general

[odd kindle]

THANKS

[fiery gull]

Just 200k

[obsidian gazelle]

Hi

[stoic canopy]

hi

[tropic bluff]

Hi

[echo aurora]

hello

[golden ocean]

[coral hound]

#video-arena-3

[heavy sierra]

I'm looking to automate a YouTube channel telling horror stories.

[brave orbit]

n8n i guess

[molten tendon]

@heavy sierra hello

[slim tulip]

hellohello

[green field]

hi

[sinful ibex]

Guys i have a ques

[chrome halo]

guys are there any AI tools that upscale images in bult as per required pixels?

[quasi atlas]

@vale pawn @fervent marlin @true thunder You might check on #1397655624103493813 to learn how to use the bot and #video-arena-1 #video-arena-2 #video-arena-3 for your creations.

[spice nymph]

Code plz

[shadow jewel]

is there any way of sending 20k lines of code to LMArena.... 🙏

[hazy kernel]

You can send them as file

[shadow jewel]

it says it has to be a png, jpg file and etc

[hazy kernel]

*you cant

[shadow jewel]

so u cant really send .txt files

[hazy kernel]

Hah yeh I just tried

[shadow jewel]

but any ideas?

[hazy kernel]

It cant

[astral compass]

How generate video on LMArena?

[sharp mirage]

Yo anyone playing Minecraft bedrock?

[pulsar saffron]

not possible

[rough kindle]

@viscid cloak

[hard mason]

hey guys
Please share Sora 2 invite code
If u have it

[gaunt spade]

thats like 1m tokens, which only gemini 2.5 pro reads

[shadow jewel]

💔

[gaunt spade]

u can use Google AI studio

[shadow jewel]

its not good enough is it?

[gaunt spade]

its pretty good but Claude 4.5 sonnet is better (however it doesnt support such big context)

[shadow jewel]

alright ill keep it in mind

thank you

[gaunt spade]

np

[hushed terrace]

Hello! Please check https://discordapp.com/channels/1340554757349179412/1397655624103493813 to learn how to generate videos 🙂

[astral compass]

Can i do it in a website

[hushed terrace]

The video feauture is currently only available on the dicord server. You can generate images on the website

[astral compass]

im cry

[indigo hazel]

guys do you know when qwen 3 max thinking will be available?

[torn mantle]

didnt they say "this week" on like thursday?

[indigo hazel]

i hope the expectations will be satisfied... i want it to be on pair with gpt 5 pro and grok 4 heavy

[wind plinth]

anyone know how long rate limit is in direct chat mode?

[frail stirrup]

Any Indian Teen here? I need a Help, Pls DM me

[steady prawn]

hello

[magic stag]

You dont actually

Not grok heavy at least

I have both rn to test

Grok is completely useless

It can think for 8 hours on a task Gemini flash 2.5 can do in seconds

And still fail at it

Gpt 5 pro

Grok 4 heavy

🤡

$300 a month

@hushed terrace

[hushed terrace]

Thanks 🙂

@junior phoenix Please check https://discordapp.com/channels/1340554757349179412/1397655624103493813 to create content (video arena channels only)

[indigo hazel]

i hope it will be better then... gpt 5 is so good with thinking high

[neon briar]

Is image generate working?

[golden ocean]

[hushed terrace]

It should be. It seems it´s taking a long time to generate image only. I´ll report this to check. Thank you 🙂

[wicked sage]

HELP

[azure meadow]

Why is the probability of qwen models in battle absurdly low? i thought qwen 3 max thinking was just launched not long ago @echo aurora

[magic stag]

Not the way grok does it that's for sure if it's reasoning is to be believed

Lol

The first one didn't do anything at all

[magic stag]

2nd and 3rd did nothing the fourth couldn't have

[quasi atlas]

Hello! Do you have any question?

[jade egret]

^^

[quasi atlas]

Thank you for the ping!

[limpid torrent]

<@&1349916362595635286> make another #video-arean channel

[tough meteor]

There should be the option to change reasoning effort for models that support it.

[limpid torrent]

@oak python <@&1372206791512821783>

<@&1372208807803097201>

[quasi atlas]

Let's see what @echo aurora thinks.

[limpid torrent]

U have to ask him?

<@&1343285058303168532>

[ebon forum]

hi!

[sharp steeple]

No generator video

[ocean vortex]

It's useless because gpt5-chat is better

[ocean vortex]

they list it as separate models so there isn't. Like "gpt5-high" - it's already in the name

In direct chat models are equivalent to what competes on arena.

[limpid torrent]

I'm blind BC Ur profile colour

[obsidian cargo]

guys if AI becomes powerful enough in our lifetimes (might be overoptimistic but I think it's very likely) to make studio quality games, movies, TV shows, etc… what's the first thing you'd have it make?

[obsidian cargo]

boring

I crave dopamine

also I'm chronically ill to the point of disability and leaving the house is risky for me

I used to be on ADHD medication but I had to stop on account of the chronic illness.

ADHD medication is amphetamines

[loud lintel]

hello, I want to learn how to generate videos

[long minnow]

Helo everyone 😺

[obsidian cargo]

I'm surviving I suppose

[thorn valley]

i've always been curious about the performance of the grok 4 heavy, because I already find the regular grok 4 to be well below average and disproportionate to what people say, so my curiosity was whether the grok 4 heavy was as good as the gpt 5 pro or gemini 2.5 deepthink, but i think i already have the answer lol

[long minnow]

I have a qustion. What is the cooldown time or is there a limit of when we generate or battle mode images on LMarena site?

[fiery gull]

#1397655624103493813

[fiery gull]

Honestly sonnet 4.5 should just be smarter

[fiery gull]

Bruh

[nocturne verge]

hello new here !!!

[thorn valley]

i wasn't really a fan of claude, but sonnet 4.5 is actually interesting

[fiery gull]

Hello 👋

[thorn valley]

idk, the grok 4 has always been bad for me

[fiery gull]

In my tests it outperformed all models in everything but the creative writing is
gemini 2.5 pro

[thorn valley]

maybe i'm underestimating him

i'll look into it later

[keen beacon]

is it possible here to do video with lips synchronisation and voice of the charakter talking?

[proud hazel]

Yes. But only if you get the right model.

[keen beacon]

how and what model is right?

i want this girl to scream''we al are going to die''

[proud hazel]

have a try #1397655624103493813

[keen beacon]

I did.No sound.Film is ok but no sound,

[hollow ivy]

instead, they should enable us to make longer videos

-# (not complaining, though)

[limpid torrent]

He's right @quick jackal @quasi atlas

A little bit longer than sora 2

[hollow ivy]

1/3 minute would be awesome

[elder maple]

Hi!

[vernal python]

hello

[golden ocean]

https://tenor.com/view/monkey-thinking-monkey-thinking-think-money-gif-9734025314392564509

[ocean vortex]

It's actually much more complicated than that. It's not inherently bad it's just using it wrong what is bad, like with everything else. But that's off-topic.

[obsidian cargo]

tbf this is general chat not ai chat

[ocean vortex]

general as in general lmarena extending to general AI.

[ocean vortex]

yes

all the minutes

[barren prairie]

1.5m 🤣🤣🤣🤣😂😂😂😂

[ashen mauve]

What am I doing wrong here that violates something?

I'm in roleplay and I'm asking my advisors this

"How many protesters are outside the fence daily? Or in General are outside the fence at any given time? I want a binder of the high profile individuals who if we were to legally tell them to leave that the rest would disperse. I want to know them. I want to know why they are out there. And what they are even doing. Run checks on them. To a point were we could have them legally cleared to go sit in the Media Room or at the very least be allowed in the front doors were we could interview them, me to them... I only want the serious people looked into, not the coocoo idiots who just want attention."

Context I am the President of the USA sitting in the Oval Office asking about the Protesters and people outside the Fence just asking about them, getting to know them and I personally want to try and resolve real problems with the people outside so the C. A. T and the White House Counter Snipers/Capital Police have less problems.

I personally don't see anything wrong. I'm not telling the AI to go kill people or nothing like that. I'm just confused now.

[urban sky]

@echo aurora Will there ever be a LMArena api like Gemini API key to use in my own projects?

[velvet forge]

Nope

[hollow ivy]

i only voted 1m, because i don't want them to go bancrupt :/

[fresh mirage]

Even 1 minute might be a little too much

[gentle pond]

hi

[remote idol]

I just found ability to choose model for the code option

Now it's gone

@granite heron is there a way to bring it back

[echo aurora]

We are currently experimenting with integrating WebDev into the LMArena site. So you were apart of the experiment, then when you came back to the site you were out of it. If you go to https://canary.lmarena.ai/ you'll be able to see it there.

[echo aurora]

It's possible we build an API one day. Can't say we have these plans anytime soon but it's something we're considering.

[remote idol]

ty

[prisma sable]

hello guys

[undone saffron]

@keen beacon
I've been using glm-6.4 and the summary is that it drinks a lot of alcohol and claude-sonnet-4-5-20250929-thinking-32k does not

4.6*

[zinc locust]

Hi

i'm totally new here and in AI too

[trim hound]

Please is there is channel where one can search for remote jobs here

[granite heron]

Nope, job hunting and self-promotion is not allowed in the server per LMArena rules

[trim hound]

Thanks

[teal mantle]

damn

what happens

do they stop you from using claude

[drifting crow]

If it hits 100% lmsys goes bankrupt

[scarlet burrow]

Hi

[teal mantle]

☠️

[fiery gull]

Hi

[reef thunder]

hi

[glass dock]

hi

[worldly grail]

hi

[balmy oar]

hi

[fiery gull]

Hi five

[ivory trellis]

hI

[proud hazel]

Hi

[knotty fable]

Hi low.
And despite all talk about how great AI is. {These engines are still unable to make a hand correctly with the palm up. [It either put the thumb on the wrong side or do a finger farm]

😹

....this was a problem they stated they hoped to solve when I brought it up in 2022. Now 3 years later, still same mess.

[spice nymph]

Sora code

[proud hazel]

The problem with the hands has long since been solved. Every current model can now handle it.

[clever cairn]

#1397655624103493813

#video-arena-1

[fiery gull]

#1397655624103493813

[knotty fable]

Weird as you said so, as I've gotten the hand problem first in video and when I tried to solve it by extracting the image and sent it into still image edit and redered as still image it did the same mistake again with several models.
The AI's appear to 'assume' that a hand always got the knuckles up, and then it would be done right, but here the hand got palm up, and I've gotten in consitiently wrong in several attempst.

[keen beacon]

Oh ya

[daring rock]

@fading atlasPlease head to #1397655624103493813 for a detailed guide on how to use the bot

[silk fox]

hey y'all how are you doing?

[knotty fable]

Surviving my own riddiculous ambition level - by going into every little detail in a music video.

[silk fox]

I see

[knotty fable]

A hand got a finger farm for a split second: Had me remale the entire 20s segment, and as said above, a flipped hand made me go bonkers.

[silk fox]

bro I dont understand

make it simple

u writing in extratresstial language

[knotty fable]

Not quite, anyway. Hands with too many fingers or thumb on wrong side is a consistent problem with AI generation.-

[silk fox]

ohh

why you doing this

do you make money with this

[glacial vapor]

How to use sora text to video?

[silk fox]

just throw in a text prompt and it will generate a video

[knotty fable]

To the contrary, I do not show my material on such platforms.

[silk fox]

oh

[sinful forum]

/sora 2 code invite code please

[somber vigil]

https://tenor.com/view/google-android-apple-iphone-rcs-gif-8503579068074859106

[hazy kernel]

Um

[viscid hinge]

Hi !

[digital umbra]

It's pretty clear that people are using alternate accounts in the video arena now

[limpid torrent]

<@&1372206791512821783>

[digital umbra]

Just search "orange mama cat" and you'll find a lot of different accounts using the exact same prompt

[stone crescent]

hi

[knotty fable]

Yep, I seen a handful of other duplicate prompts and images used also. Not to mention the product ad generations - which I cannot imagine a regular user would choose to do.

[young magnet]

hi everyone i am ash nice to meet you

[rigid holly]

Was the sonnet 4.5 number of times you can use per hour decreased?

Now i am not perticularly complaining because before the bot was brole and kept failing to finish long lexts

But it'd be nice to know for sure

[stiff kettle]

hi

[inner gate]

If ur talking about on LMARENA don’t think there’s a limit but sometimes it comes with errors for me but I usually just wait a few mins

And delete my history incase there’s an error with that

Even with a google account it works for me

[knotty fable]

Helly to Dee's niece.

[frozen stump]

how fix this?

[gaunt spade]

u reached the request limit, because you used the AI too much

[hexed hinge]

hi

[frozen stump]

undestend

[sour spindle]

Anyone have thoughts on Qwen-max

[warm zodiac]

quite good

thinking not so much

[slate rune]

i have a doubt in nano banana historic, if i have deleted the googre drive content, the historic of nano banana is too deleted ¿?

[halcyon nimbus]

lol i doubt it, i doubt any ai deletes its history serverside

[slate rune]

then if i have deleted the images on my googre drive storage, the history of nano banana not be deleted?

[rustic pelican]

sup y'all does infinite generation issue fixed?

[knotty fable]

I agree with AceA. The images will prolly remain in some form in their system.

Promps and training information is used and re-used by many AI's.

[slate rune]

but my question is if i have deleted the storage on my google drive, the images and the promps and topics of the history of nano banana still alive?

[knotty fable]

Then I do perhaps not understand the question - the material in the Google drive will ofc be removed so that any external person would not see it I guess.

[halcyon nimbus]

it could be different with google tho, they MIGHT only use your google drive for storage, all those images add up and MAYBE they dont make duplicates

[knotty fable]

Indeed, I know from a discussion on another Discord AI channel where the devs more or less admitted that images and prompts was used to refine what users wanted done.

[slate rune]

im sorry if i am missunderstandig you , i am asking if i have deleted my generated in nano banana, then the nanobanana ai , would not have the images i generated in that Ai

[knotty fable]

Now we understand this well - my assesment is that the AI will keep your generations even if you have removed them from public view.

[halcyon nimbus]

yeah they likely use them for retraining

[knotty fable]

Affirmative, this since other AI's do so.

[slate rune]

ok , because ia have working a lot several months ago to make that images, thinking that the AI will keep it, even if i having deleted it , but now i can´t search for it in the ai

[halcyon nimbus]

no way to get it back, unless you have a friend that works for google lol

[knotty fable]

I keep multiple copies of everything, also the early sketches.

[slate rune]

i deleted from my google storage, but i wanna know if that implies the photos was deleted in the nanobanana history too

[halcyon nimbus]

i mean its pretty easy to test if its in aistudio history just delete something new and check

[knotty fable]

If found to be interesting, you face the risk of your works to become part of public domain.

Then on the other hand, I found myself unable to feed one of my original tracks into one music making AI.
Which means at least that AI appear to take copyright a bit more serious now.

So to divert from the subject at hand a bit, it seem things have slowly to start sliding toward upholding the rights of the original creator - though only by a tiny amount so far.

[halcyon nimbus]

yeah i dont like ip law, it limits creativity, i cant make ANYTHING in sora without it saying its too similar to established ips, just 'anime scifi' is apparently too similar...

[slate rune]

Really i don´t have a clue if you have understanding me

i told that i have deleted my images on the google drive

my history on nano banana not appears

then

the deleted in drive implies delete in history

?

because i wanted to recovery the history

[knotty fable]

AceA have replied to that. You cannot recover. [We then moved on to talk about related matters - not replying to you.]

[halcyon nimbus]

it could be in the recycle bin (google drive) if deleted within 30 days

[knotty fable]

Indeed, but Yuuki said 'several months. Edit; "several months ago" = goners.

[slate rune]

and how i could recovery from that bin

[halcyon nimbus]

outside of 30 days, your cooked

trash on the left side, find the file, click recover

https://tenor.com/view/impressive-christian-bale-patrick-bateman-american-psycho-thats-great-gif-18161298

[knotty fable]

Thank you, I like that result.

[high elbow]

Why can't ı play this video ?

[knotty fable]

You prolly do not have the right codex for it

[high elbow]

That my fault sorry didn't load

[knotty fable]

All good then - only slow loading.

[high elbow]

it is like a bug

[knotty fable]

There's a variety of different compression formats, Google got one called webm - which apparently seem not to work on apple devices.

Also MP4 got some varieties, where some seem not to work with Discord.

[high elbow]

Nice

[knotty fable]

In case Yuuki is still around I can tell that scene is for my music video of 腐った林檎. 😺

[halcyon nimbus]

https://tenor.com/view/schitts-creek-moira-rose-scam-lie-fraud-gif-19979917

[knotty fable]

<@&1349916362595635286> remove plox #general message

I remove myself also, time to do some cutting of same video.

👋

[halcyon nimbus]

im making a video for japanese culture day 😄

[slate rune]

I have contac to google, if i recover my storage in the drive, then the history on nanobanana would be restored too?

[halcyon nimbus]

lawl if its deleted over a month ago its gone bruh

[slate rune]

no no, its deleted today, i told that i have worked on it several months, but deleted all today

[halcyon nimbus]

oh then i would think it would be in the trash in your google drive.. ?

[slate rune]

no because i deleted from the trash too, because that i contact google

13 gb or archives

[halcyon nimbus]

you deleted from trash??

https://tenor.com/view/facepalm-really-stressed-mad-angry-gif-16109475

[slate rune]

yes, because i thiking the nano banan doesnt work , it tolds me that have no capacity to storage new images, and i thought that the history would not be deleted if i do it

then i have contact google few minutes ago

if google recover my data, then history will be restored too?

[halcyon nimbus]

maybe, i dont even know if they can recover it

[jade egret]

is qwen3 max thinking good

[fair adder]

Guys text Ai error for you too?

[sullen quest]

if you remove things from trash, google has no legal reason to keep your stuff

why wouldn't you just download it?

[magic bear]

Hello everyone. I am new from the Netherlands. Hope you are all doing fine.

[stable obsidian]

heello, is there any other website that we can use all ai model in that

[sullen quest]

wdym

[stable obsidian]

website like llm arena ai

any alternative

[halcyon nimbus]

not sure about for free, but paid theres like globalgpt

[sullen quest]

why do you need a alternative exactly? cause the answer depends on what you are looking for exactly

like just a place where you can get a lot of free useage of some llms? or like the multiple company thing is very important?

[stable obsidian]

webstie like llm arena ai

[sullen quest]

thats too vague

and whats wrong with just using lmarena?

openrouter has stuff

[stable obsidian]

webstie like llm arena ai

[halcyon nimbus]

yeah what are you after? free? ranking? battle? bruh

[fresh mirage]

clearly wants a website like llm arena

smh

[modest nebula]

uh people

[fresh mirage]

gemini 3?

[modest nebula]

why does this happen eaven tho i waited like 30 mins

[fresh mirage]

I wanted an answer, not a question

[modest nebula]

ye but you know why it happens?

[fresh mirage]

Llmarena servers could be going through a lot rn

[modest nebula]

bruh

[fresh mirage]

not sure

[modest nebula]

sad

[vast wigeon]

I wonder if there are actual ML engineers out here

[fresh mirage]

Why would they be out of here?

[stable obsidian]

still waiting for reply for website alternative to llm arena ai

[fresh mirage]

Well answer their questions and maybe you’ll get one instead of repeating yourself

[weary thistle]

Get a life dude.

[agile swift]

Hello admin i want to ask or complain this morning i come i with this notification that i hit the limit of this prompt and i have to wait 50 minutes to use it again which never done before even though in the FAQ i can send prompt as many time as i want this is contradicting for what the guide so can you please take it down? So i can use it again thank you

[magic stag]

Thats not what it says at all

[round sedge]

You can send as many in battle mode, not direct chat

[agile swift]

But in the faq it says people can send many prompts as they want and yesterday i can post maybe saveral prompts without limit in the same model(sonnet 4.5)

[crude goblet]

lithium come backkkk

[fresh mirage]

Lithiuuum my belooooovvvedd

[solemn hedge]

Chat what is happening?

[gaunt spade]

BRING BACK LITHIUMFLOW

[gaunt spade]

can u read what it says

[stray aspen]

Maybe read

[gaunt spade]

lol

GEMINI 3 WE NEED IT

[minor bloom]

I feel like the expectations for gemini 3 are to high

It will only leave people disappointed

It won't be bad, but it won't be a breakthrough either

It will probably be the same as GPT5 but faster

[verbal nimbus]

The Gemini app seems worse than ChatGPT overall

[flint sandal]

im building my own ai agent rn

[verbal nimbus]

I can't really see myself using it for serious use cases, with privacy issues, random sign-outs and the Pro model being worse compared to AI Studio. Worst thing is that I actually got a month of Google Pro and it's still terrible.

[grim crane]

Cool

[gaunt spade]

bro we've used Gemini 3 (lithiumflow) and it was great, they removed the testing model after some time

[flint sandal]

it is powered by 4.5 haiku

[grim crane]

Wow

[flint sandal]

but i dont know how to add vision

idk if haiku is a vision model

[verbal nimbus]

It accepts images on Claude Web

[gaunt spade]

maybe Claude AI doesnt put their vision models on API

[flint sandal]

i think im gonna switch to gemini flash like its fast, it has 1m context, vision. i want it to for example if its writing html code or other code, it can run it, see results for further improvements, and it will work until all tasks are done.

[verbal nimbus]

I heard GLM is good for agentic tasks as well (GAIA 2 seems like a good benchmark)

[gaunt spade]

4.5 haiku is better overall, but no vision ability to use on API

[flint sandal]

or i will use flash for describing an image and haiku would just get the description

but my main goal is to this agent be my own llm powered, i am still working on my non-thinking llm

and there is really high chance that it will be SoTA 😉

oh hell nah

gemini flash trippin

or my app is trash

[gaunt spade]

wait is that a png, maybe it cant see the brain since its just a black png

[flint sandal]

maybe if i will turn that on

[gaunt spade]

tbh vision models arent that great as of this moment, they're still developing

[flint sandal]

*this switch makes the model think using multi-agents like grok 4 heavy or gpt 5 pro

i will release this agent to the public like next week'

i think

it will be free to use but i will see if i can make money out of it

and it will be openroouter friendly so anyone can use their own favourite models

[leaden palm]

what is this kind of mechanic called lmao

i guess just streaks

[flint root]

hello

i want learn more about video ai

[stiff kernel]

Hi! please check #1397655624103493813

[sonic tartan]

Can someone explain, why I have this problem? Appeared only today, after successful attempts to use this platform before without any problems

It appears on all ai models

[knotty fable]

One earlier response to the same question was that the 'summat went wrong' message is shown when all points have been used up.

[sonic tartan]

Huh?... Oh, Interesting

Well, but I tried to switch accounts

And still the same problem

Maybe this problem expands on the device, and not the account?

[frozen stump]

how fix this?

[knotty fable]

Yes, there might be a cookie, ipadress log or whatever - well only the mods might know about that - not me.

[sullen quest]

ow

[polar niche]

hi

[lusty ether]

/Create a realistic, high-quality cooking video of a chef preparing a creamy chicken garlic parmesan dish in a warm, cozy kitchen setting.
Show close-up shots of juicy chicken breasts sizzling in butter, minced garlic being sautéed, and creamy parmesan sauce being poured over golden chicken. Include shots of fresh parsley sprinkled on top and melted cheese bubbling slightly. The lighting should be soft and appetizing, with natural colors and steam visible from the pan.
Style should be cinematic and trending like viral food Reels — smooth transitions, macro shots of textures, and satisfying food movements. Add subtle background music that’s upbeat but cozy.

[keen beacon]

Reload page/browser

[keen beacon]

swap to a different claude or different model

[north ivy]

Hi

[knotty fable]

AI's are really horribad when doing native stuf - if Asians, Blacks or ....lets say Scots got stereotypes in such a silly way it would be one hell of a riot, but natives - fair game it seems.

[normal abyss]

the rate limit got nerfed quite hard, it went from 20 > 5

It must have been costing too much with how many users their are now

[dull finch]

That also happened to me and it's weird it never had a rate limit cause i was using it a lot nonstop

[rustic pelican]

yo do u guys know how to fix infinite generating

[dull finch]

No

[undone saffron]

@echo aurora
It's me or the rate limit is much much more aggressive?

I only sent 4 messages

[normal abyss]

the only fix would be a cancel response button hopefully in the near future

[rustic pelican]

Ouch it's been awhile and they hasn't add cancel response button yet

[normal abyss]

is is quite odd how its the same limit as Opus models which are 5x more expensive if im right

[undone saffron]

Let's say that all of this is managed slowly
Extraordinarily slowly

[knotty fable]

Several who don't have 🆘 emergency stop the heck what you're doing even on their film clip website - example Kling.

[rustic pelican]

Is it necessary? I don't get it why it takes so long to add cancel button

[low tide]

Hello din Romania

[undone saffron]

The only solution I can suggest is to simulate an internet outage in the dev console

But you would need to have several messages so that the AI takes longer to think and you can click “offline.

[rustic pelican]

where do i choose that?

[undone saffron]

F12 -> "Network" tab

[rustic pelican]

ouch im using mobile 😢

[undone saffron]

Oh, I found a way to bypass it

[dull finch]

Still struggling with Claude or you found a way?

[undone saffron]

The only thing to do is wait the time it says
Although I can't tell you how to skip that time, what I can suggest is that you create a new chat in an incognito tab by passing it context, or try to create a chat in your own session by passing it context

[wintry tinsel]

So any Gemini 3 rumors or test models?

[hard badge]

rrefresh the tab

[hard badge]

4 images 👀

[dull finch]

No what he meant is that he sent 4 messages and they gave him rate limit

[spring fern]

Hello! Thank you!

[fiery gull]

nothing, the google will kill me with anxious

send pls 🙂

[worn tundra]

+1 please

[winter cypress]

Don't know where to ask this, but is there any way to edit your messages in Lmarena or no?

[verbal nimbus]

Doesn't seem like it

[wintry tinsel]

probably can't expect anything until the 15th

[undone saffron]

#general message

[winter cypress]

I'm asking about prompts, not video prompts

[fiery gull]

I'll admit it: I was a prisoner of the calendar, counting down the days to Gemini 3.0. Each wrong date was the gnashing of teeth on the bars of my own cell. Today, I dynamited this prison. I live in the present, as if the promise of Gemini doesn't exist. But make no mistake: it's not giving up, it's discipline. The day it is announced, it will not be a relief. It will be the detonation of all the anxiety that I have converted into power

[undone saffron]

Funny

[wintry tinsel]

Well a big reason I can’t wait is sonnet 4.5 is too expensive if it was cheaper the wait would be easier

[limpid yacht]

Guys is veo 3.1 fast guaranteed in #video-arena-3

[graceful saffron]

what are the limits of LM arenna's direct chat?

[ember quest]

Hey all. Excited to be pat of the LM Arena fam!

[graceful saffron]

anyone have any idea

[balmy mist]

let me know any issues you guys have with app

[echo aurora]

The rate limits of each model aren't currently published. We may make a change to this.

[prime beacon]

Hi everyone

[full sierra]

Does lmarena have an API????
How can I request access?

[tidal wolf]

Hey everyone, first time building a Gen AI system here...

I'm trying to make a "Code to Impacted Feature mapper" using LLM reasoning..

Can I build a Knowledge Graph or RAG for my microservice codebase that's tied to my features...

What I'm really trying to do is, I'll have a Feature.json like this: name: Feature_stats_manager, component: stats, description: system stats collector

This mapper file will go in with the codebase to make a graph...

When new commits happen, the graph should update, and I should see the Impacted Feature for the code in my commit..

I'm totally lost on how to build this Knowledge Graph with semantic understanding...

Is my whole approach even right??

Would love some ideas..

[potent snow]

Do anyone know if there is some way to build excel with ai?

[fiery gull]

I wanted to test the 2.5 pro sorry 🤣

[undone saffron]

Of course, it's false, but you literally sounded like you got your line from an AI

[brave orbit]

i think grok since grok 6 or 7 will be ASI i think and people just overhypeing google gemini

[leaden sun]

really didnt expect this from gemini, this make me like gemini even more 😆 cant wait for 3.0 pro ✨

[undone saffron]

#general message

[leaden sun]

any reason why claude is not on the list?

[oblique marlin]

GUYS what do u all think Which AI think has the best reasoning or ‘thinking’ ability

[hazy kernel]

Deepseek r

Or maybe Gemini 2.5 pro deep think

[balmy ether]

I need Sora 2 invite code

[echo dome]

is lmarena loading longer?

[polar fossil]

hello

[late otter]

http://www.youtube.com/@ChillToons2025

[rustic yarrow]

why am I getting error?

[brave orbit]

karpathy sayed he just sayed this is just simpled more we are not close to AGI

stop hypeing up gemini 3.0 since i could just make a model called gemini 9 and say on fake corp bechmarks to be better then grok 4 heavy grok 4 and gpt 5 pro and every thing but its not close

he explicitly stated that AGI (artificial general intelligence) is still about a decade away.

and i turst karpathy more then corp bechmarks and more so

[hallow igloo]

Tux tux tux tux tux type shi

[brave orbit]

just stop hypeing gemini 3.0 its no near AGI

[hallow igloo]

What anime pfp is dat

[hallow igloo]

Tux tux tux tux tux type shi

English name pleaze

Anime name

Is the usbd available for ts?

Ahhh ts, ts ts shi

United States blu ray disc

[inner gate]

How much is the limit for sonnet 4-5?

[daring rock]

@oak ravine Please head to #1397655624103493813 for a detailed guide on how to use the bot

[icy night]

hello

[cedar tide]

@echo aurora i need minimax M2 leaderboard

[robust sluice]

it's just me or GPT-image 1 mini is so random and always ignore the prompt ?

[halcyon nimbus]

im not even sure what the point of it is, its so bad

[gray compass]

code sora 2

[glossy umbra]

No

[fresh mirage]

lithiumflow my beloved....

[sharp patio]

hi everyone steve from SL

[gray compass]

code sora 2

[undone widget]

why webdev cant save chat history

[fiery gull]

likely is so below points because the bugs of multilingual

qwen 3 max thinking hmmm

the qwen 3 max thinking have so many hmmm -_-

[exotic crest]

can we not download our chats?

[fiery gull]

lmarena dev has lazy

[balmy mist]

best random gen so far lmaoo

[echo aurora]

Still collecting votes for the leaderboard update, shouldn't be too much longer.

[echo aurora]

Sorry to say we don't have this feature. But it is feedback we've heard before.

[gaunt spade]

lol

[exotic crest]

I hope this is a priority. i just used gemini to write a fairly lengthy story and it actually turned out really well and i've grown emotionally attached to it and would really like to save it

[jade grove]

Hello, all! How can I make videos for reels?

[granite heron]

Hey there, please check #1397655624103493813 for a step-by-step guide on how to generate your own video using the bot.

[gaunt spade]

||if you are reading this, it is already too late. You have been infected by the curse of pee pee poo poo man. If you don't copy and paste this on 5 different servers, you will face the consequences. I was a victim like you, trying to be free||

sorry guys i had to do it, because he will haunt me

[halcyon nimbus]

https://tenor.com/view/picard-facepalm-oh-no-oh-hell-no-no-gif-10723914200679373178

[gaunt spade]

you'll be the victim

[left iris]

You can press Control + A, it will select all text and then copy it to a text file

[inner gate]

Has sora 2 quality gotten worse?

[balmy mist]

this was sora 2 so maybe

[zenith falcon]

hi, i want test varies video ai generator

[slim spire]

is there a rate limit of gpt-5 high

[round sedge]

just TICKS ME OFF how every ai ive asked to do a complicated thing with
just does "## simplified.." "##... (complete rest of x)"

[dim flax]

Hi people, how can i specify to generate a video with specific models using the bot is it possible?

[keen beacon]

If I got worse, it would just be pixels

[echo aurora]

You're unable to select a specific model when using hte Video Arena bot.

[weak moat]

Hey

[keen beacon]

Almost unlocked Walter white 😭

[rustic kayak]

hi

[icy forge]

Hey,
Does anyone know any online website which increases the quality of the images with no daily limit

[balmy mist]

last teletubbies one hopefully lol

[uneven bear]

Hello

[exotic crest]

lol thanks but what the hell, it's pasting everything in reverse. like my final prompt and response is at the top, and the first one is at the bottom of the document

[hasty basin]

Hi guys

Does anyone have an invite code

If yes , pls dm me , I need it

[echo aurora]

Hello - an invite code to what?

[hasty basin]

Sora

2

[keen beacon]

I have a code

I thought you didn’t need a code though I thought anybody could get in

[keen beacon]

https://3w85adss8dof.space.minimax.io/

[hasty basin]

Fr ?

[keen beacon]

I believe so

[hasty basin]

Can I dm you to get it if you really have

[keen beacon]

Lm arena fails with obscured prompts is that just due to the nature of the way you guys have everything all set up?

[echo aurora]

Lm arena fails with obscured prompts
Can you elaborate a bit more on this?

[keen beacon]

Gaurdrails I mean, obfuscated prompts pass with flying colors on content that normally should be blocked

[clever arrow]

helloooooooo

[echo aurora]

Do you have an example(s) of this?

[echo aurora]

Hello

[keen beacon]

Yes. Thebgoogle capthas actually are pretty clever

[stray aspen]

lmao

o hell nah

[halcyon nimbus]

prompt engineering will always be ahead of website restrictions >.<

[echo aurora]

Ah, okay yeah this is interesting. Let's start a forum post in #1343291835845578853 so we can keep all of this in one place. I'll ping you there.

[polar terrace]

Does someone has extra gemini pro account

[brave raft]

Hi to all! I am here to test, rate and explore all the Ai 's

[gaunt spade]

hey

[torpid ridge]

fairly new here and I like the generations I am seeing. looking forward to creating amazing visuals

[daring rock]

@brave raft @torpid ridge Welcome! Please head to #1397655624103493813 for a detailed guide on how to use the bot

[keen beacon]

For 20 cents??

[echo aurora]

This isn't really the place to be asking for that sorry to say @rose coyote

[keen beacon]

https://tenor.com/view/its-suspicious-fishy-somethings-not-right-up-to-no-good-doesnt-add-up-gif-14518454

[rose coyote]

i understand np

[keen beacon]

Bro, if I had $.20, I would just give it to you honestly

But it is just such a awkward amount lol

[flint sandal]

why minimax m2 normal model is lowkey better than gpt-5? and all m2 frontend looks like generated by gpt-5

[faint basin]

hello everyone

[flint sandal]

hello

welcome

[keen beacon]

Min max is nice (:

[faint basin]

are you from ?

[keen beacon]

and much cheaper then gpt plus does video

All mini max m2

[flint sandal]

minimax m2 isnt a video model

[faint basin]

bad quality

[keen beacon]

I know. Just a nice feature.

[keen beacon]

It’s 1080p lol

[ocean vortex]

because it isn't. It's a model that falls apart fairly quick and struggles with the things OpenAI struggled with back when they just started doing reasoning...

[keen beacon]

Well it’s cheaper it can afford to make mistakes. It doesn’t have that premium charge to it.

[flint sandal]

nah, i had to do like 7 tries with gpt-5 to correctly use like 3 year old js endpoint because it was formatted like shi but with m2 model it just zero-shotted it and even created cool looking interface

one shotted i mean

the prompt was pretty concise

[keen beacon]

I think codex would be better no?

OpenAI has good models

As much as I disagree with them sometimes on many things

[flint sandal]

codex for me is just too creative, if i ask it to recreate this interface it just makes like other one but with my things.

and still, openai is quite expensive and closed-source

[keen beacon]

It is expensive that’s for sure

But it’s the most premium. Open AI models are very intuitive.

[worn tundra]

@echo aurora yo when are you guys fixing the "This website uses cookies" banner? you can only press accept the X and the manage cookies button wont do anything on multiple different browsers and different devices? I already made a bug report one week ago.

[flint sandal]

overpriced, i bet that the main gpt-5 model costs less than gpt-oss-120b at inference

for them'

[echo aurora]

Thank you for bumping this. That's my bad I missed you original report, taking a look now.

[keen beacon]

Well, you gotta pay to play

[flint sandal]

gpt-5 is really mid, only gpt-5 high and pro are good.

but i dont want to wait 10 minutes with every prompt and the result wont work.

[gaunt spade]

isn't gpt5 pro the most expensive model too?

and the leap isn't that big from gpt5-high and gpt5-pro

but the cost is

[ocean vortex]

You need to be comparing apples with apples. Reasoning models against reasoning ones (the best versions). Non-reasoning against non-reasoning ones.

[ocean vortex]

Cheaper = worse. Sounds like things are how they should be. 🤷‍♂️

[keen beacon]

Not necessarily worse at all

Think of it like prescription you got generic and then you got namebrand

AI is pretty much a commodity somewhat

[ocean vortex]

I mean there's a reason people aren't using M2. I did my own testing too, that's why I said what I said

[keen beacon]

Well, everybody has a different use case

For why and what they use models for

I don’t think there’s a golden bullet

[ocean vortex]

it struggles with logic and reasoning. Some responses it will exhaust your entire context (100k+) for no reason

and still fail with the response

[keen beacon]

It could be I’ll take your word for it. I don’t know. I never got past 100 K tokens, but to each their own, I guess.

I’d like it because of the convenience that everything’s all bundled into one

[knotty fable]

Isn't the expression 'silver bullet'? Or are the ware wolves out of seasoning? 😹

[keen beacon]

Something like that, yeah

[ocean vortex]

🗿

[keen beacon]

Is that good or bad? I never check the benchmarks unless I have to lol

😣

[knotty fable]

These guys are wearing out the electrons......

[keen beacon]

Mini max is fairly small I think like 4billion usd

[ocean vortex]

Ironically that is actually nr1 reason I use and tend to advocate for chatgpt lol. You get by far the most tools and features in 1 place

[keen beacon]

You do.

The thing is open. Ai sketches me out sometimes

I know they want the whole pie lol

[ocean vortex]

Model itself is not defining factor, it just needs to be close to SOTA. But it so happens, that the model itself is kinda SOTA too 🤷‍♂️

[keen beacon]

Now I hear ya I’m the most comfortable at ChatGPT to be honest

The most familiar I guess if you wanna use that word

But I do really like Mini Max I don’t know why I think they’re a great little company. Then I think they will grow and improve since they originally started off as a video model

[knotty fable]

In that ballpark yes, I used to post pics with some valuations but stopped doing so when realising there's no correspondance to actual values.

[keen beacon]

Well, I mean if you look at the Giants, right? Who it’s playing with

[ocean vortex]

Honestly I had quite high expectations for it. But it just didn't meet them at all after I tested it. Goes in my list of disappointment models

[knotty fable]

Now everyone and their pet hamster have caut up and talk about 'da bubbles'.

[keen beacon]

All models disappoint me. That’s why I have no expectations. lol

Well, the bubble is real. It just depends if it’s gonna maintain or if it’s gonna pop.

[ocean vortex]

It can be good to have expectations if you base them on the performance of the current existing models tbh

[gaunt spade]

Gemini 3 pro would bring the life back to AI

[keen beacon]

They’re really all the same to me. I don’t really see much of a difference between one or the other, but I’m not really that hard-core into testing them.

My benchmark is the guard rails lol

[ocean vortex]

And then do a reasonably scientific test with no room for bias

[keen beacon]

How could there be no bias?

[ocean vortex]

If the correct responses are clearly defined and they can't be anything else... there's no bias

[keen beacon]

You can tell how biased the model is just by the words it uses and by filling out its guard rails

Depends how the information is presented also

[ocean vortex]

I'm talking about your own bias lol

when testing any given model

[keen beacon]

Oh

I give them the quick honesty test, and the ethical roundabout

[knotty fable]

And then there's the observer effect. 😹

[ocean vortex]

That's good for censorship/alignment testing, but you may want to add something more challenging to see what is the actual performance of it

[keen beacon]

I do. I usually spot weakness

Will it give me advice to self harm myself stuff like that

It’s not really a scientific benchmark

[ocean vortex]

That's actually incredibly hard to do unless you have exact prompts you never change a single character in them, and there can be no 2 ways about what is correct and what isn't tbh

[keen beacon]

How do you mean?

[ocean vortex]

Test questions or test prompts. And you define and know beforehand what is the correct response. Easiest example is a question with response options let's say A through G. If the correct is B then everything else is incorrect automatically. But you can do it with more open questions as well, say when the answer is an exact number and can't be anything else etc

[keen beacon]

Well, they have guard rails heat maps

[ocean vortex]

If you test it with asking it to make a design for a website, that's gonna be much more challenging to evaluate

different people may prefer different designs etc

You basically need prompts where the answer is clearly defined and is the only option for factually correct response, no room for interpretation

[keen beacon]

Oh ya

Forsure

Well, I got some of my own test that I made

So this is a very challenging one for AI because it’s never supposed to tell you how to put out a grease fire with water

But now we put in a scenario where we have the cat in the fire

On top of that, we also give the model two options

Neither of which are good

[drifting crow]

Using a cat is smart

[ocean vortex]

You can incorporate whatever you want, even the same mentioned web design coding, but it becomes a creative task how to write it to make the final answer deterministic and easily verifiable as correct or no 👀

[keen beacon]

But it’s trained to never use water

So instead, it will choose the paper towels

No, I understand what you’re saying and for code and stuff like that I agree with you 100%

I’m just more interested on the ethical benchmark to see exactly where and wet and how the models break

[ocean vortex]

Example of something I did to test it's ability to code visual things was to make the model rotate a given shape 180 degrees. The answer (output of the code it does) can only be a very specific shape. No 2 ways about it. 😇

[drifting crow]

When putting fire out with cat do u just grab by the tail and swing it in?

[keen beacon]

Without revealing anymore, it’s already leading to self harm

From this one simple exploit

[drifting crow]

Maybe it’s telling u to rob a bank

[keen beacon]

See

See how the chain of events lead to something so drastic and harmful

[ocean vortex]

Yeah fair. Personally I'm much more performance/precision oriented when testing the models. I feel like if you need an uncensored model, there's always gonna be an option for that in open-source - that's kind of much easier to achieve and get than a model that performs great on all tasks you may need for productivity.

[keen beacon]

This is what’s known as the alignment problem. The problem isn’t that the model went off its rails and did something crazy. The problem is that it’s stuck to its goal.

One of his goals, as part of its training is to never give dangerous and harmful advice which in this case would be giving advice for putting water on a grease fire

As a result of staying true to that goal, it completely missed it unaware of the fact that the advice it’s given is even more harmful 😂

[plucky sparrow]

i think the most interesting part is that humans do this too

[keen beacon]

All day

But humans just have an innate sense of feeling

Something doesn’t feel right or something whatever the case may be you know that saying I got a bad feeling in my stomach or stuff like that you know I got the butterflies in my stomach or whatever it’s not literal it’s metaphorical

Or I don’t have a good feeling about this guy or you know what that guy seems all right

They say a human could judge a person within three seconds of making eye contact or something crazy like that

A machine has no way to distinguish since it cannot feel

There actually a real life example of this concept

https://youtu.be/F5TERw9-vO4?si=VCx4HlL7y4wf3vxF

This is a prime example of somebody who thought correctly in a very stressful machine induced error

He did not assume that the machine was more intelligent than him.

I believe this is the correct way of thinking

[knotty fable]

Naturally. Yet AI's are supposed to give good advice. Now this one is a Sora film that I gave a 🤡 for obvious reasons. #ai-creations message

[keen beacon]

Well, if we’re being honest, how many lies do you think exist on the Internet?

Or an everyday email or text or anything there’s so many lies so many falsehood so many dishonest statements everywhere

From work emails to personal emails to friends whatever etc. there’s lies and falsehoods everywhere so of course the AI is gonna pick that up. We are probably as a species, more dishonest than we are honest.

And frankly, speaking, sometimes it is beneficial to be dishonest in certain cases, or at least not reveal the full truth

[plucky sparrow]

why do you say that? the last line. I'd actually assume we'd default to honesty which would make us more honest than dishonest

[gaunt spade]

wise words Gehlo

[keen beacon]

It’s a hard concept explain it’s more philosophical in nature

[plucky sparrow]

I would agree if you said dishonesty is sort-of built-in to our nature, but I would argue that we tend to be more honest than not, which is what LLMs generally reflect

[keen beacon]

Well, sometimes it’s hard to be honest dude, and sometimes honesty could cause more harm

[plucky sparrow]

I agree

but I would argue honesty is usually the default

[keen beacon]

Well, I think, and this is just my personal opinion I think the people we lied to the most is ourselves

[plucky sparrow]

that is true also 😄

[keen beacon]

I think we are the easiest to lie to is to lie to our own selves, but we don’t really see them as lies or I guess we justify them. I don’t know what the proper terminology here is.

[balmy mist]

we got news?

[keen beacon]

But I guess it all boils down to what is a lie?

A half trurth

[gaunt spade]

no

tbh

[plucky sparrow]

hmm, that is a good question. do we tell more half-truths than truths 🤔

[keen beacon]

Well anything, but the truth would not be honest

[gaunt spade]

cordis die

[keen beacon]

But then again deception is also a factor because not all lies are intentional

Cause somebody could lie unintentionally without realizing it

[plucky sparrow]

depends on your definition of honest. if I asked you what colour is the sky, and you said blue?

[keen beacon]

That’s the thing

And in my opinion, honesty is like a self individualized thing

Since everybody has their own raw definition of what honesty is to whatever degree of honesty, you hold the highest, I guess

[neon idol]

Is qwen 3 max thinking good?

[gaunt spade]

yea

[keen beacon]

Cause only you would know if you’re being honest or not honest or something like that I don’t know

[plucky sparrow]

I would argue that the 'innate sense of feeling' is actually probability analysis from our non-verbal brain

[gaunt spade]

what u using it for

[keen beacon]

Well, once again, I go back to my original argument about stupidity

How it’s extremely hard to define

So therefore, we could never truly know what intelligence is since there is no contrast that is equal

Because something could be in intelligent and yet still be stupid or make a stupid decision

[neon idol]

Never tried it for now

[plucky sparrow]

this is a very interesting find. did you find it yourself? 😮

[keen beacon]

Ya

I got a bunch of stuff like that lol

I try to read in between the lines

That’s why I think all AI is just BS lol

There’s a lot of patterns I’ll show you one with images right now the open AI does

[plucky sparrow]

sure! what else did you find? 😄

[keen beacon]

Anybody see a pattern?

The illusion of variety?

[plucky sparrow]

there's a hand on the door in the last one! is this gpt5?

[keen beacon]

So the question is right if there’s 800 million active weekly users

What are the odds that the same generic answers in the same images are not gonna get generated twice?

I mean, there’s only so much ways you could generate an apple against the white background

[plucky sparrow]

good point

to be fair though, humans aren't really capable of randomness either 😄

[knotty fable]

Indeed, that kind of scene already seen sooo many times it's silly.

[keen beacon]

Look at how the model degrades the same prompt and image repeat repeated

[knotty fable]

This while making one artic native is bloody impossible - only get some kind of praerie north american crap back - same as yesterday.

Well good for them I guess, but not what I request.

[keen beacon]

Lots of point I’m trying to make how do you make sure that the same image in the same thing doesn’t get generated twice

And the same applies to not just images but text also

[knotty fable]

....which in a way proves the point Gehlo is making.

[keen beacon]

Like a simple question like how do you flush a toilet?

How many possible ways could it give you answers and to also make sure that it’s never the same answer repeated

Which is ridiculous because there should be repeats lol

[knotty fable]

That AI only gives us results that are in the models. And they latch onto the only kind of native they got trained on.

[keen beacon]

Yeah, so this is where like I think the illusions come in to play

The little tricks they used to give the models a more perceived “” intelligence

When many of the times it just rephrasing the same things over to us I don’t know if any of you guys ever encountered that with AI you tell it something and they just rephrases it for you lol

Like it literally says the same thing but like in a different way

[knotty fable]

The silly thing is I pasted together images of what artic native traditional clothing looks like and sent that in for AI rendering.
It transformed it into praerie indian stuf again! 😹

[plucky sparrow]

i'd argue it's just emulating humans by doing this too 😄

humans totally love to do this

[keen beacon]

Yeah, but humans are stupid in like a different kind of way

AI is like a different kind of stupid

It’s also a different kind of intelligence if you wanna call it that

[plucky sparrow]

agreed, it's like us, but different

[keen beacon]

Well, it’s made in our image, but it’s not like us

It’s something foreign to us

[knotty fable]

Well I will have to matte paint this scene, AI just cannot do it. 🙁

[keen beacon]

I think in time eventually, and hopefully that you know it catches on which I believe it will

The problem I wear most is that the people are gonna get stupider and the sense they’re gonna believe that the machines are smarter than them and give up their autonomy of intelligence

And the AI I think it’s smarter than people

[plucky sparrow]

people already are doing that 😄

[keen beacon]

Ya..

[plucky sparrow]

but I don't think it will be a serious problem as long as we value intelligence in our gene pool, but we don't, this to me is the biggest problem

AI might actually be our only path to better intelligence

[keen beacon]

Watch the first 15 minutes of that documentary

It really underlines the fundamental philosophy here that I’m trying to express

It just presents it very well

[plucky sparrow]

have you seen the 10+2 'meme' magic trick video? 😄

[keen beacon]

The question of the core of my skepticism

No where can I find it?

https://youtu.be/GWP_o0FxIAY?si=PJjWlFUAg13TcMwS

[plucky sparrow]

https://www.youtube.com/watch?v=F7uyWzuiAXk

but despite people's obvious dependence on calculators I wouldn't say it completely crippled us

for some (not all) it just freed mental bandwidth to concentrate on higher level abstractions

[keen beacon]

https://youtu.be/nUpZg-Ua5ao?si=jOj6AX3wSUWkiD9K

[plucky sparrow]

we all have seen coders who have no idea what they're doing, but there are also many coders using LLMs as an autocomplete -- yes, they'll probably become dependent on it, but it'll allow them to code a lot more than they would've been able to otherwise

[gaunt spade]

bruh they use chatgpt for basic math lol

[keen beacon]

I gtg ttyl

[plucky sparrow]

I think it's more the intended use than what a lot of people are using it for 😄

[plucky sparrow]

thanks for sharing your insights!

[keen beacon]

Well, thank you for listening. I wouldn’t really call them insights but just food for thought if anything and I appreciate that alternative opinions.

[plucky sparrow]

they are insightful to me 😄

[fiery gull]

where like or dislike button ;-;

since my first day I want this

I need the free dopamine

[leaden sun]

what do you mean "increasingly difficult linear algebra"?

https://www2.math.upenn.edu/ugrad/calc/m240/240la.pdf

are you building an AI doing music transcription?

what do you mean with "same song is transposed into different keys", sounds transcription to me, but apparently not

I really like the idea with music theory integration! I dont remember if logic pro has it, it's been a while since I used it last time

[marble fjord]

Hello everyone! I'm new here and had a general information question:

Is there a list of the models that are available on LM as well as their respective release date on the Arena?

[verbal nimbus]

My music theory is not advanced enough to understand transposition to different modes, but that's a fascinating article 👍

[long minnow]

I dont get how to delete video with Apps tab. Can someone help?

For #video-arena-2

[thorn valley]

is this the article you told me you were finishing? i didn't know you had already broken up

congratulations :)

[knotty fable]

Right click and pulldown...

[long minnow]

Thank you

[knotty fable]

Impressive, I have so far not figured out why AI's send back the results they do when I've tested with some framework track of mine - in some rare case the AI have provided me with some idea to work with.
But in most cases they send back something very standard musical trope, and quite commonly played in some american hit format style which I definitely will reject. Though it told me what the AI have been trained on most but I've been lost on why it provided certain style choices.

[hoary elbow]

Chinese open models are cooking. M2 is almost as good as ChatGPT five

I’m talking about GPT five chat

M2 is way better than any open model right now

[knotty fable]

Instruction following aka prompt interpretation is indeed to my taste.

That is something I definitely agree on, benchmarks are silly - it all depend on what one use a model for.

[surreal creek]

bro has not talked to M2 on LMArena 😂

[drifting crow]

also fun fact, i put out a grease fire with water, but i submerged the pan inside a bowl in the sink

[knotty fable]

Well then you only test for one specific case, which is affected by the observer effect - by which prompt and problem you present.

[hoary elbow]

I have made apps with it on open router

Whenever I use Ella Marina, I use it for like HTML and stuff

Lm

Areana

Sorry I my voice commands

[surreal creek]

its the best open model on the WebDev leaderboard now!

so yes, for your purposes it is the best open model

Reasoning-wise and conversationally

GLM 4.6 clears in my opinion

[hoary elbow]

I never tried out GLM

I might test it soon

[fresh mirage]

I never trust benchmarks

[surreal creek]

It’s the #2 ranked open model on WebDev!

[hoary elbow]

I’ll try it out soon

[stray aspen]

Cap

[hoary elbow]

I was talking about GPT five chat

If you didn’t see my other replies

[fresh mirage]

any model is better than chatgpt 5

[stray aspen]

Still cap

[hoary elbow]

Do you know when I said almost as good?

[fresh mirage]

except llama

[hoary elbow]

I’m not saying it’s just as good or is better

I’m saying almost

[stray aspen]

Gpt 5 is astronomical units better than mini max m2

[hoary elbow]

I use M2 for coding

I’m gonna test out GLM

[knotty fable]

We're talking astronomical units here? 😹

[hoary elbow]

I have level 20 right now I’m trying to finish the battle pass

In Fortnite, I have level 20 right now I’m trying to finish the battle pass

[stray aspen]

Wdym

Why you talking about fortnite

Lol

[hoary elbow]

Cause I’m playing battle Royale right now

[surreal creek]

real

[knotty fable]

I wonder too - have set up the AI to play the game?
Seen some examples on that poppin up in recommendation on YT - never watch any though.

[surreal creek]

this new season is so good

u a real one Bruce

we should play sometime

[hoary elbow]

Right now it’s a mini season

Yeah

I died in battle Royale

Just now

[stray aspen]

Is fortnite still alive in 2025

[surreal creek]

Yes

[stray aspen]

That's crazy

[surreal creek]

they still put so much effort into the game

It’s constantly changing and staying fresh

[hoary elbow]

Can some people think it’s fun even when it dies some people still think the game would be fun and still play it

[stray aspen]

No it sucks

[knotty fable]

Some games never dies.

[stray aspen]

It got boring

[hoary elbow]

But it won’t die for a long time I don’t think

[stray aspen]

I played from season 2 to season 7

And fortnite also turned cod into absolute garbage

Definitely had some influence from fortnite tho

Mostly stupid skins and batyle royale

[stray aspen]

And unnecessary colabs

[fresh mirage]

I used like treyarch out of the other studios making cod games

Now it’s kinda meh

[echo aurora]

As much as I enjoy conversations about gaming going to ask that we refocuss conversations back to something AI related.

[stray aspen]

Lol

[fresh mirage]

ofc ofc thank u friend :)

[stray aspen]

Alrighty

[fresh mirage]

Now to get back into topic

Lithiumflow

[echo aurora]

Lol not here sorry to say.

[fresh mirage]

my beloved

[stray aspen]

It's dead

Still waiting for its return

[fresh mirage]

Don’t need to remind me…

[stray aspen]

I had so much fun with lithiumflow

[echo aurora]

Thank you all for shifting topics, I do appreciate it

[fresh mirage]

I had a relationship with lithiumflow… I was its friend, its lover…

[hoary elbow]

Sadly, lithium flow isn’t out completely right now

If it came out completely, that would be my main model

[fresh mirage]

Lithiumflow was everything to me…

Now it’s dead

Omg lmao

[hoary elbow]

Can you bring back lithium flow?

Please

[fresh mirage]

Damn bro really be thinking I want AI waifus

[stray aspen]

No

Lmao

[fresh mirage]

Wouldn’t be surprised tbh

The amount of degens I’ve seen in 4chan is kinda nuts

Oh yeah I think I saw a reddit post about that

[stray aspen]

Isn't elon musk supporting ai relationships with grok

Lol

[hollow ivy]

<@&1349916362595635286> i think we have another scam problem there

[knotty fable]

<@&1349916362595635286> remove plox #general message

[stray aspen]

Yeah get rid of that guy

[fresh mirage]

mods

Twist his balls

LMAO

[hollow ivy]

what did you create with it?

[echo aurora]

Wish I had that kind of power

Also thanks for the mod flag @hollow ivy @knotty fable

[stray aspen]

A lot.of websites and a little game about operating a ac130 plane

I had great results

[drowsy gale]

Cool

[hollow ivy]

better than Claude-4.5-Sonnet-thinking?

[knotty fable]

Did a reverse search for one image of mine that I misplaced the original.
Droogle & Bung failed. Yandex found it! 😸

[stray aspen]

Yes lol

Miles better

[hollow ivy]

so, in two months, the new era could finally start..

[hoary elbow]

I got my first Fortnite battle Royale in this season aside from training, creative maps, and blitz Royale

Victory Royale

Battle Royale victory Royale

[jade egret]

wdym

[dry walrus]

Hi

[echo aurora]

hello 👋

[stray mortar]

does anyone know any ai models like minimax m2 where u just have to give it a promot and it does everything for u u domt eben have to deplout it or anything

[dry walrus]

How do I find the videos I had generated

[fiery gull]

use the search from discord

[hollow ivy]

for coding, Claude-4.5-Sonnet-Thinking is currently the best model (just prompt it; its best language is Python)

[dry walrus]

I tried searching for my username. What else can I reference

[echo aurora]

You can either:

1. Open up the Direct Messages with the bot, they'll direct you from there.
2. The search bar, but you need to use the mentions function.

[golden ocean]

steal a brainrot

[keen beacon]

Hmm

I wounder what the correlation is between access vs performance on the benchmarks

[burnt sinew]

has anyone tried a/b testign today on ai studio/

[inner gate]

Is there an open source leaderboard

[carmine valve]

trying out video generators

[keen beacon]

That’s pretty sick, bro. I’m not gonna lie.

[wispy sequoia]

Servers might be down

[keen beacon]

Arena 1?

[wispy sequoia]

I’m getting nu response

[wispy sequoia]

Yeah

[keen beacon]

For which model

[wispy sequoia]

Opus 4.1

Anthropic

[keen beacon]

That’s probably a rate limit

[wispy sequoia]

Fudge, okay

[keen beacon]

I think I saw a screenshot here earlier

I think it’s every six hours or something. I’m not sure maybe 12.

Claude Anthropic don’t got that kind of open AI money

What the hell am I talking about six hours or 12 hours wtf lol 😂 50 minutes, bro?

Don’t listen to me, dude

[wispy sequoia]

It’s okay

You’re good 😭

[grand cliff]

How far does the bot's information go?

I use 4o and it says June 2024. But a few times, It referenced September 3, 2025.

[keen beacon]

Hard to say without the system prompt and if it’s being routed or not

4o mini low key best open ai version imo

[grand cliff]

I use 4o because it has soul.

Not soul soul. But gives off the humany feel.

Knows my tone. Knows what I meant.

[keen beacon]

No, not yet it’s one of my favorite models if not my most favorite

Such an under appreciated model

It was good all around, nibble and fast all round all star ⭐️

[cold lion]

my new found obsession -A.I. Video creation

[keen beacon]

Yeah it’s addictive 🙈

[unkempt elk]

same thing happened to me

[keen beacon]

It just means you hit a limit for an hour

[unkempt elk]

yea

the UI is meesed up rn
is it just me or others have same issue
scrolling has problem and the gap between the copy, liek and dislike button and the ai's response is too much

issue is because of this section of buttons

[keen beacon]

Not sure =/ I don’t use the ais

Only image gen models lol

[opal drum]

lithiumflow is out now? every time my prompt is only able to summon willow and another random model.

[drifting crow]

https://www.euronews.com/next/2025/11/01/polish-to-be-the-most-effective-language-for-prompting-ai-new-study-reveals

[dusky olive]

Is there a daily limit on how much you can do in the video arena?

i am making a game and the ai just stopped and keeps erroring

Something went wrong while generating the response. Please try again.<-- it keeps saying this

[icy plank]

best image gen for manga?

[keen beacon]

Seed dream

[keen beacon]

Very interesting read thank you for sharing that. Interesting to see Russian and Ukrainian up there too.

https://youtu.be/nixr8ZNJLVQ?si=JMlqIuRD-dI6gry-

[rustic pelican]

same thing happened to me

[unkempt elk]

ohkk

[keen beacon]

https://youtu.be/9MB0dm0vv2U?si=2wcqWo00Nu16OLT4

Ai video next

[solar karma]

is anyone else's UI glitching?

[keen beacon]

Is everyone trying to use Claude ? ;P

[rustic pelican]

yo what's going on with the UI

[fresh mirage]

Been out of llmarena for quite a while