general

[surreal zephyr]

Overthinked and broken

Avg opus

Inverted axles even

Gpt is reliable.
Opus is creative but dumb

Make your pick

💔

[cyan kettle]

programming

[honest verge]

I'm picking mistral

[surreal zephyr]

(Gemini is on crack and has dementia)

[gusty helm]

gemini

[surreal zephyr]

Gpt's all the way

Opus did worse than even gemini 😭

[honest verge]

Actually Gemini result looks like a mountain

[surreal zephyr]

It looks like poly bridge

Water

[honest verge]

But opus 4.6 did oil

[surreal zephyr]

Crossy road colors

[surreal zephyr]

And 2d

🥀

Opus vs gpt

[cyan kettle]

which gpt

[surreal zephyr]

5.2. High

[topaz condor]

the first one , I guess

[honest verge]

Try xhigh

I'm begging you

Send prompt

[surreal zephyr]

Overthinks and forgets

[honest verge]

If you can't I'm going to do this myself

I want to test it anyway

Maybe it can

[cyan kettle]

where do you access xhigh

[gusty helm]

tbh I dont see major differences between the major players in coding department. But then again I use it as a tool, check on apis and give advice at best. Not to vibe code entire projects. They all mess up at some point; my my goto would be gemini, closely followed by claude (rly more convenience/personal preference on their personality). GPT gives me most bugs/yaps too much while I prefer more on the subject concise answers.

[surreal zephyr]

"low poly animated water, high performance optimised, for game. using threejs and other needed stuff. realistic colors"

[surreal zephyr]

Because everything except gpt is same overthinkers on crack

Gpt is professional even if it thinks less

[gusty helm]

nah, seen GPT spew some plain wrong stuff with confidence

[surreal zephyr]

Gpt>claude>gemini

Rn

[surreal zephyr]

Which model

Im talking 5.2 and 5.3

[gusty helm]

yeah you may have a point there, would need to check whatever license copilot has, but im sure not 5.3

(cant rn)

[surreal zephyr]

@honest verge wheres that one ss of gemini thinking hes made by anthropic

[surreal zephyr]

5.3 is codex only rn, but even 5.2 beats opus and gemini on most tasks consistently

[surreal zephyr]

You arent a dev

[honest verge]

Opus 4.6 did better than thinking lol

Waiting for xhigh

It still thinks

🥀

Opus 4.6 looks good

[topaz condor]

What is this?

[surreal zephyr]

Water

[topaz condor]

lol

[surreal zephyr]

Waves bad imo

[hushed gyro]

Hi errors again on NB Pro

[topaz condor]

do you think it is water?

[honest verge]

WHAT THE HELL DID XHIGH THINKS

WHAT

NAH THIS CAN'T BE TRUE

[topaz condor]

Which OS do you use?

You seem to testing it on linux

[surreal zephyr]

Lmao

Ask it

[honest verge]

Xhigh still thinks

[surreal zephyr]

Lol

Ask it

[honest verge]

This thing sucks

[surreal zephyr]

Xhigh sucks indeed

But ask it

I wonder why

Is it sys prompt or your prompt

Or crack

[topaz condor]

You have your discord server in your profile and what does it do?

I can suggest new features as possible as I can

[honest verge]

It still thinks 🥀

It thinks for 10 minutes already

[topaz condor]

It takes too long.

[honest verge]

And It did absolute trash

[topaz condor]

maybe, there seems to be a backend issue?

[honest verge]

I won't even show it

It's too bad

[topaz condor]

like a network issue

[honest verge]

No

[topaz condor]

try catching logic is crash

lol

[honest verge]

It just thinks too much

Probably spends all tokens

[topaz condor]

ttft and tps are too bad, you mean?

[honest verge]

Ye

[topaz condor]

critical problem

[sturdy mica]

where can i see the public chats

[topaz condor]

they are important factors

[sturdy mica]

arent all chats public

[topaz condor]

this is public channel

[sturdy mica]

where can i find all the public chats

[topaz condor]

you see it now

[sturdy mica]

i mean like that people have put into arena.ai

the conversations users have had with the chatbots

its all public, no?

[surreal zephyr]

Game

[topaz condor]

p2e game?

what engine did you use?

[surreal zephyr]

Rblox

[topaz condor]

If you are python developer, you can run this code and analyze
THe output png is very funny

This is created by calude opus 4.6

[rain bay]

actually its weird

[golden ocean]

real

[toxic verge]

Maybe after 50 rounds and triples checking for hallucinations and verified for accuracy .

If we are to believe some people here then isn’t that a dying occupation?

[golden ocean]

https://cdn.discordapp.com/attachments/1442924145615568956/1442924156546056272/MConverter.eu_v14044g50000d485kpfog65rnjameha0.gif

[night moat]

nano banana always error please fix

[honest verge]

This is cruel

Opus 4.6 thinking can't even make anything

Never use it

Use mistral instead

[left lodge]

They are limiting the generation to be under 6 mins. Which is diabolical, models need more than 6 mins to give outputs worth their cost like opus 4.6

[quasi belfry]

does anyone know how to fix this: https://discord.com/channels/1340554757349179412/1470147097352208394

[hushed gyro]

Ffs why do the errors keep appearing what tf is happening with this platform bro

[sturdy mica]

retard

[loud crag]

Although the errors are painful, I remind myself I'm getting to generate everything totally free, and then I can deal with it.

[left lodge]

You are paying with your data. It's not free

[loud crag]

Yeah that's fair.

[prisma cipher]

I have already found the definitive solution (probably) to the typical error using Opus 4.6.

[golden ocean]

https://github.com/lmarena/arena-rank

[humble tusk]

this has been going on for 30 minutes, I even cleared the cookies, but it's still "Generating..."

[weak dagger]

Reset the page or try again.

[night moat]

still not working

[harsh wolf]

Yo guyz, is gemini 3 pro preview 2k down ?

[mystic raft]

Is there any other websites to use Nano Banana Pro for free?

Because Nano Banana is always down

[fiery widget]

hello guys i got stuck in using gpt 5.1 high

[honest verge]

Guys use mistral

[prisma cipher]

Basic browser-based CS 1.6 game:
https://019c41e8-8c95-77ee-b0d9-5752537a0ed4.arena.site/

[sturdy mica]

@echo aurora will we ever be getting search with every single model, its entirely possible, along with that why not have image and raw text file uploads?

raw text files like png, jpg, html, js, py etc

only pdf files is pretty silly

[honest verge]

Use mistral instead

[astral helm]

yeah it's too much

[sturdy mica]

bro

mistral sucks

are you joking

[honest verge]

No

It's the best model ever

[shrewd citrus]

interestingly 4.6 thinking doesn’t work on other websites either

[sturdy mica]

because its

holyyyyy

for the 100th time in this server its thinking is summarized by another ai

sometimes the ai just doesnt work or denies to do it

[shrewd citrus]

nah I think it just thinks for wayy to long

and it reaches some sort of limit even before it outputs anything

i thought the issue was only on arena

but surprisingly the exact same issue happens on other websites too

[sturdy mica]

im telling you thats how it works

its not thinking for too long?

?????

the exact same issue happens in other websites because its in the api

thats how claude works

retard

[shrewd citrus]

no it thinks for like 5 minutes

[sturdy mica]

bro

[shrewd citrus]

compared to 4.5 thinking which gets its thinking done in a minute MAX

[sturdy mica]

because it has a thinking token cap

[honest verge]

Use mistral

[remote sun]

I sent several messages that violated the Terms of Use while trying to edit them to comply with the rules. However, even after fixing the content, none of my messages are reaching the AI in this chat anymore due to a Terms of Use violation error. Could you please help me resolve this?

[shrewd citrus]

so why isn’t there a cap on 4.6?

deadass why are you so pressed over ts tho 😂

[sturdy mica]

because there isnt

i dont work at anthropic

[shrewd citrus]

https://tenor.com/view/speed-dad-laughing-speed-dad-crying-speed-speed-ksi-speed-trying-not-to-laugh-gif-1466628274697419264

[hollow imp]

Yupp is scem

[hollow imp]

That there is hard 6 min limit, if thats reached when a model is generating a response it cuts off the response and throws Something went wrong with this response, please try again. error

I reported this way back on November of 2025 but they haven't done anything.

Models are literally made to think for hours and here they have a hard 6 min cutoff 🤦

And btw this 6 minute limit is on every single model available not only opus 4.6.

[remote sun]

I sent several messages that violated the Terms of Use while trying to edit them to comply with the rules. However, even after fixing the content, none of my messages are reaching the AI in this chat anymore due to a Terms of Use violation error. Could you please help me resolve this?

[shrewd citrus]

someone said yesterday that if you tell it to think for 30 seconds it might work

[sturdy mica]

were you joking here

[shrewd citrus]

it’s a weird idea but it’s worth a shot

[sturdy mica]

was this a joke or was this a real statement

[stiff trellis]

Any way to get this into CLI? like api or anything

[sturdy mica]

No.

There will never be

[sturdy mica]

this is not a platform to use AI for free

this is a benchmarking platform

[rose sky]

How’s this song I created?

[sturdy mica]

how many seconds did that take

@rose sky

[golden ocean]

24 seconds

[sturdy mica]

yeah about that

about that many seconds

[fossil fable]

ah

at last

the first model i think is better than openai's flagship

took like 5 years but whatever

[plucky sparrow]

we need non-nerfed gemini back 😭

[fossil fable]

g3p was close but i still preferred 5.1 at the time

[sturdy mica]

what the hell is this spacing and which model is better than openai's gpt 5.3 codex?

who's g3p?

[fossil fable]

for agentic models 5.3c is still king im talking about chatbots, useful questions about general knowledge, expressiveness, delivery, depth

4.6 tops

[sturdy mica]

i think for coding 4.6 still tops

and is ahead of 5.3c

in all cases

[fossil fable]

i think it's basically a tie

[sturdy mica]

we will see when gpt 5.3 releases

non codex

[fossil fable]

i mostly use ccli as a terminal agent, not a programming agent

this use-case is really underrated because 99% of people have never touched linux

[tender patio]

This is just not true

Ive had 0 fails with 4.6

[fossil fable]

Leaderboards mean literally nothing

[tender patio]

means something to me

[tender patio]

and?

[inner relic]

Is ai smarter

?

[fossil fable]

a year ago:

look at these benchmark scores! (utterly clueless)
6 months ago:
benchmarks are actually trash; crowdsourced polls like lmarena are generally more accurate at rating general use
2026:

https://tenor.com/view/chudbob-soyjak-spongebob-meme-spongebob-wojak-gif-11677224362360143388

[inner relic]

Advanced npc in video

They flanked me

[shrewd citrus]

that’s actually such a cool system

you can make a hide and seek game maybe

with those npcs

[inner relic]

Good idea, I will try them in a maze

[tender patio]

Thats just pathfinding no?

[sturdy mica]

old uses messing with new

[weak dagger]

#1343291835845578853

[tender patio]

i make rbx games

Working on this rn

so id say i know my fair share of rbx scripting stuff

[inner relic]

That's impressive

[tender patio]

https://cdn.discordapp.com/attachments/954434991549448202/1333598793681342509/togif.gif

[shrewd citrus]

how do you do the models

[tender patio]

Blender

But since the style is studs i make most in studio

[sturdy mica]

holy template game

[tender patio]

Not a template

[sturdy mica]

literally a template your building on

YES IT IS I HAVE SEEN IT

[tender patio]

Litteraly isnt

[sturdy mica]

THAT IS A TEMPLATE

[tender patio]

send it to me then

[fossil fable]

rblx will rob you of all your labour and effort

[rose sky]

I’m sorry, but what do you mean by, “How many seconds did that take”? Is it how many seconds took to generate the song or something else?

[honest verge]

Use mistral

For this

[tender patio]

ive already made money of rblx

[fossil fable]

holy trash

[fossil fable]

1% of the money you generated

[tender patio]

You arent even in any dev servers sooo....

[tender patio]

?

[sturdy mica]

how many cents have you made bro

[tender patio]

let me send screenshot 2 sec

[fossil fable]

it has some of the harshest royalty margins on the internet lmao

[sturdy mica]

i make games dude

your game is literally a cash grab

[tender patio]

really? i didnt know

[sturdy mica]

thats not a good thing that your making cash grab games

[tender patio]

i really do not care

[sturdy mica]

why are you bragging that your making slop

[fossil fable]

r$

lmao

[tender patio]

convert it to usd

you can devex

[sturdy mica]

like 50 bucks

[tender patio]

i wasnt

[sturdy mica]

you totally are

[tender patio]

sure

[tender patio]

when

[sturdy mica]

you dont care that your game is slop

your acknowledging that its slop and dont care

[tender patio]

yeah

1.4m rbx is 5,340.33usd for 2 weeks of work

[fossil fable]

Cash-grabbing is PERFECTLY valid bro

in a world and economy like THIS the hustle is sacred

The problem is that he's cash-grabbing for a 50 billion dollar corporation

Not for himself

[tender patio]

mhm

Nothing scummy about what im doing

just making games and kids play it

Good for you 😄

Ive also wanted to start on a passion project

[fossil fable]

but it generated $20,000

[tender patio]

Yeah

[sturdy mica]

what do you mean

[fossil fable]

[sturdy mica]

what does this mean

[tender patio]

rblx takes a huge %

i have ugc up for sale rblx takes 70%

[sturdy mica]

send me your profile

url

[tender patio]

Sure

[fossil fable]

rblx really needs to be destroyed it's comically evil

[tender patio]

https://www.roblox.com/users/5075711763/profile

[golden ocean]

im down

[tender patio]

Sometimes

[golden ocean]

to make rlbx games

https://tenor.com/view/cat-sunlight-follow-lie-groom-gif-5081640

[tender patio]

Compression you really just love to argue with people

Yeah

[brittle hollow]

https://tenor.com/view/rage-gif-18338667365126853250

[sturdy mica]

yes i suppose so

[tender patio]

https://tenor.com/view/gato-cat-rigby-gato-coqueto-rico-gif-4061476440942373994

[honest verge]

Then I'm going to do it

[scarlet spire]

They'll grow out of it, eventually.
(or whenever they don't, and end up sectioned at 3AM on a very quiet psych ward.)

Volition is a powerful mediator

[tender patio]

https://tenor.com/view/scarecrow-pam-true-this-post-above-pam-gif-12781760158064032282

[sturdy mica]

i just dont have anything to do right now

[scarlet spire]

That, that you won't grow out of

[tender patio]

mhm

[scarlet spire]

The seeds of tomorrow are being sown yesterday, ya kno.

[sturdy mica]

i dont know what that means

[scarlet spire]

Exactly.

[sturdy mica]

whatever man

[brittle hollow]

Is the argument in the room with us

[tender patio]

Just scroll up

[sturdy mica]

i dont even know what arguing was happening

[sturdy mica]

i think i was upset because some kid makes slop games on a kids platform

and everyones siding with him i im not sure whats happening here

[brittle hollow]

I don't see any arguments though, just passive aggression

[honest verge]

Noo my hotspot

[golden ocean]

https://cdn.discordapp.com/attachments/559403230438883351/1468572366748778734/attachment.gif

[sturdy mica]

that grass is literally brighter than the alpha minecraft grass

this game is so good for children's attention spans

exploiting children for money

[tender patio]

Its just grass

[tender patio]

How am i exploting them? I dont pressure them to buy the stuff and they get what they bought

its their choice

[sturdy mica]

yeah its their choice

the average player is 5

[tender patio]

I think youre just a little lost and trying to be mad to be mad

[fossil fable]

lmao

[scarlet spire]

For my photographs

[sturdy mica]

im assuming your game has random gamepasses that pop up randomly

and the alert when you open the escape menu right

[tender patio]

Nope

[sturdy mica]

that seems like the kind of game to have it

[tender patio]

Nope

[sturdy mica]

so im just gonna go with Yes

[tender patio]

I dont do that

[sturdy mica]

i know it does ive seen the kit

thats a UI kit

[tender patio]

Its not

[sturdy mica]

your using a UI kit and it has that

[tender patio]

the icons are

but the ui no

[brittle hollow]

Meat It's not that deep. You're in an AI server which is more "slop" than some random dude using "templates" to create "trashy" rblx games

[sturdy mica]

the UI is a kit yes it is

[tender patio]

Bro its a frame with a ui stroke and an icon

takes 3 seconds to make if you have 2 braincells which you dont

[left lodge]

Bro atleast edit it a little bit before pasting 💀

[tender patio]

Youre just coping

[sturdy mica]

whatever guys keep justifying exploitation of little kids on video games to make money

its really cool

[scarlet spire]

But do you? And do you have spinal nerve cells too?

[sturdy mica]

maybe youll all learn

and grow up a bit

[sturdy mica]

not the cope reaction image

[tender patio]

I sure do, ive got a lot more than him

[scarlet spire]

The public wants: developers
The public gets: projectors.

[tender patio]

This guy said "grow up"

Youre the child

[brittle hollow]

The only thing being exploited here are the parent's credit cards, im sure the kids are having the time of their lives playing hennings neon green grass rblx game

[tender patio]

The us isnt the world

[sturdy mica]

LOL we are not bringing up politics with this 😂

holy christ you guys are bad

[fossil fable]

you're missing the part that it's a REFERENCE

[scarlet spire]

Do you still remember what you said five minutes ago, without looking it up?

[fossil fable]

to how FRIED humanity is right now

[brittle hollow]

It's compression against the world 💔

[scarlet spire]

mechanically recovered animal protein

[tender patio]

[tender patio]

https://media.discordapp.net/attachments/919788930348499004/1083057183165530132/IMG_5730.gif

Whats wrong with that?

[fossil fable]

use a steal command

[brittle hollow]

True robloxian!

[sturdy mica]

you arent a real developer broski

[tender patio]

How come?

[fossil fable]

what

[tender patio]

Just because i actually make money and do something with my life doesnt mean you have to get mad because of it

[sturdy mica]

being a developer for money isnt being a developer

[scarlet spire]

Why do we even need AI to generate slop, if teenagers are most effective recruitees likewise.

[fossil fable]

"unrelated to this community" mods how is rbx related to this community

[tender patio]

Pay them with a mcd meal and theyre happy

[golden ocean]

because ai makes slop faster

[scarlet spire]

I concur.

[sturdy mica]

yes but i dont know why you say that with this reply

why are you replying to that

they are in discord servers because [random fact]

[tender patio]

https://tenor.com/view/spongebob-squarepants-patrick-star-patrick-jug-meme-gif-11214063207978068321

[fossil fable]

ah

right

rbx is pure slop therefore it fits the general theme

[brittle hollow]

https://tenor.com/view/spongebob-patrick-spongebob-patrick-smile-spongebob-patrick-spongebob-patrick-neptune-gif-15553908463759634649

[sturdy mica]

are you guys all in this discord server to talk to the ai models for free to make your slop games

[tender patio]

Yeah

[sturdy mica]

i know @tender patio is

yeah

[scarlet spire]

I was here to provide some front end fixes as a tampermonkey script but I never ended up doing so, as the interface DOM ID values kept changing.

So here we are.

[sturdy mica]

thats great humbledeer

[golden ocean]

to talk to my ai girlfriend sydney

[tiny jolt]

Agentic LMArena when?

[tidal grove]

Hello

[brittle hollow]

Theyre ai also

[tender patio]

https://tenor.com/view/pollop889-gif-3245691328781964997

[sturdy mica]

alright well im almost done with this api

[tender patio]

cool

[sturdy mica]

can i have like 50 dollars worth of r$

[tender patio]

No

[sturdy mica]

20 dollars

[brittle hollow]

can i have like 50 dollars worth of r$

[tender patio]

no

[sturdy mica]

4 dollars

[tender patio]

maybe later

[sturdy mica]

every single AI youtuber

GPT 5.3 CODEX IS INSANE..

CLAUDE is KING

[toxic verge]

I guess in the other hand only one way to learn is to experiment

[tender patio]

https://cdn.discordapp.com/attachments/1276918863975682079/1299435078153273435/attachment.gif

[surreal zephyr]

Except 4.6 sucks

And codex is great

🥀

[scarlet spire]

They... all did what you asked for?

"A low poly ..." or whatever you asked for

[surreal zephyr]

Compare how good it looks

I asked for good perf realistic 3d water

100%

LOL

[sturdy mica]

what the hell is this pfp dawg

[surreal zephyr]

Ai slop

[sturdy mica]

who logs onto discord and sets their profile picture as that

[surreal zephyr]

A bot

[sturdy mica]

or some random 30 year old

but that guys a bot

yeah

[sturdy mica]

is that really true

[surreal zephyr]

Nah those often set a cat, adolfh, or a tank, or a car

[surreal zephyr]

Yup

[sturdy mica]

can i get the prompt and try this

[surreal zephyr]

"low poly animated water, high performance optimised, for game. using threejs and other needed stuff. realistic colors"

[sturdy mica]

alright let me try it

opus 4.6 and kimi 2.5 thinking

opus 4.6 thinking

i dont even know why im using thinking

if any frontier model cant do this without thinking then its worthless

kimi's approach is better just looking at the thinking but then again opus's thinking is summarized

[sturdy mica]

video size couldnt be smaller mate

[surreal zephyr]

🥀

[sturdy mica]

why are you testing the game in like a 100x200 window

[surreal zephyr]

2000x10

🥀

[inner relic]

You talk too much

[surreal zephyr]

Stupidity is annoying

[sturdy mica]

@surreal zephyr your just like me

🥹

[surreal zephyr]

You're

[sturdy mica]

yes it is

[surreal zephyr]

🥀

[sturdy mica]

okay shut up chud

https://cdn.discordapp.com/attachments/1134158923704844332/1429197899329441822/attachment.gif

crazy

[tender patio]

pure ai

[sturdy mica]

kimi 2.5 thinking

[surreal zephyr]

Meh

[sturdy mica]

opus 4.6

[surreal zephyr]

Whats the buggy rock

[sturdy mica]

claude says that they are low poly rocks

"Low-poly rocks for scene context"

[surreal zephyr]

Truly a 4d rock

Opus js ahead of us

[golden ocean]

REAL

[surreal zephyr]

Opus knows 4d

[north obsidian]

-4.5d

[sturdy mica]

4.6d

get it

@surreal zephyr asked claude to fix the rocks and they did

it did

i had to tell kimi to also fix it because of side by side chat and for some reason it made it all entirely better

and redid it all

[keen beacon]

why does claude stops generating and it says something went wrong

[untold siren]

Hello

[hushed gyro]

It's been 2 hours, and the errors have not been fixed.

[surreal zephyr]

Nice

[surreal zephyr]

It stuck in loop and timeout

[surreal zephyr]

Its been 2 days wym

[sturdy mica]

your right claude 4.6 is not good

kimi and claude have the same performance but claude overthinks

it's unreliable

ill keep using kimi until gpt 5.3 is out

unless that sucks too

[surreal zephyr]

Claude overthinks a lot. Its very creative. But overthinks

And hallucinates

[surreal zephyr]

5.3 codex is peak too

[sturdy mica]

?

[surreal zephyr]

Only ever model that nailed 3d tank tracks

With suspension

And animations

[sturdy mica]

no way codex made that

[surreal zephyr]

1st try

[sturdy mica]

wow

[surreal zephyr]

5.3 codex in cli

[sturdy mica]

ive never used codex 5.3

wish i could try it

but i dont have chatgpt plus or anything

and its not on anything like perplexity

[surreal zephyr]

Better than opus imo

[sturdy mica]

i wonder why openai wont release the api for it

i think they said somewhere that its too smart

and could find bugs and exploits in commonly misconfigured websites and such

but then why release it on the codex app????

https://tenor.com/view/weird-worm-eminem-eating-an-apple-gif-13073780328856045851

[surreal zephyr]

Gatekeeping

[hushed gyro]

Oh works for me 80% of the time

[surreal zephyr]

It crashes when it gets stuck in a loop

[sturdy mica]

no it doesnt

[hushed gyro]

So how to solve?

[sturdy mica]

it crashes when it uses all of its tokens

it has a hard limit

[surreal zephyr]

It has 6 mins timeout

[sturdy mica]

its not actually minutes

[surreal zephyr]

Use better model

[sturdy mica]

people keep saying that

its the tokens

not the time lol

[surreal zephyr]

Dunno

[hushed gyro]

NB Pro is the best model wym 😭

[sturdy mica]

its not setting a 6 minute timer every prompt lol

[surreal zephyr]

Idk thats what i read

Js repeating

🥀

[sturdy mica]

genius

[surreal zephyr]

Theres no way output token limit is so low tho

[twilit urchin]

Hello there! I am Mia from the cold north, Finland☺️

[golden ocean]

so true

[limber panther]

im ambatucam from the snow kingdom

[golden ocean]

buss

[tender patio]

what prompt did you give it?

[golden ocean]

ambatubus

I think

[wind escarp]

hey can i use an mcp with arena ?

[plucky sparrow]

you can't

[wind escarp]

ok, thanks for answering

[plucky sparrow]

np

[surreal zephyr]

i forgor i used cli

realistic tanks physics simulation

[proud bobcat]

GLM 5 and Qwen 3.5 spotted

We’re eating GOOD

[proud bobcat]

No way

That’s actually lit

[queen veldt]

Glm 5 will be op

As long as there's open source models we'll be fine from the ai bubble

[mortal coyote]

gemini image geekin out !???

ive scrolled half a dozen reels , this is still the same

[golden ocean]

gemini image is also scrolling half a dozen reels

[proud bobcat]

I still cannot believe OpenAI won’t make GPT-OSS 2

It’s like their biggest hit on the API

Nearly 8% of their market share comes from GPT-OSS 120B alone

Or maybe more I forgot

Yeah their entire market share is being held up by GPT-OSS 120B (7.8% of their market share) and GPT-5 Nano (5.6%)

As of February 2nd 2026

I’ll probably recalculate

[zealous path]

Hello

[shrewd citrus]

6 reels isn’t enough bro 💀

[bright junco]

Help, does anyone have a way to fix this? It’s been frozen like this for hours already.

[analog swan]

is opus 4.6 for free in Arena?

@daring rock

[keen beacon]

I send an image and a message in another chat that isn't frozen, and when I click send, I'm taken back to the chat that's frozen.

[proud bobcat]

Yes

All models are free

[analog swan]

aw sick

[proud bobcat]

Opus 4.6 has limits BUT

It’s quite lenient

Like 6 messages I think per 2 hours

[analog swan]

what in your opinion is the best for coding C#

game mods/cheats

[proud bobcat]

I tend to use Opus 4.6 for the coding, and Kimi K2.5 thinking for the code review and bug reports

Goated combo

[analog swan]

alright bet

[hoary grove]

Guys, I have a website project, how do I upload it? Only pngs, etc. can be uploaded

[left lodge]

Use k2.5 thinking for frontend , its great on that part, it also supports image inputs so you can show how exactly you want the frontend to look like visually

[hoary grove]

Guys, I have a website project, how do I upload it? Only pngs, etc. can be uploaded

[echo dome]

bring back cloudflare since reCAPTCHA just getting annoying

i just did captcha in correct way and then "please try again"

[golden ocean]

fr

[hushed gyro]

@echo aurora can you please tell the devs of your company that there is a serious issue that needs to be fixed. the nano banana errors.

it's been 2 days and there is no genuine addressing.

[surreal zephyr]

Opus bad

[golden ocean]

they're working on the new nano pineapple model

[hushed gyro]

wym

it gotta be cap

[echo aurora]

Good morning/afternoon/evening everyone

[echo aurora]

I'll be addressing these reports in the bugs threads.

[surreal zephyr]

Hi

Why opus 4.6 so bad

😔

[echo aurora]

I have a lot of pings to get through so apologies in advance if it takes me awhile to get to your question/issue. I really appreciate the patience!

[echo aurora]

Big disagree

[surreal zephyr]

It cant even reach leaderboard lol

[golden ocean]

it's nowhere near pineapple-4.6

[surreal zephyr]

This too

Loses against kiwi 2.5 and sonnet 4.0

[echo aurora]

wdym it's currently ranked #1 on multiple leaderboards?

[surreal zephyr]

Nonthinking

[spare rune]

There’s a psuedo effect or it’s not but someone said generate exactly when it becomes a new minute

[echo aurora]

Big agree

[spare rune]

It has worked for me

So far

[frosty shuttle]

💀

[spare rune]

Omg

From what I can see

That’s a 90% increase

[surreal zephyr]

sorry but that looks more like 2000% increase

[spare rune]

Mb

I understand estimated Claude opus 4.6

[obsidian cargo]

Claude Opus 4.6 believed me about hytale and bubsy but is thoroughly convinced I made up DK bananza and am gaslighting it into thinking it's real

[hollow imp]

@echo aurora

https://www.theverge.com/tech/875309/discord-age-verification-global-roll-out

If you can't read: https://archive.is/Xjvi6

the ending of this article is killing me bro how,,

[crude snow]

hi

[echo dome]

pineapple can you return cloudflare

recaptcha got worse

[north obsidian]

Good morning

[echo dome]

it's Extra annoying when you click the literal correct images and it says "try again" because the car/crosswalk/whatever was small and in the background

[obsidian cargo]

Did you mean to reply to someone else??

[surreal zephyr]

Opus

4.6

Believed

= happens always opus dum

[hushed gyro]

what is up with this pineapple thingy? An inside joke or smth

[hushed gyro]

what so at XX:XX:00

[burnt solstice]

Hello

[hushed gyro]

now both models???!!!!

@echo aurora I think this NB Pro issue has been happening for so long I think that you should tell the devs that this is an URGENT issue, or else chat will be on fire tonight...

[echo aurora]

Yeah shared this in #ai-news . I haven't yet fully wrapped my head around this change, but if there is something that fundamentally effects this server, I'll keep you all updated if there is something impactful.

[echo aurora]

I'm going to share a longer message about model errors in #1417174113092374689 in a bit. Crafting the message now so keep an eye out there.

[hollow imp]

Fundamental effects to everyone because of this

https://www.malwarebytes.com/blog/news/2025/10/discord-warns-users-after-data-stolen-in-third-party-breach

[surreal zephyr]

No way

Im quitting then

Moving to GG

Gadugadu

[hollow imp]

Steam chat?

[woven terrace]

hey guys

[chilly osprey]

hey

[woven terrace]

quick question

My impression is that once a model is released, they usually dont change ranking in the scores until new models come?
So unless gemini 3.5 or some other model appears soon, it is likely to take a while for Claude to lose pos1 rank?

[abstract tundra]

will seedream 5.0 drop on arena today?

[woven terrace]

I am speaking about text leaderboard, style control off

[azure nexus]

hi

[pulsar knot]

hi

[oblique sentinel]

I mean they update regularly but the 14 pt gap between opus and gemini won’t change unless a new model comes

but opus 4.6 thinking still needs to be added

[woven terrace]

Possible entrants would be grok 4.2?
what about this Gemini 3 Pro GA that I hear rumours about?

[oblique sentinel]

Yes 4.20 may release by end of month and same with 3 pro GA

[woven terrace]

so opus keeping the nr1 spot for this month is not a lock?

[echo dome]

help

i keep getting "please try again" when trying to verify in correct way

like i picked all fire hydrant and i still get "try again"
its so annoying

same with cars

fix the captcha or just replace captcha with other captcha that are not annoying

or just use hCaptcha

it's privacy-focused alternative

[scenic spear]

Can I unarchive arena chays

[drifting crow]

Sounds like something a robot would say

[echo dome]

i was trying to say that hCaptcha should be better

[drifting crow]

Im sorry you didn't click all the hydrants so I can't be sure I'm talking to a human right now

[echo dome]

hold on let me click all the hydrants so you will see

[drifting crow]

I think they fail them on purpose to increase the amount of model training they get for free

[proud bobcat]

right now Qwen 3.5, GLM 5, and DeepSeek V4 are the biggest contenders to tip the scales

[drifting crow]

They could easily detect ur human from the mouse movement anyway

[proud bobcat]

GLM 5 will be a MASSIVE contender because they already had nearly SOTA scores on just 355B parameters, and the new model is going to be 755B (from logs found on github)

so it may just be double the scores it already has

or double the gains at the very least

[echo dome]

here's proof that i'm not robot

left is i clicked all fire hydrant
right is "please try again" and goes to same one as left

[drifting crow]

Does top left have a pokemon hedge

[echo dome]

wait is it me or it looks like pokemon shaped grass?

wait no it doesn't

[wheat thorn]

i was having the same issue but it fixed itself

[echo dome]

for me its not fixed

idk why

[weak sluice]

aw man!

[robust sluice]

it's weird how turn off adblock can fix this problem (Im using Brave)

[echo dome]

well

[quartz pike]

la

[wicked sage]

hi

[grave plaza]

i just love Kimi k2.5

[echo dome]

i made feedback about recaptcha

https://discord.com/channels/1340554757349179412/1470476115129663705

i found the request.js in arena.ai
i sent it to google gemini and he said this

[echo aurora]

For those that are having issues with our captcha system I'd encourage you to review this message: #1451574502369656842 message

[echo dome]

maybe use hcaptcha?

[echo aurora]

Team is review options for changes to the current system.

[echo dome]

wait so they reviewing options for changes?

[echo aurora]

We're always going to be open to alternatives if those alternatives are better.

[echo dome]

that's good
i hope i will see other captchas after reviewing

recaptcha? more like Tracking

[echo dome]

also i was talking about recaptcha
since recaptcha v3 is now tracking your ip address

but since you said that
you should pick privacy-focused alternatives

[echo aurora]

Agreed, I'm excited to see what the adjustments look like. Would note changes like this aren't the kind of changes we'd put out an update for. When analyzing these systems it's important we're looking at it through real-world usage.

[echo dome]

hCaptcha and Cloudflare Turnstile are better tbh

but alright

[exotic bough]

hey

why we can only one video now ?

[surreal zephyr]

good change, less slop

[glossy umbra]

5.3 codex released via api?

[exotic bough]

what is slop

it do better videos?

[surreal zephyr]

noooooooo

🙏

[grand cliff]

Note to self

Don't refresh the Site when its generating

It might break the AI and make it generate endlessly without stopping, unable to continue

[burnt sinew]

No it will work just refresh it in a bit

It just will only show when its done and after you reload it

[grand cliff]

I already tried

Still is as of now.

[verbal nimbus]

What browser are you using?

Btw I've encountered the problem too

Usually VPN related

[echo dome]

Microsoft Edge without touching settings
also using zapret dpi (not vpn)

[golden ocean]

https://cdn.discordapp.com/attachments/1276918863975682079/1299435078153273435/attachment.gif

[echo dome]

it only works with traffic lights

[verbal nimbus]

Hmm maybe a VPN will actually help in your case

• private browsing

[echo dome]

uhm reCAPTCHA uses several methods to detect both VPNs and private/incognito browsing

[verbal nimbus]

I had the same issue yesterday, had to switch VPN + browser

[verbal nimbus]

Yeah, but you're in Russia right?

[echo dome]

yes

[verbal nimbus]

Maybe it just doesn't like the IP

Like sometimes nothing can fix it except changing the VPN server

[echo dome]

my ip is good and from my country how is he doesn't like the IP

[echo dome]

it can provide temporary relief from reCAPTCHA loops, but it is not a permanent fix and can sometimes make the problem worse if done too frequently

[verbal nimbus]

Idrk what's going on, but that's what seems to work

[verbal nimbus]

Yeah this has been a problem for a long time

[echo dome]

so at this point I'm waiting for new captcha system like hCaptcha or other captcha

[verbal nimbus]

The best way to replicate the problem is just to connect with Tor

[echo dome]

idk why they picked recaptcha instead of other privacy-focused alternatives

[verbal nimbus]

Maybe because people were misusing the site and making private APIs

[echo dome]

maybe

but at least i got recaptcha js code

[left lodge]

Rate limits exist you would still get the same amount of queries

[weak nova]

persistence good image model (n 1/1) just got it in battle, 4096x4096 blurry /smudged details up close but really good interpretation of the prompt

[echo dome]

recaptcha is also unfriendly to extensions

simple: don't use grok
it's problematic

[left lodge]

There was some damn word play lol

[verbal nimbus]

Sometimes you don't hit a rate limit but it asks you to sign in, not sure why

[left lodge]

Cause Its a completely diff system

[cunning hill]

When can i generate videos??

[scarlet spire]

Morning, noon, evening and night!

[cunning hill]

Where??

[verbal nimbus]

Mornooninght

[cunning hill]

i want generate video ??

[verbal nimbus]

I haven't generated a video in months, not sure 🤷‍♂️

[grand cliff]

Still 'Generating' it says

[verbal nimbus]

Watching AI videos gives me headaches lol

[echo dome]

and ofc reCAPTCHA is objectively insecure and illegal

(i think)

[wheat thorn]

nooo

[verbal nimbus]

Does it work on mobile

If so I wonder what happens if you enable mobile mode from Dev console in private tab, I think it changes user agent

[wheat thorn]

i will try

[echo dome]

i tested on mobile and i didn't see this

(recaptcha popup)

[verbal nimbus]

https://developer.chrome.com/docs/devtools/device-mode

Firefox can do this too

If you're desperate and wanna try

I think captcha takes a lot of factors into account, including device type; they're more lax on mobile

There's ways captcha can tell though, so not sure if it'll work. Unless you go full mobile emulator and use BlueStacks or smth

[woven harness]

thought this was just my problem lol

[grand cliff]

Yup....

I just checked again

Still generating

Stop generating button + Edit button fr

[shrewd citrus]

It will generate

if it’s text give it 1 minute

if it’s an image give it 2 minutes

if it’s code give it 5 minutes

and if it’s a video give it 10 minutes

after those times you can refresh the page and the output will be there

[grand cliff]

Text

[vivid sandal]

Is there a way to fix generations that get stuck???

Like I have a prompt, i liked what it was generating, but then I accidentally pressed home and it went to home page, and I had to go back but now it's stuck...

Oh, it looks like people are having the same problem as me

[echo hawk]

Well, it looks like they just removed the normal 1K model of Nano Banana Pro (gemini-3-pro-image-preview)

<@&1349916362595635286>

[grand cliff]

Nope. Didn't work.

Kept it up for 5 minutes.

[shrewd citrus]

what’s the prompt?

[grand cliff]

It's uh....

Well

I put the idea of Tokito as a prodigy, and the Demon slayer mark into the Roman Empire/Era

It was so good too.

[shrewd citrus]

what did you mean by 1st one please

Are you using battle mode or side by side chat?

[grand cliff]

4o gives me these options

It gives me ideas

I can choose 1st, 2nd, 3rd, or 4th.

No 4th here

Or I just shift direction

So I picked '1st one'.

[shrewd citrus]

oh i see

is it still stuck on generating or does it say error @grand cliff

[grand cliff]

Still stuck

Till this hour

[shrewd citrus]

hmm maybe log out and log back in

[grand cliff]

Almost an hour

[shrewd citrus]

log out first and close the website

[grand cliff]

Did just that. It got stuck

[shrewd citrus]

that sucks

[grand cliff]

Verily

[shrewd citrus]

sorry but I can’t think of any other solutions 😔

[burnt sinew]

Try again

[grand cliff]

Too late. Didn't work

[burnt sinew]

What

Like does it still say generating

[grand cliff]

Yup

For the past hour

[burnt sinew]

Ctrl shift r?

[grand cliff]

Nope

[hollow snow]

Why is gpt 5.3 not in the leader board ?

[wind ember]

should it be?

[north obsidian]

Because it is awful

The best is mistral AI

[wind ember]

thank you

i agree

[hearty cloud]

Have they removed Nano Banana?

[slim gorge]

dang anthropic is really taking the top

ig we just gotta wait for google's response...

[tribal remnant]

Yeah

[surreal zephyr]

holy rigged

4.5>4.6 normal>4.6 thinking

💀

[slim gorge]

huh

[surreal zephyr]

opus 4.6 is awful lol

[slim gorge]

maybe from your experience... lmao

[surreal zephyr]

its gemini 3 tier hallucinating lmao

lol

[undone geyser]

what image models are upcoming?

[surreal zephyr]

paper banana

[slim gorge]

guess we gotta wait for gpt-5.3-codex to be on arena

i wonder when gpt 5.3 is coming out... like not codex

[surreal zephyr]

codex can do this btw

[shrewd citrus]

is it right tho

like there’s a chance it was trained on that book too

[surreal zephyr]

thats suspension physics simulation with tracks

in js

and html

[surreal zephyr]

opus fails miserably lol

codex was the first and only that didnt fail it

❤️‍🩹

[gentle pasture]

most obvious insider tradign here

[slim gorge]

i feel like gpt is smarter but opus/sonnet still generates better code but cant tackle complex stuff

[surreal zephyr]

opus is more creative

gpt is better at code actually

opus is better at front end

i wouldnt let opus touch backend though

[slim gorge]

my only wish is 1m tokens context window on codex

[surreal zephyr]

400k is not bad, and it does better for most of the window

opus has still only 200k for most cases

and it hallucinates after that

as well as costs more

[slim gorge]

in my experience after compaction it just kind of starts the task over from the beginning lol

[tired mantle]

how to fix this?

[slim gorge]

try again

xd

[tired mantle]

still

[surreal zephyr]

try again

xd

[quartz pike]

guys if seedream 4.5 was alr peak imagine seedream 5

[surreal zephyr]

imagine gpt 5.3 noncodex

[quartz pike]

i absolutelly love their ai models they are so cheap and good. aswell as their vid model

[surreal zephyr]

😩

[slim gorge]

i used to believe the leaderboard propaganda but now i realize gpt is still the best

[surreal zephyr]

yeah

theres no way opus thinking got on top

its paid sponsored ad 100%

opus thinking literally doesnt work

it couldnt get anywhere near top10

[slim gorge]

i dont think its THAT bad but yeah

[slim gorge]

💀

[surreal zephyr]

another alt of that bot

😭

[undone geyser]

cannot wait for the next nano banana model. I mean what you think in your opinion i can do with paper banana? make comics?

like idk, a picture of me in marvel style, starring in a comic about saving people in a bank robbery? why is it called paperbanana?

[quartz light]

bruh ur still showing the same misleading thing

[surreal zephyr]

opus could never

[quartz light]

does that use hinges

or constraints

?

[quartz light]

https://api.websim.com/blobs/019c3fca-69b4-758e-8fd3-efb6425e5e98.html

[surreal zephyr]

neither

custom based raycasts + welds

it was a port from js version it made

[surreal zephyr]

LOL FAILED IN 3 WAYS

[quartz light]

?

[surreal zephyr]

opus so bad 😭

[quartz light]

stfu

blocked

[surreal zephyr]

the track doesnt deform

lol

it doesnt bend around wheels, it just guesses

the track doesnt connect

and the suspension doesnt work

absolute fail

[quartz light]

nonthinking

[surreal zephyr]

still

it failed

[quartz light]

• the wheels do move

[surreal zephyr]

they dont move up down

with terrain

[quartz light]

u cherrypicked

[surreal zephyr]

they clip below terrain

😭

[quartz light]

ego

[surreal zephyr]

WYM CHERRPICKED

[quartz light]

ego

[surreal zephyr]

I LITERALLY SHOWN 1ST TRY

ONE SHOTTED IT

[plucky basalt]

🍿

[surreal zephyr]

hes ragebaiting atp

[quartz light]

he has been doing ts yesterday too

blocked him

he acts like im the ragebaiter

[surreal zephyr]

https://media.discordapp.net/attachments/945168127191285810/1127666475981877278/attachment-36.gif

[scenic spear]

Opus 4.6 in arena is how many context please?

[echo sinew]

Hello chat! Friendly reminder of the server's #rules : ✅ Treat others with Respect. Be kind, assume good intent from others, and keep disagreements respectful. It’s encouraged to share your disagreements, but only if it’s done in a respectful and productive way.

[surreal zephyr]

exacly!

[shrewd citrus]

67

[surreal zephyr]

iv been using 5.3 codex for a few hours, and i can confidently say its WAY better in practical use than any other model i used, including both opus and gemini

[whole swallow]

https://tenor.com/view/your-not-a-sigma-sorry-you-ain't-him-bro-is-not-him-gif-14261336311648900928

[surreal zephyr]

its the only one with good memory

no hallucinations

and follows prompts well

as well as good spatial reasoning

[whole swallow]

Gpt is known for following the promps at the letter

But with planning that is SO bad

[surreal zephyr]

if you prompt it good

[whole swallow]

Opus brainstorms with you

[surreal zephyr]

it makes WONDERS

[surreal zephyr]

yea

opus is creative with you

gpt does the math instead of guessing if you ask it to

[whole swallow]

Yeah? But i din’t want a slave, i want a confrontation

https://tenor.com/view/family-guy-whip-angry-dad-gif-17511996347660987928

[surreal zephyr]

gpt is way more professional

imo

opus is creative

thats it really

[whole swallow]

Yes

[surreal zephyr]

opus is best at creativity

but it sucks in other fields

[whole swallow]

Opus boss -> gpt codes

[surreal zephyr]

opus if you dont know what you want and you want model for u to decide

[whole swallow]

Exactly

[surreal zephyr]

gpt if you want model to keep bothering you with questions to precise what exacly you want

but then gpt NAILS it

[whole swallow]

Yes

What did you code bro

[surreal zephyr]

game

gta online- like

but in rblx

for fun

[whole swallow]

Can i see screenshot

[surreal zephyr]

this too

also jet and heli physics

ai cops, npcs

gta online like web system

missiles, flares

custom car and tank physics based on raycasts

thermal and night vision

ai turrets, sentries, cram's

airstrikes

war thunder like armor system

quickchat

drone system, reaper and suitcase drones

nukes with custom vfx

a bit more i forgor