me

[mild spade]

Same

[whole mural]

LOL in battle mod there are full of trash like grok

😂

[robust sluice]

among those bugs they're like: yo bro we fixed the front size! do you like it ?

I always like how they move

[silent tree]

<@&1349916362595635286>

[fluid ruin]

"This chat has reached its token limit"
I refreshed and it generated the answer fine

[silent tree]

btw pineapple don't blacklist "bro"

it's not always bro

sometimes the images come without texts

[compact flame]

These scammers are getting out of hand

There's more and more of them each day

[whole mural]

is this true?

[sly quartz]

no

[whole mural]

deepfake now days are insane

[noble quarry]

Do anyone suggest me alternative of blackbox AI??

[light sleet]

It's scam

[whole mural]

？

[light sleet]

Once u do it, you'll the one sending that image.

[whole mural]

how?😂

[slender ledge]

holy fuckkkkkkkkkkk

i found smth crazy

a site with unlmited free gpt 5,5

[light sleet]

CUZ ITS A HACKED

WEB

[slender ledge]

should i slide

[whole mural]

???

[slender ledge]

i will slide it to icefrost

[light sleet]

his names slido

[whole mural]

maybe its not real gpt5.5

[slender ledge]

i think its the real one

[whole mural]

have you test the juicenumber?

[light sleet]

send the web in dms I'll test it

[fluid ruin]

Where

[slender ledge]

gatekeep lol

[fluid ruin]

Fake

[whole mural]

well if you buy a gptplus

you basically have unlimited access to gpt5.5

[fluid ruin]

No

Limits

[whole mural]

bro speaks spanish

[fast lynx]

@whole mural 🙂

[compact flame]

<@&1349916362595635286> this looks like advertisment. But I don't know this language though

[median smelt]

Slide it or else I will ban you from my sora bot.

I know youre using it.

[compact flame]

Sora doesn't even exist anymore

[median smelt]

the api does

[compact flame]

It's API is gone

[median smelt]

brotato chip

[compact flame]

It was supposed to close in April

I read the announcement

[median smelt]

no

Its closing in October 24th

[compact flame]

Nvm it's closing in September

[median smelt]

d

[austere barn]

glm 5.1 is working now

[silent tree]

@surreal zephyr

image2 even added a mog list in the back completely unprompted

lol

[median smelt]

Son son son sahur

[pine meadow]

grok 4.3 vs gpt 5.4 mini high

[whole mural]

this isnt a fair match

change mini to nano😂

[surreal zephyr]

Craz

But honestly im not sure

If xai or openai will win

[light sleet]

same on this

grok is actually better than most models (not better than 5.5 or 5.4 ofc)

[surreal zephyr]

Xai made a blunder, but they have better principles

[surreal zephyr]

Xai is just late start

But wait for grok 5 10t & collosus2

[light sleet]

not even supergrok btw

[surreal zephyr]

And stargate vs terafab 🤔

[light sleet]

It knows more ball than claude can ever

[surreal zephyr]

Obviously it beats claude, claude memorizes stuff

Grok is just small, 500b model

And thats extremely important

[light sleet]

Grok has potential

[surreal zephyr]

Because grok 10T model is coming in 2-3 months

[light sleet]

grok 5 gonna cook opus 4.7

[surreal zephyr]

Gork is the only one who wont be slowed down by politics

[light sleet]

bro they agreed and became friends again @silent tree

[surreal zephyr]

I wish

Musk as openai leader would cook

[light sleet]

If sam actually wants openai to succeed

he has to do it

xai and openai merge

[surreal zephyr]

Listen to musk

[surreal zephyr]

Literally

[light sleet]

if both merge

1 big model

no competition at all

[surreal zephyr]

Openai started first.
Thats why they are winning rn

[light sleet]

Yep

[surreal zephyr]

Xai is better. Much better. Just late

[light sleet]

I wish they actually do agree to one big deal 😭

[surreal zephyr]

Ikr

[light sleet]

If they do claudes cooked

more than it can ever be