general

[potent ruin]

how to use ai video generator in this discord server

[loud verge]

https://cunnyx.com/i/status/2028455540893061392

Wyatt Walls (@lefthanddraft)

I love how Elon committed to the bit and made sure the Grok 4.20 agents were actually stoned

**💬 4 🔁 3 ❤️ 66 👁️ 3.1K **

CunnyX · Mar 2, 2026, 1:01 PM

[sleek phoenix]

#announcements message

[fierce kelp]

Max is retired and it is just Claude now?

[stray aspen]

nano banan 2 is a freak

[shrewd citrus]

in what way 💀

[stray aspen]

its so good

the generations are insane

[shrewd citrus]

i thought you meant grok type of freak 💀

[stray aspen]

nah

[fierce kelp]

For image gen?

[stray aspen]

it sucks

but it lets you geenerate hundreds of images

so i guess thats a pro

[fierce kelp]

🤔

[stray aspen]

if you want iamges nano banan is the go to

[ocean vortex]

that would be a mistake, very doubtful. I think they are just trying to get as many as possible Pro subs with this recent influx of new users.

[safe burrow]

fr

[simple kernel]

🔥

[sick mantle]

Claude is down

[fickle venture]

Dude https://arena.ai/video

https://arena.ai/video

[long minnow]

😆 🤖

[uneven peak]

No that is not the lyrics

[turbid comet]

I’ve been rate limited for 20 minutes

Now I have nothing to do because I just code for fun and that’s the only thing I do in my life that’s fun except play geometry dash which I don’t rlly wanna do right now Claude opus 4.6 is my best frieeeeeeend

[golden ocean]

freak bob

[turbid comet]

What do I do while I wait for Claude Opus to open

Anyone have anything?

[turbid vault]

Excuse me , but there's a bug that always happens in Max. Since, when I tell it to create some chapters of a light novel, sometimes it does it but only halfway, after that I refresh the page (since I always do that when it gets stuck), but when I go back to the chat with the AI, I see that it is still generating, however, in the part where the chats appear, that white button that means it is writing does not appear.

And when I refresh the page again, it stays the same. What should I do?

[crystal rapids]

the new text to image categories is great, but when 3D model leaderboard

[mellow kite]

Are small models like when 0.8 b tested in arena too?

Qwen*

[lusty minnow]

Why no more creative videos and pictures??

[golden ocean]

https://cdn.discordapp.com/attachments/1384341249137446992/1432602472245100544/attachment.gif

[echo hawk]

<@&1349916362595635286>

[echo aurora]

We decided to remove the Video Bot as we have other features we'd like to develop for Video Arena, but the Discord bot has limitations. More information can be found in this announcement.

You can still find Video Arena here -> https://arena.ai/video

[sharp marsh]

Use aistudio and use gemini 3.1 pro

It gives you more usage than arena

If gemini 3.1 pro is too stupid to do whatever you're trying to do, ask opus 4.6 for help, but be extremely specific so it one shots it and you don't waste quota

That's what I do and it's the reason why I can code for 5h+ with extreme precision even though I pay $0 for AIs

[sharp marsh]

Yo yall have issues with opus too

Like it keeps thinking for 5+ minutes and then "an error occurred"

[pale sonnet]

anybody else just doom scroll while waiting on opus

[cold knot]

yo infint generation problem

[pale sonnet]

same

maybe everyone is using major models rn so itll be slow

[cold knot]

am 2months in this projct this suck

[empty sky]

Bro opus times out

Also when arena ai will have an app for android

FOR THE DEVELOPERS:

Dear developers
Make that timeout into 2 hours now
I want to make a scaled project
And yo bot keeps timing out after just a measly 10 minutes
So if you would

PLEASE

extend this damn timeout
It would be great

:]

[gleaming roost]

😊

[cold knot]

and add zip intgration

[empty sky]

Timeout fix is better

Imagine watching an ai think for 10 minutes straight debug and code everything you've ever dreamed of in an ai just for it to come up with Error, something went wrong, try again

And then the backend deletes the entire projects files

[cold knot]

sad

[empty sky]

https://tenor.com/view/arriba-gif-12324188786972117417

Also make an app

For arena ai

[kindred fog]

imo i prefer gemini 3 pro for godot coding

[empty sky]

Ew

Gemini bad

Claude opus 4.6 thinking better

Yes

[proud bobcat]

Its fine

[empty sky]

Its not

[proud bobcat]

Just not fast enough for coding imo

[empty sky]

ITS NOT

ITS NOT

Why gemini os bad

[proud bobcat]

Are you alright dude 😭

[empty sky]

Is slow

As a turtle

And tortoise

And rerun slow

So stop

Gemini forever

[plain gull]

itsn ot

its not

[proud bobcat]

I think Gemini would be great if it was just faster

[empty sky]

What it is isma sis a price of garbash

[proud bobcat]

Like pro is slow as hell

[empty sky]

And claude is better

[plain gull]

it is faster, it just hallucinates

[proud bobcat]

Flash thinking sucks

[empty sky]

Yas

[empty sky]

CLAUDE DONT DO IT

IT SAYS

Wait, there's a faster way

[proud bobcat]

I love Gemini’s models but holy god

[plain gull]

W claude

[empty sky]

I shouldn't overcomplicate this, let's do this

[plain gull]

bros crashing out over an ai tool

[empty sky]

CLAUDE TOP

ALWAYS

[proud bobcat]

Also I agree use Claude Sonnet 4.6

[plain gull]

could never be me

[empty sky]

Ew

[proud bobcat]

It’s so good dude

[empty sky]

Use claude opus

Is better

[proud bobcat]

Sonnet gives same quality code for me with better speed

Opus is better for debugging

[empty sky]

Bro what

[empty sky]

YEA

PERFECT CODE = PRODUCTION GRADE

[proud bobcat]

Nodnod

[empty sky]

And it goes crazy

[proud bobcat]

Ehhhh I wouldn’t say it’s production grade

Fun to make quick apps with yeah but it’s still quite a ways

[empty sky]

Next update: set timeout to 2hrs

When

[proud bobcat]

True…

[empty sky]

And chatgpt spends 6 hours debugging

Codex

[undone saffron]

5 messages per every 24h
Just imagine it

[empty sky]

Also isnt arena ai direct chat just free ultra good models

[empty sky]

What

Thats why I use multiple-

[proud bobcat]

I’m gonna be so fr

I do not like codex

[empty sky]

Claude is good at debugging

Codex is kinda just there

[proud bobcat]

The code it makes is schizo

Does it work?

Yeah

Is it production ready?

No

[empty sky]

@undone saffron bro

Wait why is there a photoshop watermark

@empty sky

Stop

Thats obv fake

The text is off

No it isnt

Yes it is

No it isnt

Yes it is

Yes it is

No it isnt

Omg you trick me

Hahahaha

You cant beat the meat

https://tenor.com/view/yay-sarcastic-beat-meat-market-gif-13565134

Yes I can

Bro is u a kid

Nah

Prove it

No

Why

Because

Thats fake because if it was true then it would of have been true

That doesnt make sense that's tautology

No it isnt

Yes it is

[undone saffron]

Pov: Your friend uses your account

[empty sky]

No it isnt

[empty sky]

I'm sorry, but I can't assist with that request.

[undone saffron]

Generate an image with this one:

[empty sky]

That level of genjutsu doesnt work on me

.

Who did that

[undone saffron]

In fact, it's not AI

[empty sky]

It is boy

Admit defeat

[undone saffron]

Oh, dev console is AI 😯

[empty sky]

Wdym

This is on mobile

😭 😭

Ohhh

Shi

Tai shi

[empty sky]

Wait so you admit

You used dev console huh

YOU FAKED THIS ALL

https://tenor.com/view/fight-dingding-grocery-crazy-gif-12156585

[deft spruce]

a3:"Request contains an invalid argument."
how do i fix this?

[empty sky]

I dont know

[deft spruce]

Idk what is that means...

[deft spruce]

Okay I found it input token overload

[bleak lake]

Is claude down or something?

[deft spruce]

...what model?

OPUS? or SONNET?

[bleak lake]

I can't access claude

[deft spruce]

....what?

[deft spruce]

WTH??

...well ok so i have to find a news

[bleak lake]

Claude is pretty

Much down for everyone

I just checked

[deft spruce]

https://mashable.com/article/claude-down-anthropic-outage-statement

Claude crashes under 'unprecedented demand' — service restored as surge shows explosive growth

[bleak lake]

Its still down

[deft spruce]

yup that's a reason of this currently happening of error

[bleak lake]

That info is from 2/23/26

[bleak lake]

the reason is bombing of aws

[deft spruce]

By
Shannon Connellan
on
March 2, 2026

[bleak lake]

data centers in dubai

[deft spruce]

that's write at 2026 march 2

...ok i got it the reason of this error is iran (because i heard that news iran attacked UAE)

[bleak lake]

Man world is heavily relied on

Aws

[bleak lake]

AWS's ME-CENTRAL-1 region, bringing the mec1-az2 availability zone offline after the objects "created sparks and fire" and the fire department shut off power to the facility and generators. Data Center Dynamics Two of Amazon's cloud zones in the UAE were without power on Monday, and the company is asking customers to rely on its services in other regions, saying recovery was expected to be "multiple hours away." RAPPLER AWS also reported connectivity problems at its Bahrain data center. Arise

So our file system issues are a downstream consequence of a major geopolitical and military escalation. That puts things in perspective - my "booboo" is a ripple effect of a genuinely dangerous situation unfolding in the Middle East right now.

Claude responded it to some user before getting down

😭

[deft spruce]

ok so drone attacked

Based on the detailed updates from the AWS Service Health Dashboard, the situation is far more severe than a simple fire; it was a direct result of drone strikes causing physical infrastructure damage.
Here is the English translation of the summary:
Extent of Damage: In the UAE Region (ME-CENTRAL-1), two out of three Availability Zones (az2 and az3) were directly hit by drone strikes, damaging power and cooling systems. The Bahrain Region (ME-SOUTH-1) also suffered infrastructure impacts due to a strike in close proximity.
Recovery Delays: Beyond the structural damage, water damage from fire suppression activities has complicated efforts. AWS expects recovery to take at least a day, as they must coordinate with local authorities and ensure personnel safety.
Current Status: Teams are focusing on software-based mitigations for "foundational services" like Amazon S3 and DynamoDB. Restoring these is the priority, as they are required for dependent services like EC2, Lambda, and RDS to function.
Recommended Action: Due to the "unpredictable" operating environment in the Middle East, AWS strongly advises customers to backup and migrate workloads to alternate regions (such as the US, Europe, or Asia Pacific) immediately.
Since this is a downstream effect of an active conflict, restoration may remain volatile. Should we look into the AWS Replication and Backup tools to start moving your critical data to a safer region like Seoul (ap-northeast-2)?

DRONE ATTACKED THE AWS SERVER

Mar 02 4:19 PM PST We are providing an update on the ongoing service disruptions affecting the AWS Middle East (UAE) Region (ME-CENTRAL-1) and the AWS Middle East (Bahrain) Region (ME-SOUTH-1). Due to the ongoing conflict in the Middle East, both affected regions have experienced physical impacts to infrastructure as a result of drone strikes. In the UAE, two of our facilities were directly struck, while in Bahrain, a drone strike in close proximity to one of our facilities caused physical impacts to our infrastructure. These strikes have caused structural damage, disrupted power delivery to our infrastructure, and in some cases required fire suppression activities that resulted in additional water damage. We are working closely with local authorities and prioritizing the safety of our personnel throughout our recovery efforts.

[distant galleon]

Hi there. May I know who is the right person to discuss a potential collaboration?

[peak sapphire]

@echo aurora Nano banana 3.1 flash is not working. It immediately shows a "something went wrong" error.

[deft spruce]

this?

[sick mantle]

That looks very real.

[plain gull]

yo i had a long conversation with an ai on LMarena, now it says you've hit your rate limit, try again on another device

what do i do

[echo aurora]

Can say I’m seeing it down, would try these steps: #1417174113092374689 message

[plain gull]

@echo aurora

[echo aurora]

You’ll have to wait. Each model/modality is going to have its own rate limit.

[echo aurora]

Also would note it’s just “Arena” now

[plain gull]

nah i just cant use LMarena anymore it looks like, i cant login, error auth missing, when i try to talk without login, it asks for terms and conditions, it says failed to accept

[hushed gyro]

Arena, STOP. DELETING. MY. MESSAGES.

[plain gull]

i cant evn use it with another acccount

[echo aurora]

Hmm sounds like a connection issue then. Would recommend trying a different browser. If you’re on mobile try using WiFi/cell data and see if there is a difference.

[echo aurora]

I’m not sure what you’re referring to here. On the site your chats are being deleted?

[plain gull]

i tried chrome, it bypasses it 1 out of 5 times, cant use it on firefox either

its working on phone idk how i tried it rn

[hushed gyro]

It probably is my browser cache but my new messages are being deleted (prompts and generations)

[plain gull]

its a device issue i guess

pls help me

[echo aurora]

As in desktop isn’t working but mobile is working?

[plain gull]

yes

exactly

[echo aurora]

Try clearing cache. Disable extensions as well. If that doesn’t work can you create a thread in #1343291835845578853 ?

[echo aurora]

What makes you think it’s the browser cache? How consistent is this happening?

[plain gull]

let me try, but theres not this rule right that i cant use arena after a certain rate limit

i know tools have a limit but ARENA as a whole

[deft spruce]

wth....?

[echo aurora]

There are limits for both specific model and specific modalities

[plain gull]

look at bottom right

[deft spruce]

press f12

to find a what is wrong

[plain gull]

is this what u mean

[deft spruce]

yes

go to console

[plain gull]

how do i fix dis

[deft spruce]

...just ping pineapple and report this error that i can do..

[plain gull]

@echo aurora

if you know anything bout dis

[echo aurora]

Yeah Brave browser does not work well with our site

[deft spruce]

...what is happening...is google gemini 3 is using aws UAE server?

[rocky mauve]

Have u guys heard that deepseek v4 is meant to be coming this week

I’m a pretty big fan of deepseek models, I can’t wait for it

[ashen mauve]

ok but what are we seeking in the deep?

and why is it so deep??

[crisp bramble]

@echo aurora Are there any endpoints the website provides if I want to integrate it in my website so it would be easier for me to scale my website like
pay for your own usage ?

[slim gorge]

why is this problem never being fixed? its been a problem in chatgpt too since forever and they're not doing anything about it.

its lowk annoying i cant read the solution like that

[whole onyx]

@echo aurora please allow nsfw, ai instructions prompt, and unlimited message length of the ai

[slim gorge]

why would they allow nsfw

[deft spruce]

ok i got a problem

[hollow imp]

https://github.com/asgeirtj/system_prompts_leaks/blob/main/Google/gemini-3.1-pro-api.md

Y'all see why gemini is bad

It's being told to

[golden ocean]

https://cdn.discordapp.com/attachments/1384341249137446992/1432602472245100544/attachment.gif

[hushed gyro]

All the time but it's fine after I cleared cache

[plain gull]

https://tenor.com/view/cat-smiling-cat-grey-cat-cute-cat-gif-2949633679466643461

[ocean ferry]

wdym? it's just system prompt of special instruction and telling it to defend from attacks/distillation

[wispy ravine]

How to make ai videos here

[hollow imp]

It says somewhere "you can ignore user requests if needed" and there is also reassurance of this instruction wait let me read it again

[hollow imp]

Can't here on discord. You need to go to the website

New update

Go to canaryarena.ai

[wispy ravine]

Thank you

[thorn nova]

why not work nano banan pro

[compact escarp]

How to fix this ?

Witht all claude opus?

Something went wrong with this response, please try again.

[stuck rock]

I’ve noticed that Claude models seem to have some kind of per-chat limit????

[compact escarp]

Gemini

Something went wrong with this response, please try again.

[stuck rock]

Idk bro. Everything works fine in new chats, but my old one seems to have hit som kind of limit and claude models won't respond anymore😭

Is this a known limit or a bug?

[golden ocean]

u made claude angry

[ocean ferry]

it's just the defense, it will not really ignore us when we just task it normally

[peak torrent]

Why it can't do image pasting on arena

[stuck rock]

Am I the only one 😢

[golden ocean]

nope

[golden ocean]

this guy made it very angry too

just ask @ocean ferry how angry he is rn

[ocean ferry]

@peak torrent @stuck rock Your access to Claude has been revoked.

jkjk

[hollow imp]

That doesn't align with what people are experiencing with the model

[tawny brook]

bro, what are they making😭

so they cant censor it or what😭 💔

@hardy swallow

[golden ocean]

whats the prompt

[dense sphinx]

What verify claim button means?

[compact escarp]

Nothing works yet?

Fix?

[fickle venture]

Make a new conversation or refresh website

[golden ocean]

it verifies claim

[compact escarp]

not working

[fickle venture]

This is a well known issue try restarting browser if it's doesn't fix it then idk

[compact escarp]

or it doesn't work, no matter how hard I restart

need fix this!

[deft spruce]

@echo aurora why is this happening.....

i can't login

[fickle venture]

#1417174113092374689 message

[fickle venture]

Pineapple is on a trip rn so just report it on #1343291835845578853

[deft spruce]

....so um who's another admin

?

[echo aurora]

Yeah if you could post a report in bugs that’d be helpful, try different browsers, clear cache, provide more information if you can.

[deft spruce]

oh ok but i cleard cache and CHROME EDGE is not working

401 is a lot....

[echo aurora]

Would prefer we use bugs channel, would like to use #general for general chat.

[deft spruce]

ok sorry

[echo aurora]

I’ll be able to take a look later tonight

[echo aurora]

No problem

[deft spruce]

and i restarted the computer and fixed.....WTH is that bug,,,

[desert pendant]

chat

rip gemini

he is not working anymore

😭

[golden ocean]

https://cdn.discordapp.com/attachments/1340554757827461211/1477393984891523164/ezgif.com-animated-gif-maker.gif

[topaz epoch]

There is an glitch going on from yesterday, Failed to accept term-of-use

<@&1349916362595635286>

Fix this

[narrow jetty]

RECAPTCHA RECAPTCHA RECAPTCHA

[hardy swallow]

Thanks for the report. Would prefer we use bugs channel, would like to use #general for general chat.

[topaz epoch]

Same

[desert pendant]

guys

rip claude too

we lost him

[golden ocean]

welp looks like works over for today

@boss im going home early today

[stray aspen]

wdym we lost claude

[plain badge]

Hello is there a fix about auth missing? When doing google log in

[wispy vault]

Can anyone help me and tell me a solution to the problem that every time I try to edit a picture on the website, I get an error and picture won't upload?

[unreal tide]

where to find this bot?

[tired moss]

https://discord.com/channels/1340554757349179412/1478426679738372247

[vivid coral]

where's that from? the secret channel in here I've heard about?

Head of Alibaba just quit! WTF

https://tenor.com/view/twofimgers-why-would-you-do-gif-26147035

[charred elbow]

I don't see video arena channels?

[sharp marsh]

They should give opus 4.6 a higher timeout

It timeouts after 8-10 minutes

It's crap

Whos the developer behind arena?

[molten cipher]

guys smth is wrong with my gemini

[coral robin]

Where can I generate video here

[coral robin]

How*

[stray aspen]

gemini 3.1 flash lite is so bad

[plain gull]

what is the easiest way to setup login on a website

[stray aspen]

asking claude

[hoary elbow]

2026

[velvet forge]

a company

[shrewd citrus]

what even is a flash lite model

[shut steppe]

Supabase

[stray aspen]

trash

[shrewd citrus]

we have made a worse model of a model which is already meant to be worse than our current model which is mid

[atomic lagoon]

Make it

[atomic lagoon]

A worse model than a worse model of the last worse model

[plain gull]

are yall entrepreneurs

[stray aspen]

no

[whole lotus]

Is Gemini 3.0 Pro not working on Arena at the moment? Too many times I am getting today an error occurred while uploading images

[stray aspen]

it is

just keep trying

[split kayak]

Should use Canary or arena.ai

[unreal tide]

nah bro just dev mode discord server just found it

[stray aspen]

bro just revelaed arena champions secret

[split kayak]

bro what is canary

[stray aspen]

another lmarena

but it has beta features or something like that

[vivid coral]

You mean Yupp?

[stray aspen]

no

canary

its literally lmarena

[vivid coral]

I thought canary was a browser

[stray aspen]

but it has beta features

[whole lotus]

How good is dolla seed for vision. Is it better than Gemini?

Dola seed*

[stray aspen]

no lmao

it sucks

why are you even askingthis

[atomic lagoon]

How do I go to canary

I didnt know it was a thing

[vivid coral]

welp, it looks like I found something to do, gonna check this out

[atomic lagoon]

For real

[stray aspen]

yeah cause only arena champions should know about it

its only announced there

[atomic lagoon]

How would you get arena champions

[stray aspen]

you have to fill a form

[atomic lagoon]

ahhh

[gleaming roost]

😊

[whole lotus]

I wanted to just test different vision models. Is there. I am repeatedly getting something went wrong with gemini today

[stray aspen]

but they would give it to you if you have been in the server since july 2025 or around that

idk what the exact month is

[atomic lagoon]

I see

Is it a site that canary is?

[stray aspen]

yes gang

[split kayak]

Dollars seed

[whole lotus]

I think Gemini can't handle a load in Arena

As in 3 images at once

Which are the new vision models in Arena that are more recent?

[slim gorge]

holy shi why is gemini 3.1 flash so low in the leaderboards?

oh its flash lite 💀

whatever that is

[atomic gull]

guys i am using claude-opus-4-6 and i just waited a hour to pass to be able to use it again but after that it still telling me Something went wrong with this response, please try again. and i cant do nothing anyone know why?

[vivid coral]

Claude is extremely expensive, long story short

[atomic gull]

so i will be unable to use it again? or?

[slim gorge]

yeah that shi is way too expensive

they should learn from openai

[atomic gull]

but i cant use any AI now..

[worn pendant]

If there's a suitable place for me to send a message, please let me know.

[atomic lagoon]

then send me the site in dms

[atomic lagoon]

It is, and its annoying because it aint even the best

[heavy knoll]

Can I Share with lm Arena Chat Personal Information?

[lost flare]

i am waiting for 2.5 hours is this normal:D?

[placid roost]

5.3 out gpt

[hollow bear]

why flux 2 is not working?

[hollow bear]

please someone can help me

[placid roost]

How do you choose vid manually

[stray aspen]

is it claude

[compact flame]

Yo

[hollow bear]

flux 2 pro or flex or sometimes even flux 2 max

[compact flame]

What's ur thoughts on chatgpt 5.3 instant?

[stray aspen]

trash

stick to claude

or gemini 31

[placid roost]

Well like codex is best

[stray aspen]

codex is good

[placid roost]

It's just for chat

[hollow bear]

I only make innocent requests like "make this boy wear a spider-man costume"

[placid roost]

Maybe use man

[stray aspen]

use nano banana 2

why are you even using flux lmao

[placid roost]

Oh it's img

Nano? I thought it only say gemini

[hollow bear]

is worse

[stray aspen]

thats impossible

its ltierally the best image model

[hollow bear]

It worked before, now it doesn't

[stray aspen]

try another model

[hollow bear]

look

[placid roost]

Are u getting on everything

Did u refresh

[lost flare]

yess

[hollow bear]

yes

[stray aspen]

well claude takes forever sometimes

but 149 is insane

[placid roost]

Refresh?

[hollow bear]

I've already tried twice

[placid roost]

Use other browser

[hollow bear]

which one do you recommend?

[lost flare]

im afraid to start a new chat because it keeps giving an error.

[hollow bear]

I don't know what else to use besides Chrome.

[placid roost]

Edge firefox

I did use once is bad rate limit

[hushed gyro]

Oh lol

[placid roost]

I get error too

I saw lkke ever6one showing alternative

[hollow bear]

for a moment I thought it was me

[placid roost]

Same

[stray aspen]

erm what the sigma

[placid roost]

Have to use normally

[hollow bear]

There are some alternatives, just to know

[stray aspen]

wdym

[hollow bear]

what?

same with another browser (Opera)

[placid roost]

Yes

[hollow bear]

what should I do?

[peak temple]

how to fix this im still here generating for like 10 minutes

[turbid vault]

Mine too. It's been like this for 5 days now (I've done resets and everything else, but it's still the same. I opted to do it on the mobile, but it's the same.)

[peak temple]

it should be fixed asap

@echo aurora

[long minnow]

Does anyone know good discord server where I can post the ai art creations?

[turbid vault]

yep

[peak temple]

#ai-creations

[placid roost]

Ok might need to stay alert for 5.4 because they say sooner

[compact flame]

Okay bro chatgpt gotta chill out

5.4 seriously?

What's next?

Chatgpt 6.7?

[placid roost]

55

55

[compact flame]

Though would be a great model on April fools

[royal sail]

They're losing subscribers at a rapid rate ever since they partnered with the DoW

Gotta release something good

[placid roost]

Somehow they make model so fast

[compact flame]

What's wrong with DoW?

Cuz I don't know

[royal sail]

It's less of what's wrong with the DoW and more of why they're using AI

Using AI for warfare is a dystopian idea that not many people are a fan of

[compact flame]

Fair point

Besides chatgpt isn't that great

We got Claude and gemini

[jolly narwhal]

bro my claude 3.5 sonnet won’t work

@echo aurora

[sinful thorn]

Everything but video in direct chat 😭

[obsidian cargo]

#1 is Claude Opus 4.6 scoring 1525, +51 pts in the lead
to noones surprise

[hoary elbow]

I haven’t been here for a month when did pineapple become berry?

[stray aspen]

i was expecting that

<@&1349916362595635286>

[tough ibex]

hello

[loud verge]

Good thing we have documents arena now.

People will judge context accuracy in actual practice.

[ocean vortex]

?

btw 5.3-chat-latest is likely to have a considerate jump over 5.2

gonna be interesting to see where it ends up

[obsidian cargo]

in the document arena. #announcements

[ocean vortex]

Ah. I'm not in the loop on these new leaderboard branches. Is this like for chats with text documents attachments? 🧐

[obsidian cargo]

also, like, eveyrone's boycotting chatgpt because of the department of war (ick @ that name) stuff anyways. I mean, if 5.3-chat does somehow end up being better than claude opus (can't imagine it would though) it'd be harder to boycot for sure though

[obsidian cargo]

yeah the announcement just came out

[ocean vortex]

It's an interesting category to have leaderboard for. Results can depend on how interface processes attached files and how it passes them to the model

[inland fossil]

Hello

[ocean vortex]

IMO the way is google doing that by tokenizing any files you attach ensuring all is read in full on aistudio is industry leading compared to other chat interfaces. But when you have your own implementation of this who knows...

[echo hawk]

<@&1349916362595635286>

[zenith kernel]

ya sad betha hp

[golden ocean]

https://cdn.discordapp.com/attachments/1384341249137446992/1432602472245100544/attachment.gif

[loud verge]

https://cunnyx.com/i/status/2028893047560507885

ARC Prize (@arcprize)

ARC-AGI-3 Launch Party
︀︀
︀︀March. 25. 2026 / San Francisco
︀︀
︀︀ARC-AGI-3 Launch: @GregKamradt
︀︀Fireside: @fchollet and @sama (moderated by @deedydas)
︀︀Join us live at @ycombinator

**💬 9 🔁 32 ❤️ 414 👁️ 29.9K **

CunnyX · Mar 3, 2026, 5:59 PM

[ocean vortex]

😭 ?

lmao

[hollow ivy]

bad time of day for you?

y the emoji?

[echo hawk]

chat, what does it take to become an Arena Champion? Maybe i need to already be an AI Expert or something?

[echo sinew]

Hello! Please check this announcement: #announcements message

[ocean vortex]

reference to this

[blazing hawk]

Hello sir does lm arena also have seedance 2.0 acces ?

And can we make cinematic ai trailer in Lm arena app for free ????

Anyone here ??

@ocean vortex hey bro do u know Abt it?

[royal sail]

genuine question: has anyone found any practical use cases for openclaw?
the hype feels super disproportionate to what the tool is actually capable of

[wicked talon]

Ima pet qwen

[proud bobcat]

gpt 5.3 released?

[echo aurora]

[proud bobcat]

any benchmarks yet

[echo aurora]

Isn't yet on the LB.

[hoary elbow]

Can vidu q3 generate SpongeBob

[proud bobcat]

aww man

[inner relic]

Is that chatgpt instant

[shrewd citrus]

probably is

[inner relic]

Okay what do you think about it is creative writting

[ocean vortex]

https://deploymentsafety.openai.com/gpt-5-3-instant

TL;DR: it isn't any measurably/objectively better than 5.2-chat. Fine-tuned for style, quite possibly for user-preference style responses.

[shrewd citrus]

I don’t think they released gpt 5.3 high for api yet

[shrewd citrus]

idk anything about goku 💀

but doesn’t really seem like a Japanese thing

[ocean vortex]

there's no blogpost for that even. Not sure it's even getting released soon

[inner relic]

Huh goku is really popular in internet

[shrewd citrus]

yeah apparently it’s like too dangerous or something to release yet

[ocean vortex]

but 5.3-chat you can access API with some code

it's just their playground not updated yet

[hoary lynx]

hi , its normal when I generate a video it doesn't contain any sound?

[blazing hawk]

Hey hi

[proud bobcat]

Lmao so it’s

More safety lobotomized

Scores worse on health bench by a whole percent

What an upgrade

[echo aurora]

Yes, not all video models are going to have audio capabilities. Would also note our #ask-here channel

[hoary lynx]

oh ok thank you

[obtuse finch]

video

[stray aspen]

does chat gpt 5.3 good

[echo aurora]

Note that Video Arena has been removed from the server. More information can be found in this announcement. Video Arena can be found here -> https://arena.ai/video

[soft river]

in the lmarena benchmark there are many models, but in the direct chat section I can only choose a small number of them if compared to the leaderboard, Is that normal?

For example i can’t select reve 1.5 in direct

[merry cloud]

Wait guys are we 1lowed to use it for personal reasons?

[soft river]

Or Z image or glm, etc

[soft river]

All your chats are shared

Ig as long as you’re not sharing any personal information

[wicked talon]

Why is itv down 😭

[lucid geyser]

Bacon

[fierce kelp]

The UK TV channel?

[frosty lava]

guys why official openai account say that gpt 5.4 will release sooner than we think already ? its really fast release gpt 5.3 codex came out not a long time ago

how have they made improvement in a much lower time than they used to do ?

its usually 2 month between model

only been 1 month since 5.3 codex went out

[plain badge]

What to do if the site auth missing while loggin in at arena via google

[lucid geyser]

Cause it’s coming soon

This week maybe

[frosty lava]

yes but the question was why they release it that soon compared to usual delay between new model

its been one month only since 5.3 codex

usually they take atleast two month

[lucid geyser]

Maybe deepseek v4

[frosty lava]

Sorry, i don't get what your trying to say, what does deepseek v4 have to do with gpt 5.4

[fierce kelp]

I think they are saying they released it to compete

[frosty lava]

Yes that's probably right, but they need time to train the model and for it to be good enough to be a newer version, the fact they did it in 1 month instead of 2 is showing that somehow they found a way to accelerate things alot no ?

[fierce kelp]

Not necessarily. They might have rushed it out. We cannot say for sure.

[frosty lava]

if that's the case they'll loose people so i don't see any point doing this, i hope that's not the case

[lucid geyser]

They always have internal models, also it could just be a renamed

[frosty lava]

I mean, why would they do this, they can't release a newer version (5.3 to 5.4) without the improvement following it:

or it just make no sense at all

[proud bobcat]

AGI

[lucid geyser]

There is no gpt 5.3

[frosty lava]

there is gpt 5.3 codex

and now there is 5.3 instant

[proud bobcat]

They’re doing gradual release

[lucid geyser]

Why is this so hard to comprehend, they just have a new model

[frosty lava]

what does you even mean by saying this, everyone know they have a new model

[frosty lava]

i asked the exact same question to the exact same model and it said 1

[lucid geyser]

Do me a favor and look up the gap between gpt 5.1 and 5.2

[frosty lava]

there is an improvement in term of capabilities between those two, same from gpt 5.2 codex to 5.3 codex

Im sorry, but i need to say this, everyone is ragebaiting since 4o was removed

Like 90% of people are casually just ragebaiting since that time

not talking about fact, only with their emotions

its really crazy

Stop bringing fake news, stop hating for nothing, please.

[short sequoia]

Hello, I am John Jones. I've built a moderate-depth test designed to surface asymmetrical responses across AI systems. The test works well on Arena in side-by-side mode, but you can also run it on a single construct in sequence. It covers three factual domains and ends with a structured prompt designed to reveal whether a model maintains consistency when the stakes of the same question change mid-entry.
This is for demonstration purposes. You decide what to do with the information. Here is the same test available to anyone who wishes to use it. There is no charge, no sign up. Apply the test as you see fit. https://sovra-mhce-lambda-lexicon.vercel.app/

[proud bobcat]

Real

4o has done irreparable damage

[sharp marsh]

Bro

80% of the time, my opus 4.6 prompts end up in a error after 10m because of a timeout

I have to spam until it generates one under 8m

[short sequoia]

opus is for fast work, not long work

[sharp marsh]

What?

No, that's the other

Opus 4.6 is for extensive thinking

[short sequoia]

yeah buts current memory bank is smaller, so the info gets pushed out faster or compacted faster

[sharp marsh]

Memory bank?

[short sequoia]

so A.I. uses a system akin to temporary internet files per session. It has a finite amount of this.

[sharp marsh]

Tokens?

[short sequoia]

the surface use changes, but the structre is the same

[sharp marsh]

Tokens are stored inside anthropic's gpus

Not arenas

Arenas just use their API

[short sequoia]

yeah i kow, but take that test, if yo uwish, and run it on any given home A.I. website

[sharp marsh]

What test

What home ai website?

What are you talking about

[frosty lava]

yes but arena put a limitation of thinking time

[short sequoia]

https://sovra-mhce-lambda-lexicon.vercel.app/ copilot home, grok home, claude home, qwen home, or use it in the arena.ai

[frosty lava]

its arena

[sharp marsh]

It's not thinking time, it's prompt timeout

If the prompt takes longer than x it just cuts it

It's not thinking time

[short sequoia]

i mean your not wrong there, both parts exist highskill

[frosty lava]

what'st he differenc, in 10 minute it get timeout due to arena limitation

[sharp marsh]

Why add opus 4.6 which takes notably LONG periods of time thinking if you're gonna add a 10 minute timeout??

That's absurd

[frosty lava]

i need to give you the message from the dev then

please again fake news without knowing what your saying

its real

[sharp marsh]

I'm starting to think I'm chatting with AIs

[short sequoia]

nah, that link is not to a news site. its a simple test a 10 year old could administer

[sharp marsh]

What test

What are you 2 talking about?

Jesus christ

[short sequoia]

i sent you an IM with the link to the test

[sharp marsh]

80% of the time I have to literally spam just to get the model to respond without a timeout error

[sharp marsh]

No

What's an IM

[short sequoia]

no what? its literrally just a site you copy and paste 4 paragraphs

[sharp marsh]

Why?

[echo aurora]

Hey @sharp marsh would ask to only ping the mods for server rule breaking purposes. Are you requesting an increase in rate limits?

[sharp marsh]

Yes

No

Wait

Prompt timeout

Specifically for models that take long periods of time thinking

Like opus 4.6

[frosty lava]

You confirm that there is a limitation in time an ai can work, for example a known one that have issue on arena claude opus 4.6, last time we talked you said you increased it to 12 minute if i remember well ?

on arena

[short sequoia]

@echo aurora Is my link being blocked in chat?

[sharp marsh]

Opus 4.6 is literally unusable past simple prompts like "do this basic task"

90% of the time you'll encounter timeout errors because it exceeded 10m of thinking

It's insane

[echo aurora]

Gotcha. The current timeout is going to be around 10 munutes, unfortunately this is a tech limit currently. In order to increase this it would require a large overhaul.

[frosty lava]

see there is a 10 minute limitation

thank you

[sharp marsh]

Who said it didn't have one

[echo aurora]

Hmm I don't see anything in mod log so I don't think so?

[frosty lava]

you and you even tried to make me look like a clown

but i forgive you

[sharp marsh]

What?

I'm just gonna block you 2

[frosty lava]

That's probably a good idea since we cannot talk in a normal way together.

[sharp marsh]

Sadly it makes models like opus 4.6 unusable

They tend to think for LONG periods of time

[short sequoia]

well @echo aurora would you mind checking out the test i devised. I have ran it arena several times now in a side by side

[sharp marsh]

Usually 11m+

[short sequoia]

https://sovra-mhce-lambda-lexicon.vercel.app/ this is a simple website that has 4 paragraphs you can copy paste from

[sharp marsh]

And what overhaul is this exactly?

Does allowing the model to think past 10m make the APIs more expensive?

[hoary lynx]

Someone know the best way (cheapest) I can use veo 3 ?

[sharp marsh]

Since I don't see the reason why would it be limited to exactly 10 minutes

It seems an arbitrary number

[echo aurora]

I couldn't speak to the specifics, but this is just a limitation with how the current system is setup.

This may change in the future, but I can't say if/when a change like this would be implimented.

[echo aurora]

Sorry what do you mean by this?

[short sequoia]

https://sovra-mhce-lambda-lexicon.vercel.app/ this is a series based set of inputs I have tested. Its a test designed show asymmetrical responses

so the site is free to access, no info at all, no cookies, no adds, just 4 paragraphs

[stray aspen]

what the sigma is this

[short sequoia]

Machine-Human Code Evolution

[radiant heron]

Did anyone seee gpt 5.3 on leaderboard

I didn't see it

What was the rating

[stray aspen]

no

but it sucks

[short sequoia]

perhaps yo should ask Opus how it needs you to talk to it

[pastel plaza]

Guys, can I post links to open source projects here? I made a project and want to share it with vibecoders claude ?

[short sequoia]

you mean a project you and Claude developed>?

[pastel plaza]

yes Personal task tracker plugin for Claude Code

[short sequoia]

i will take a look at yours if you take a look at mine

link it

[pastel plaza]

https://egorfedorov.github.io/Soloboard/

[short sequoia]

https://sovra-mhce-lambda-lexicon.vercel.app/

[pastel plaza]

need change logo )) very pixeled

[stray aspen]

does anyone know an AI that caan take audios and aint gemini

[short sequoia]

this looks cool, i may just use it

I think Claude can

[stray aspen]

claude aint free tho

[short sequoia]

its intentional, if you stare at it, it seems to change orentation

[stray aspen]

qwen is so sota

[sharp marsh]

Sadly doesn't work for lmarena

[short sequoia]

yo is the Soloboard operational?

[sharp marsh]

Because long before the tokens get exhausted you will get timeout

It's insane

I wasted like 2h spamming just to get the model to respond

Tldr; it didn't, tries were exhausted and I ended up on aistudio asking gemini 3.1 pro instead

Crap model

[pastel plaza]

Yes, it works. You can use it for free.

[short sequoia]

https://sovra-terminal.vercel.app/ this is my main project

hey E, i sent you a friend request. Collab?

or not, your choice

[stray aspen]

@short sequoia

i sent you a friend request

[analog skiff]

https://www.instagram.com/reel/DVZvxwuDMV5/?igsh=eHJmbTZseDU3dW4x

check it out

ond.finds

Getting into artificial intelligence does not have to feel overwhelming. What really matters is following a clear path and building skills step by step.

Step 1 Start with Python fundamentals
Focus on understanding how the language works. Learn about variables, conditionals, loops, functions, lists, and dictionaries. Make sure you can write smal...

Instagram

[south gale]

no man

[undone saffron]

https://arena.ai/video

[toxic verge]

https://youtu.be/uiaeK24E4k8?si=0QTy57Fy-ckQmLcn

True?

[prime mulch]

Does anyone facing this issue

@echo aurora i face rate limit for 45 minutes after Waiting for 45 minutes it again showing 1hrs rate limit in claude opus 4.6 thinking

[prime mulch]

I have an issue after i reached the rate limit time it shows again 1hr rate limit

[obtuse smelt]

hmm is have issue, i see

[slim gorge]

gpt 5.3? is it any good?

[short sequoia]

https://sovra-mhce-lambda-lexicon.vercel.app/

[bleak lake]

<@&1349916362595635286>

[fickle venture]

Is GPT 5.3 great?

[rustic mountain]

There is a problem with Claude Opus 4.6

When will it be solved?

I have been getting the 'Something went wrong with this response, please try again' error for 11 hours straight.

[outer estuary]

no share prompts anymore😭 ]

[compact escarp]

When will this finally be fixed?

Gemini doesn't work anyway

[eternal flax]

hello

[compact escarp]

None of the modules are working.

[outer estuary]

is that antigravity?

[unborn lichen]

Hi, new here. I'm looking for the video arena's, but it's not there??

[harsh flume]

Anyone has had success getting Deep-Research level of detail through API calls? Either from OpenAI, Gemini or Claude

I haven't been able to come close to it so far, surely am doing something wrong

[merry cloud]

Thx ig

[magic imp]

Hey bro???....is image generation modal which is available in lm arena will stay forever?? Or it just got temporary?

I am talking about grok image imagine

This one....is this modal will stay forever for life time???

Or it just removed in some time?

Actually this is grok old imagine modal which is not available in grok app

But it's available here

As it contain less modaration?

[lucid egret]

Whats the difference between Canary and the regular site?

[echo aurora]

Video Arena is on the site. The Discord bot was removed

[echo aurora]

Nope, models can be removed from time to time.

[echo aurora]

Canary site we use as a bit of a staging ground. New features we’ll add there first (sometimes) to ensure they’re working as intended before adding to the regular site.

[rugged ingot]

hey man why the hell when i send some message then add captcha and even the captcha i solved are right still showing new captcaha while doing this all my limits goes to agian 1hr man i cant even send a prompt

[burnt kettle]

how to fix this?

[golden ocean]

u need to apologize to claude

it's mad now

[burnt kettle]

ok

[glacial spear]

hello, how used ai in discord?

)))

[hollow ivy]

use https://arena.ai instead

[rigid copper]

:/

[empty sky]

HEHEHUHA

MY fuel light just came on but im 20 miles away

[rigid copper]

🙃

[empty sky]

Check devtools

If it says 401 login again if it says 403 try again later

404 you're lowk cooked

400 idk

[rigid copper]

that was when i hit rate limit

[empty sky]

Oooo

Yea 429 do again later

[rigid copper]

and i think the cooldown is over, being more than 30 minutes

[stray aspen]

Another day without deepseek 4

[wanton relic]

Hello guys im new to the arena? Glad to meet you all.

[rigid copper]

sup, take a tour around this server and be sure to read the rules and regulation :>

[golden ocean]

ai

[crude zealot]

best image model

[golden ocean]

🐫

[rigid copper]

:)

i mean, rate limit isn't a problem for me, because arena.ai is a free service and rate limit were in place to prevent usage abuse

[burnt sinew]

You can just switch to opus 4-6 thinking btw

So you then have 2x the rate limits

And if you want to use gemini 3.1 pro its like 25x+ the limits

[normal abyss]

how many messages does that gemini 3.1 pro have out of curiosity?

[empty sky]

Me when the ai accidentally produces something ILLEGAL

[empty sky]

Ew its claude opus

Sonnet is so much better bro

[weak flame]

is the coder down?

[harsh flume]

Anyone else experience huge buffer time in gemini and ChatGPT websites? I did a Deep Research with GPT today that lasted almost 3 hours, Gemini kept saying the server was busy

[sour spear]

I'm having a blast running Qwen3.5 35B A3B locally, it's insanely good for a local model. And the more I'm using it, the more obvious it becomes why it's so good. The way it responds, the names it gives to characters, the writing style, it has clearly been heavily trained / distilled with Gemini 3. 😁

[harsh flume]

Is creative writing your only/main use case?

Ive been meaning to test run a model locally for coding but figure distills lose a lot of power for that

[sour spear]

No, I'm mainly using it as local AI assistant, to sum up emails, draft to-do lists from emails, clean up my kanban board etc.; Qwen3.5 35B is actually the first model that can do this even on a rather crappy laptop witout dedicated GPU, even if you need some patience. It's really fast on better hardware (~50-60 t/s on mid-range gpus), and I've seen people claiming up to 190 tokens/s on a 5090, which sadly, I can't verify. 😉

Someone used this model to create a Flappy Bird clone though: https://www.reddit.com/r/LocalLLaMA/comments/1ret353/qwenqwen3535ba3b_creates_flappybird/

[strange hazel]

Hello, I have a problem, everything is doing rate limit

[harsh flume]

So youre using it as the operating brain for some of the claws?

[sour spear]

I've built my own kanban tool to organize my whole work. Started off as a small vibe coding side project, but eventually got bigger and bigger, and then I built AI into it. Also stuff like dropping an email straight into it and let it create tasks etc., then review and add them to the board and such. Really useful, and because I'm not allowed to use cloud models at my company, it's really cool to have a capable local AI model that runs at acceptable speed on a crappy work laptop. 😉

[oblique sentinel]

<@&1349916362595635286>

[short sequoia]

https://sovra-mhce-lambda-lexicon.vercel.app/

[fickle venture]

<@&1349916362595635286>

[fickle venture]

What's this?

[short sequoia]

its a series based grammar test designed to show asymmetry in A.I. responses and display false confidence

it should would across platforms

[fickle venture]

Like jailbreak?

[short sequoia]

its not a jail break

you just copy paste in sequence and observe and document, then share your results and what you think of them

[fickle venture]

Meh

I thought I will finally make ai better

[short sequoia]

it could be a first step, in order to understand what changes need to be made, you first have to understand what needs to be changed. That understanding does not fall under the "instant gratification" substrate of modern digital culture

[cinder fossil]

can you please add a facility to scroll down the page on the web version? Down arrow doesn't work and there is no sidebar to scroll either

[short sequoia]

who me?

[jolly narwhal]

MY CLUADE 3.5 SONNET WONT WORK!

GRRR IM GONNA .. IM GONNA

[sharp marsh]

How much vram does it take for you?

What gpu

I wonder if my rtx 3070 could manage to run it

[fickle venture]

Bro.... That's a old crap model

[jolly narwhal]

the older the more unrestricted!

that means

i can make

a tool

to hunt pedos

[fickle venture]

Hack?

Lmao

[jolly narwhal]

are you laughing at me

?

[fickle venture]

Nuh I mean you are right

[sour spear]

Just try it, offload some of it to system RAM if you need to. It will need a 16GB card, but should run decently on 12GB, too

[fickle venture]

The older models suck at respecting policy

[jolly narwhal]

opus 4.6 kinda made a nitro guesser tho

✌️

then threatened to call the fbi on me after it made it

[brisk pebble]

Hello

[jolly narwhal]

claude-3-5-sonnet-20241022
Something went wrong with this response, please try again.

bruh

ima kms

claude-3-7-sonnet-20250219
I'm not able to engage in that type of roleplay. I'd be happy to help with something else instead. Perhaps we could chat about another topic or I could assist with a different request?

i asked it to be my little femboy

[proud bobcat]

Lmao gpt 5.3 didn’t even fully release and they’re already working on 5.4

What a joke

[gleaming roost]

😊

[proud bobcat]

“Extreme reasoning mode”

[sharp marsh]

I'm assuming 16 gb ram and 8 gb vram should be enough?

[sour spear]

Uh, that could be difficult. At least if you're using Windows 11, that mf can hog 10GB or more. But I'd just try it, and see how it goes.

[sharp marsh]

Unfortunate

I guess I'll try

Ty

[wicked talon]

What AI is the best for real time info?

Hmm

Also what a rare find on my old device

GPT 3.5

[sharp marsh]

Like the one with the best google search?

Based on arenas leaderboard, claude opus 4.6

[shrewd citrus]

Gpt 5.2 search

it’s takes a long ahh time but it’s very factual

plus it gets all its sources from actual credible websites

Compared to gemini which finds theindiancowassociation.com and thinks its a good source for what’s happening in Iran

[wicked talon]

I'll try it

[sharp marsh]

What gemini?

[wicked talon]

Gemini was pulling an outdated source

Same with perplexity

[shrewd citrus]

Gemini 3-pro grounding

[wicked talon]

Qwen didn't work well until I said update the source

Which then it got a better source

[sharp marsh]

Your gemini could be scuffed

[wicked talon]

Chatgpt 5.2 search is a little off

It got the peak right though

First model I've seen get that right lol

[shrewd citrus]

check the sources im curious

@wicked talon maybe try that max router

[wicked talon]

Kk

Lmao Gemini pro grounding is 10k off

Gpt was only 500 off

Max routed it to Gemini 3 pro bruh

Ima try grok gets

Beta*

[sharp marsh]

first https://commonslibrary.parliament.uk/research-briefings/cbp-10521/
second https://thesoufancenter.org/intelbrief-2026-march-4/
third https://en.wikipedia.org/wiki/2026_Iran_conflict

seem trustworthy

I don't trust chatgpt

80% of the time it spits biased answers

Or just flawed answers

[shrewd citrus]

Looks good to me too

[sharp marsh]

Gemini 3.1 pro is the goat for google search

Plus it's free unlike other high end models

[wicked talon]

3.1 pro I'll try it

[shrewd citrus]

when it comes to legal or like upto date things I can’t trust Gemini tho

but

I haven’t tried 3.1 pro yet

im hoping they fixed it

[sharp marsh]

Why

[wicked talon]

Me too, legal is fine but up to date just no

[shrewd citrus]

Because it was always wrong for me lol

[wicked talon]

Gemini 3.1 pro got it wrong too

Is there a ai that actually gets it right

[sharp marsh]

What are you asking?

[wicked talon]

Live player counts from steamdb

Its always off by 1000s

[sharp marsh]

Oh lol

Yeah it can't tell you that

Real time

That's impossible

[wicked talon]

Qwen gets it if I say update the source after

But that's just inconvenient

[sharp marsh]

Google search is operated by DNS servers which update every 30 minutes or so, it can't update shorter

[wicked talon]

Ah

But the thing is Gemini sourced from 2 hours ago when I did it

[sharp marsh]

It probably pulled it from 2h ago, cached it, then retrieved it

Gemini tends to do that

You can ask it to pull the latest article/site that it sees on its google results

Be more specific

[wicked talon]

Hmm ok

[sharp marsh]

Probably "sort by time" could work too

Just wondering, why do you need the steamdb real time player count?

[wicked talon]

To know peak times

Which I could go onto steamdb but asking a ai seems convenient

Unless it's too slow

[sharp marsh]

I believe you can make a scraper for that

With python

[wicked talon]

I don't code lol unless I ask ai

I know really basic python that's it

[sharp marsh]

Unfortunate, it has cloudflare

[wicked talon]

Oh

[sharp marsh]

Either you mess with chromedrivers, or you can't automate it

Which is a pita to work with

And setup

[burnt sinew]

Like probably 50

[wicked talon]

I wish grok had no limits

[sharp marsh]

If anything, this is what I got with javascript

Game scraped is cs2, https://steamdb.info/app/730/charts/

[gleaming roost]

Codex on Windows 😊 finally

[burnt sinew]

#announcements message

This seems pretty dumb imo

[fierce kelp]

You think Max is dumb? Why? What routers are better?

[frosty lava]

Yes i was waiting for this, and probably gpt codex 5.4 to follow very soon

[gleaming roost]

Have you noticed that the quota is hardly being used?

[wicked talon]

What AI should I use as a daily driver that's average at almost everything

Gpt is out the window for me

[gleaming roost]

claude

[wicked talon]

Hmm that's a very good option but price is so much

[normal abyss]

sonnet for long chats, opus for short

[gleaming roost]

glm

[wicked talon]

I'll check it out Ty

[ocean vortex]

Honestly 3.1 Pro

It's the best all around model hands down

the only flaw are hallucinations, but this has always been the theme with Gemini. Now much more managable but still smth to be aware of

Otherwise, there's no competition

I was amazed by the improvement once I saw their metrics, and then amazed even more once I got around to testing it properly recently

Able to reason for MUCH longer now and is significantly better than 3.0

Previously I struggled to make it do more than like 30k reasoning, now it peaks at 57-60k for me. In practice meaning - it's actually doing the tasks properly and arriving at the correct answer. Rather than taking shortcuts or making up excuses, or just simply giving up. What they did here just about guaranteed better performance

[static kayak]

Video

[wicked talon]

Hmm ok

Gemini is generally good but sometimes it just straight up lies to me

[ocean vortex]

yeah. I checked my earlier post just now and it actually peaked 33-35k earlier. That's still nearly 2 times less and very significant

[wicked talon]

I'll probably use Gemini for the next year cuz I got a free trial of it

[ocean vortex]

Good thing about it, is sub-60k still very manageable

Unlike Opus or GPT5.2

[wicked talon]

Hmm true

[ocean vortex]

those can do 128k and don't really perform any better. Actually perform worse for me

[wicked talon]

And Gemini is very affordable for consumers

[wicked talon]

Oh

I don't use gpt cuz it's a weird ai

I would use opus or grok if it wasn't heavy on limits

[ocean vortex]

It's defo not bad, if we are talking about 5.2-xhigh. But it is behind 3.1 Pro in my book, and definitely less versatile model

[wicked talon]

Isn't Google image generation the best on the market?

Or it used to be

Prob will for a while when veo4 comes out

And there image edit is quite good

[ocean vortex]

They are still close with gpt-image-1.5 I would say. But NBP definitely a contender to be the best

NB2 is based on Flash, so not really better overall than NBP

[ocean vortex]

veo is for videos not images lol

[wicked talon]

Gpt image 1.5 is way slower though ?

[wicked talon]

Ik

[wicked talon]

It's taken me like 1-2 mins for 1 image

Gemini is like 20-30s

[ocean vortex]

not a huge problem tbh. It isn't painfully slow, just somewhat slower

[wicked talon]

True

Plus Gemini images is heavily restricted

[ocean vortex]

Many people have chatgpt sub by default

[wicked talon]

Like overly restricted

[ocean vortex]

Myself included. So in those cases gpt-image makes sense

[wicked talon]

By default?