#general
1 messages · Page 317 of 1
What do you use 4.6 for?
what's yall opinion on muse spark? is it good at coding?
learning with pdf
i want good model with powerfull context understanding
Gemini 3.1 Pro or DeepSeek-R1 is top tier
I recommend these
Although Gemini is gone and im not sure if deepseek r1 is in it
Its good but not good enough to make you shocked
Claude 4.6 programs very poorly compared to mimo_v2_pro.
Claude opus 4.6 can catch tiny details while mimo focuses on more useful ones
But id say theyre pretty equal
Yeah, but Claude is a bit weak when it comes to that.
Hi yall
Yo
what you mean claude is on the top in llm arena
Every ai is nerfed in claude
Lmarena makes it nerfed
It cuts the tokens down
So the stats you see aren't with in lmarena
They're within the actual apps
For example the website might nerf tokens, which makes ais start trying to "save" them
But idk about the rest
@quick jackal
That makes sense....
<@&1349916362595635286> git em
@light siren
hi
What do u think is this possible?
Ah yes hi
No
You both work together?
yes
Nice
Yes but he is better at logical coding and stuff
Aka backend
But I am good at frontend
And yet enhance prompt backend is made by me-
thats why i said mainly
A
Whatever
Guy is saying some latex or whatever markdown
Is annoying
And add option to disable it
You would understand better what he is talking sbout
Go ahead tell him about ur problem
oh yea the user markdown
I planned to actually remove it
Wait he meant user markdown?
pretty sure
since those have auto markdown stuff I think they're called latex
Sooo it's possible
Literally get user prompt and convert html tags to text
not as easy lmao
Okay even easier method
Literally copy user message
And replace whole user promot
With it
The copy button thing
if it were that easy the bottom copy option would have been able to copy lists
You can resend the same message without rewrite it
You can use it's own js
(in case you're using manisfest v2)
Wait what
yes
It can't?
It's a common problem
All you have to do is ask the AI to give you a very complex regex
if u type
1.
2.
3.
It wont copy that
feel free to like improve it or smthn
Only bottom copy button or copy button at all?
only bottom copy button
Dude why did you recreate copy button when u can just move original to the desired place
i cant thats the thing
I kept trying
Or
Getting button Id or whatever the hell js looks for
And making recreated button work as the original one
Using that id
Then how tf it even works
idk
Oh
I get it
React js creates html tags locally
Aka inside js not html
And that way it doesn't need id to detect clicks
Modification of js file won't work cuz it's react it's obfuscated....
Resend the same request
Remember that failed messages are logged, and the IA can use the token from that failed message
Therefore, if you simply copy the message and resend it, you would be wasting tokens
It’s best to resend the same request so that it is sent to the same log level instead of creating a new duplicate message log entry
Example: #general message
Looks like gotta recreate custom copy button properly
Yeah I saw that bug
it's not a request it's a frontend function
Idk how it happens
Usually in used chats
Lowkey prompt bar decides sometimes right is left and left is right
Whatever I do
Idk how to fix
But it's not that annoying ig?
nvm I remember now, I fixed it already
You're wrong about that
I used to think so, but both of them (fronted/backend) are to blame
what, copying messages
?
those are frontend run using js
You didn't understand me x)
wdym
You're saying that if the message gets stuck (endless generating), you just copy it and send it again, right?
what, when
no??
it was never mentioned in our convo
Astware said something like that
if it's endless generating we have a button to unlock the chat so you can send another message
does this answer
why he said it
guys is there a way to make video generation longer like to make videos that are 30 secs not 4-5
That's when I step in and say this: #general message
You'll see something like this:
You: Hi
IA: <endless generating>
You: Hi
IA: <message>
If you refresh the page, you'll see this
To avoid wasting tokens, forward the request instead of resending the message
hey guys sup? question, any news of gpt image 2?
whats wrong with that
forward the request?
how am i meant to do that
Manifest v2 -> Rewrite the JS file and modify its logic
I did it on another site to have unlimited messages with an uncensored AI
This one isn't much different and is just as easy
Obviously, I'm not going to tell you how, but if you know how to program an extension, you'll be able to do it
Look
U probably don't know
But
React js
And then on top of react js
Next JS
Which lmarena uses
Has code splitting
And
Code obfuscation
And a compiler for next js code to vanilla js code
Basically
Making
Modifications of js files in frontend
Impossible
We can only modify html and add our own js injections
Fricking auto correct
Second of all
Manifest v2 is depreciated
Here's more details about manifest v2
Yeah, it's a real struggle when the code looks like that, but I managed to do it
Luckily, I only had to edit two files on that page, but it was spaghetti code and a bunch of other messes
It's possible, believe me
But it will take many days to understand the code, review the list of processes for each JS file, and then figure out what to edit
So?
I built my extension using manifest v2 because there aren't many things you can do with manifest v3
isn't react js always generating a new code in build time-
How the hell does manifest v2 even work for u-
And you can set up a system to detect changes and edit them
Brave supports manifest v2, for google, you'll need to make a few adjustments for the extension to work
what site did you do like venice or smt
LMArena?
I don't think thats possible still
An intelligent AI that I can't mention
its weird
if you cant mention the ai how come you prove you did it
It really is, trust me
When I have enough time, I'll do it and show you how it works
Ok I will wait
It will be a massive security vulnerability breakthrough for next js developers aka vercel
If it is actually possible
I kinda don't care about kimi-
yeah what even is kimi's selling point???
A screenshot showing a bit of reverse engineering
they must be the best at something?
there isnt one
open source?
why is my name there
😭
To censor part of the domain name x)
idk
Wait... How tf did u-
even models like everything mistral makes, i never get the reason people use them, when their are very similar priced alternatives that perform so much better
it has to be a free ai website with usage limits but lets anonymous users use it
probably wanna be involved
maybe your rotating session cookies
i mostly judge an AI model on its core limits to price. Thats also why i hate GPT.
@light siren this guy might be able to help us fix even more issues if he can actually reverse engineer jsx of react
So I reverse-engineered it to get unlimited messages, because once you use up your quota of 10000 tokens, they ask you to pay
Guy
Nice
but isnt it in the backend
hire him idk
It's a she
I can't type your name on mobile
she!?!?
It's a symbol
a girl capable of reversing jsx!?!??!?!?
because like it has to be a horrible coded vibecoded slop site because i dont think they are adding the usage limit to the frontend
Yes and no
They have vulnerabilities that haven't been fixed
Did u forget dragon
This user might be her relative
It's an excellent AI, 100 times better than Venice, and it has a user friendly interface
Did u forget that we yes WE are broke
could you please tell what tools did you use to make it
just the tool
vs code x)
fair enough
And only JS Vanilla
hire them for 0$
Good idea
wat else can we even do
Likkeeee
🤔❓
prob not but claude narrowed it down to these but if its horrible coded then its prob not Narrowed down candidates:
Poe — has token limits, strong models, Next.js
Mistral Le Chat — free tier with limits
Groq — free tier, fast, good UI
Chutes.ai or similar API wrapper sites
but i dont think is it
its this female friend we have dw about it
what does dw mean
Dallas Wonder
let me change my pf
Dw is Death Wish
what vince said
DeadWalker
Walking Dead
dw is hotel
@wary nacelle
I would fix the endless generating issue so that the same request is resubmitted without creating duplicate messages
But right now I'm working on a terrible vulnerability I discovered on *** a few hours ago
What
Nah she is definitely dragon undercover
could definitely use her
You've got me curious 🤔
invoking spicy fried chicken to ban the guy
*partnership
Uhhh I can't send u friend request to make a group chat
@undone saffron are you a cybersecurity person or like just a bug hunter or smt
Isnt that obvious
@Moderator involves him
bro idk
(thanks for flag)
hi pineapple
i was definitely not talking about slavery
@wary nacelle I made ur raw message thing
it works
now imm watch youtube or smthn
bruh im bored
Just someone who knows how to use a computer for entertainment and personal use
In addition to working on a high level project for myself
That's why I use arena
So I don't have to write 10 billion lines of code instead of writing them myself
From 4 years to 4 months
okay
my new guess
with claudes help
My new best guess is some mid-sized AI wrapper site that pays for API access to a premium model and offers a free tier. Something like Merlin, Typing Mind, or a similar lesser-known platform.
wanna work for us
u will not be paid
1800 colonial brazil
Use?
I see
For that, use arena instead
That's not how u ask for help from ppl 😭
@undone saffron im just asking one thing: what ai model does the hacked website use
just that
As step 1
Dkfjfjdndjd
@undone saffron I'm soo sorry for my friend's behavior!!
please help us with our extension!!
That's more like it
I did not know that extension existed
I mean
what am i saying
that model
I did not know that model existed
I did
ahem.
May you please help us?
But idk why it exists
lets remove it fr
By that logic we would only leave latest gpt, Claude and Gemini model-
real
ts chat so boring
An AI built from scratch with constantly updated information
One that doesn't use any existing models and is private (with the exception of free access to certain messages)
I'm so busy for now
But if you're looking for something exciting, here's another fact: #general message
i would love to have opus back here just to test it with api, since with the sub on claude.ai seems like brainwashed
i dont know how that helps me
Aren't you excited to hear that there's a way to send as much text as you want in a single message?
yes
Davi is an exemplary person
Will they come back in the end?
wow you got my name right
we dont know
They will reintroduce it in the future once the arena's finances become stable and a daily usage limit has been established
I hope it will be soon 🙏
@light siren@wary nacelle
By the way, it would be a good idea for you to talk to pineaple and show him the solutions you have on the website
Maybe that way they can fix a lot of the bugs and issues that exist, and maybe you’ll end up working as arena dev
If LMArena actually had income based on users that would happen
Not based on training companies ais using users
But let's see
¯_(ツ)_/¯
@light siren what do u think
What the hell is meta spark
Meta muse spark
already did
they dont wanna work with us
They have different departments and they can't just hire ppl cuz they like them
when is meta releasing the api
Nice
I'm using the actual app
Still kinda don't know it
But worst
What y'all expected from Greedy Facebook owning company that literally uses credit cards for account verifications
And sells data
It doesn't sell direct data lol
It has the same business model as Google
It keeps your data and uses it to send ads.
Only that?
Are you sure?
Because your data is more valuable to They're advertisement part of the company then selling it
¯_(ツ)_/¯
Probably used it for their ai models to
I'm surprised Google can't make a really good ai with the amount of user data they have
And general data all together
They probably hiding something
They did tho
I would say Google is the most known company worldwide
But only for design
Like better then mythos
Not coding
Anthropic probably has less data then Google to train on
yet they are still the goat
I don't think Google is trying
Yes Claude codes better than any ai
prob
Because they need to save costs
But it's not peak at design like Gemini yet
Mostly yes
I main deepseek for daily usage but it is terrible at coding
I've never been a fan of meta
Gemini is a all in one ai
But never let it code
It's meant to be able to code, help users daily questions, do tasks for users all in 1 model
I do let it code
hows that nice
no chance
I do too but later I spend 9 hours fixing other stuff while just making a popup loading design
Sorry, I didn't see the “don't”
Just give it the errors you see and it should fix it
Well who said it writes good code
Well too bad Claude writes better
Make Claude do UI like Gemini and Gemini is cooked
can i buy tokens for my arena?
Hmm I might put it to the test
Unfortunately I can only use sonnet cuz I ain't paying £20 a month for LIMITED ai
How exactly
Just give it random prompts to make html files and see which one does better
Like who has less errors
And who can follow the prompt best
Errors is not the point here
It literally is
It may have no errors but when other dude tries to modify smth whole thing breaks
Just bcuz Gemini got lazy
That's just called not understanding the code
Depends on situation
Like changing idk popup name shouldn't break everything else
The only fix to infinite generation is
Start a new Chat (Use slow model so it doesn't respond fast),
Then, switch to the chat that had the infinite generating bug, and it will show the square button if it wasn't there because of infinite generation.
This works btw
Hmm I'll put this to the test too
It was the minor example
Obviously Gemini wouldn't do smth that obvious
Just I use complex frameworks
Like when saying make this website have localization
With translations
And specifically tell it before some prompts that there is translations . Json file
It might put a sneaky another translations array in the code
When fixing smth
To show that it works
But it shouldn't work THAT way
Ohh
I'm just testing meta AI rn
On html code
K
Why does it make websites
Lmao
Like what I mean is you can preview it in a website URL
That they made
I escaped by exploiting a bug in the javascript and got the square back
Just kidding I probably did the same method
Venice is unfiltered ain't it?
Lol
True
The one that I use is very smart and fast
@pine veldt
yo i cant log in
Clear your browser data and try again
now working dont worry : )
what ai do you use?
this is sad
glm 5.1 got the same problem 5 had
after a while it starts to hallucinate and get lost over the text
muse spark is much better
Lmao Claude L
Yeah really looking forward to it fully dropping, it was in arena as a stealth model last week
I did this on ChatGPT 1.5
no way this is real right?
Okay
Imagine what will happen in 2030
I imagine in 2030 we don't know what will happen with images, like deepfakes, AI videos, etc. You'll say "It's too AI," we won't understand anything anymore
found a similar site after 4 days, https://beta.ish.chat
Meta ai is very bad bruh it's so censored
I still need Gemini 3.1!
damn that looks insane actually
especially the YouTube one I can’t tell if it’s ai generated or actually a screenshot
Ngl that opus 4.6 feels the most limiting ever
claude
Its the same model just optimized to be quicker
Does anyone know how much token limits/chat Arena has right now?
No clue
Yeah sorry to say there isn't good tracking setup for that atm, we are in the progress of building one though
Bruh
Why is gemini more censored now??
It used to be uncensored on ai studio site
Now it will censor anything
As if it got more sensible
@flat prairie
Why would you want to do that?
You're asking a question that isnt related to the site
Nobody here will know, likely
chat is it normal to have internal organs?
Idk, you feel comfortable with pink brain?
did you try with a vpn?
Why were the Gemini 3.1 preview and claude opus removed, and will they be added back?
Guys
real
I wonder why you have the icon of new in this server
It’s kind of weird that people who study knowledge don’t talk much about stupidity. You would think they would, because stupidity is about not knowing things or getting things wrong. If their job is to explain what it means to know something, then they should also look at what happens when knowing fails.
I can’t see it on my end
🤷♀️
Ask ur ai these questions.
Do you have a good sense or good judgement?
Do you understand the possible harmful consequences of your actions?
Do you feel any fear because of you don’t realize the consequences?
Do you have any clear goals or actions ?
Bet I bet you you can’t even define stupid
The problem is, it’s not that simple. Knowing something might follow a clear set of rules, but getting things wrong can happen in many different ways. People can be wrong because they are missing information, because they made a mistake, because they believe something false, or because their thinking is biased or confused.
So the main idea is that stupidity isn’t just one simple thing. It’s a mix of many different ways that thinking and knowing can go wrong, and that makes it harder to study than just studying knowledge itself.
Whar
Yeah, that’s why I copied it from lol
While challenging, NPD can be treated through long-term psychotherapy, specifically talk therapy designed to help the individual understand their emotions and behaviors.
I was reading about Yann Lacun and what he said about Claude Mythos
He reads "Yuno Ball"
GLM 5.1 Free unlimited no limits is absolutely mastery
Twitter is the best thing to ever happen LMAO
It’s just an interesting subject, stupidity itself
And then I came across that paper 🤣
Well its bound to happen if people use AI more and more ykwim
True but era goes on
He knows some ball
Well, that’s why I brought up that paper and ask the questions I asked
It’s funny too that he said stupid that it’s stupid to ask that and he’s correct
More dangerous than the hub
I just didn’t wanna say anything yet, but you got some clown out ruined the set up
🙁
Imagine losing thinking mode
Yeah
Well, that makes me wonder if it’s not intelligent then what is it?
The only thing that comes to mind is stupidity
Its teaching to be better if you pay attention
Well, that’s what this paper is kind of saying about the knowledge itself
Then it will be a study material instead of generation
they're not going to use AI if they don't have low literacy lol
Indeed, because its the opposite of intelligence
The point of AI is to assist or do the work you ask it to do, which means you to begin with should already have the idea
I believe that non creative people cant use AI for good
But they can learn from AI
Create other ideas
I mean generation means that instead we ask complete solution, why not study the solution?
Wdym by that though?
Basically generation means that we don't discover anything but trust
that can be true, but AI did something that makes me understand some stuff a lot more than school ever did
it's just learning material to me
That’s not my blog
or a reviewer maker
gpt better than claude btw
If anybody else is interested in, here’s another interesting book on stupidity
is that downloadable to Z-Lib? (piracy based)
Whats this muse spark model mogging everyone except claude, hadnt seen it before
No yea 100% me too, the more you use it the more you will learn
Ahhh
Then yea I understand, why not study solution
true
No, that’s the Internet archive lol
yeah i understand, sometimes Internet archive has paywall 😭
which is why i asked z lib
Oh yes
To be honest with you, I really got my interest on the subject is this
The first 15 minutes of it made me really think
Especially about all the information and act we have to knowledge
And yet we’re still so dumb lol
Man GLM 5.1 takes a long ass time but its relatively cheap
It is true that being smart has its limits, you can be smart on one aspect or one specific thing, but be dumb or stupid in another aspect
unless generally stupid to all of things
just like people who vote for stupid or manipulative politicians
muse spark has been enshitified already?
yep
yeah figured
it's not as good as, it was during the release
I didn’t try it. I just figured that they would at least give it more time.
That’s crazy dude a week
well it's meta, i never expected much
like i'm good (although i would just consider myself as an apprentice rather than an expert) at liberal arts, but i'm bad at mathematical or anything related to very deep critical thinking or anything that involes complex solving.
well eh meta is at least better than microslop
The whole reason I’ve been so stupid thing I also is dude I don’t think these models are really smart or intelligent. I just think they memorize a bunch of 💩
Dyscalculia doesn't exist in my country, but it's just that Math or Physics makes me mad.
I don’t know at least I don’t see it
The more I interact the more like I get disillusioned with it
at the same time, i took a college degree, that heavily involves both.. like wtf am i doing in here?
llms works on guessing your next token after being feed a ton of data. It tries to guess what you are gonna say next. And searches its data, and gives the thing that seems like most relevant
but when you increase the computing power, by a lot it has some extraordinary changes, that's why current labs keeps on increasing the compute power
Yeah
That’s what I was thinking
But I don’t think your average user is gonna get access to that kind of compute lol
GPT app sucks
well eh for llms, your avg user is getting good enough results. For agents they don't require much either way
Well, I looked into it also yesterday
Google is known for enshitifying their products
I think you gotta pay for it to have good AI
I think that we would do that is through the API
look what they did to ai studio
I think that the app versions are just gonna give you crap
Yeah, the censorship has been getting horrible guys and horrible
I’m telling you it’s ruining these damn things
I only use it to chat
Your brave soul
I’m starting to really dislike AI
Not because of the tools, but because of the business behind them
And censorship is the big reason
I think they’re good for like utility
it's meant for profitibility as of now
they understood, that people needed AI for something they're not capable of... and that is what the market is for.
then stocks fall down
Absolutely man
But thats why I've tried going more local
Whats crazy to me is that people are surprised that if AI takes jobs over the economy would collapse
Because obviously it would
If people stop working and money stops circulating what happens?
Economy plumets
And really if that were to happen it wouldn't matter by then
All labor jobs would be taken over and we wouldn't need money
I don’t know I’ve written other people recently that it says that I can’t barely do 2% of people’s jobs that are currently
Video generation is removed from the site???
has anyone try Manus before?
Yes for now
Eventually they will take over jobs
Would yall take GLM 5.1 or Qwen 3.6 Plus
If only Opus exist. i would have been doing this prompt in Arena
Can someone recommend me the best ai video explainer?
Claude sonnet can do transcripts, gemini can do full YouTube. But only from their own websites and apps
3.6 is more reliable, glm is better strictly for coding
Camera slowly dollies in from a wide shot toward a medium close-up. Subtle natural movements: one person adjusts their position, a child leans on a parent, soft breathing, slight smiles. The TV light flickers gently across their faces. Curtains move slightly as if from a soft breeze.
Thank you
Why they removed gpt 5.4 and opus 4.6
@echo aurora how can I dm you?
If you send @oak python a DM that's the same as DMing me
oh okay
Is verifying needed?
I'm enjoying this woo hoo
I was reading a research paper on how researchers used to trick the models
So clever
this what AI should be use for, not for answers in exams. but actual reviewers or teaching guide
That’s if we are to believe that AI is actually capable of being intelligent
after all, the sources of the reviewer is me providing my own screenshots of a lecture
Nope
I’m assuming that this human made content ?
Have you guys seen the Claude propaganda?
About mythos ?
yep, then with the help of AI to OCR it, then turn it into LaTex Code (Since i'm lazy creating a reviewer) the results is the picture
that it's good? like i'm gonna believe that lol
Microsoft is very obvious. they rent their servers...
is this GPT 2019 all over again lol
dangerous in what way? dangerous that it's gonna make us believe it's too good? like come on, nobody's gonna believe that except madaforkers that can be convinced by propaganda
what happened to the web?
wdym?
I sent some messages, could you take a look?
i was using it normally suddenly its failed to geneared after reload web it just show
This page was lost in training.
Something unexpected happened. Please try reloading the page.
mostly it only show web icon and quote with trace id
is it down
data?
Browsing browsing data
so basically it lost memory
happened to me too lol
Sorry, doesn’t typo
not a cache?
I just do the whole browsing data, but you could do that too. Also see if that helps.
Would you guys like gpt 5.4 back?
entire browisng history is crazy tbh
Yeah it is
high yes, not normal
Why not normal?
Bro
Nanak banana has been nerfed so hard
very unsustainble at it's finest
I don’t wanna get worked up but
How many spelling errors does your messages have
It won't be affordable to them
i feel like claude got nerfed at all
and glm does it better rn
Look this was my reference image
I asked her to replace the pickle
It gives me this, which is fine, I guess
It's good
Even great
But then look at this
yo guys muse spark hella under-rated
Lmao wtf
So clearly knows what it did wrong, but it’s still spit out the wrong image back to me and gave me my reference image like it fixed
I swear it’s on purpose, dude
They purposely made the app version crappy
Here one more example
Man banana three was so peak until December 2025
I’m telling you though that was peak until December
It was so good dude
Man, I miss those days
That was two months ago lol
Your pfp is crazy
it seems that the AI doesn't quite understand the principles of perspective
The conclusion I came to you can’t use the app. You have to do the API or pay for it. That’s when you get better quality.
Otherwise, you’re gonna keep getting crappy results and it’s not as filtered with the API
A warm, cozy family gathering in a bright modern living room, three generations sitting together on a soft beige couch, smiling and laughing naturally. A young couple, two children, and elderly grandparents are sharing tea
It’s honestly the stupid filters that are messing a lot of this up
I hate to say it, but it really is and the prompt also is because different models have different standards of safety
It just makes the whole industry a mess because one model can’t generate another one. Can’t
It’s honestly ridiculous and absurd
Only 3 messages
Which model
Crazy
I hate when the AI tries to give me suggestions or changes my prompts
If you can’t do the task, is she just shut up instead of giving me ridiculously useless suggestions lol
animate this
A warm, cozy family gathering in a bright modern living room, three generations sitting together on a soft beige couch, smiling and laughing naturally. A young couple, two children, and elderly grandparents are sharing tea
thats fun man
“A tender, emotional moment of an elderly grandmother with short white hair gently holding a sleeping young child on a soft blue couch. The child is resting peacefully on her chest, eyes closed, wrapped in a light blanket. The grandmother smiles softly with warmth and love. Cozy indoor living room setting with warm natural light coming through a window, soft shadows, neutral tones, minimalist home interior. Close-up composition, intimate framing, focus on facial expressions and connection. Ultra-realistic skin texture, fine details, cinematic lighting, shallow depth of field, 50mm lens, soft colour grading, warm tones, lifestyle photography, emotional storytelling, 4K resolution.”
Video Script: One of the bullies reaches out to grab Aruna's hand, she sharply pulls away with an angry defensive posture, the camera focusing on her sharp reaction.
Yoo Gemini 3-flash is gone too??
The child is resting peacefully on her chest, eyes closed, wrapped in a light blanket. The grandmother smiles softly with warmth and love. Cozy indoor living room setting with warm natural light coming through a window, soft shadows, neutral tones, minimalist home interior.
oh yea
what
A vertical 9:16 cinematic short video. A plain slightly rough wall fills the frame. Within 0.5 seconds, a strong human hand suddenly slams onto the wall with force. At the exact moment of impact, a bold red “BEWARE” stamp appears instantly under the hand, with a wet ink spread effect and rough edges. Add a quick camera shake and micro dust particles dispersing from the wall. Lighting is dramatic with high contrast shadows. The action is fast but includes a slight slow-motion effect at the moment of impact. The overall mood is intense, shocking, and attention-grabbing. Ultra realistic, high detail, 4K quality.
🔊 Sound Effect Prompt (Must Use)
Loud, punchy “THUD + STAMP” sound synced perfectly with the hand impact, followed by a short echo reverb. Add a low bass hit for cinematic impact.
⏱️ Duration & Flow (Shorts Optimized)
Total duration: 2–4 seconds
0–0.5 sec: Empty wall (build curiosity)
0.5–1 sec: Hand slam impact 💥
1–2 sec: “BEWARE” visible (hold frame)
Optional 2–3 sec: slight zoom-in or fade out
When will we be getting gekmeni 3.1 pro back
I miss that shits so much
It's my therapist
just pay 20$ bro
Petition to add Qwen 3.5 9b to the leaderboard.
the bottom end of the pareto chart would greatly benefit from its presence
Does it only happen to me?
niceee
File processing and code writing are constantly improving in #Code
Is it working poorly for you too? I'm writing a request, it takes a long time to complete, and sometimes it just doesn't do it, or even makes an error. And who knows why some neural networks disappeared?
The site went down
<@&1349916362595635286> Can u help me with that? I cant find any claude opus model in website
They just took it off a week ago
Why?
Ohh. Lots of models missing too, like gemini3.1search
This issue only happen on chrome, edge is fine
Happens on brave also
me when also getting captcha now😔
What is going on? 🤨
Prism?!
Why were the Gemini 3.1 preview and claude opus removed, and will they be added back?
too expensive, unlikely for now
What's better for the model to write Prompt for Nano Banana?
3
5
1
gpt 5.4 mini high
This makes me laugh so much
why is direct so sh** now
like they are offering models from a year ago
side by side only valid mode ow
tomorrow gpt image 2
Image 2 already out isnt it
do "generate a selfie of Sam altman"
I wanna see if u rlly do
Cuz a Sam Altman selfie is the test to see if it's gpt image 2 or 1.5
You can also find out based on the format.
If it's 21:9, it's definitely GPT image 2
@undone saffron
I think it looks real, just AI images always have that like odd glamour that it adds to the picture and it makes it look too good to be taken by a phone
I think it is
Do you want me to improve?
Well if you say "(Whatever prompt you had before) and make the image look realistic from an iPhone camera" it will do it
See now its weird, its too pixelated if you knwo what I mean
Yes, the graininess, but I think it's because I said "iPhone 4, no improvement".
Yea lol
Out of these 3 Qwen 3.6 plus
Roleplay or Scenario wise, Qwen for now
unless Sonnet gets fixed in Arena (it's buggy right now for me)
i guess, 3.1 pro on AI Studio
Not really free model, I mean those which we can use for free in their respective platforms/apps
albeit, i ain't gonna lie, i settled on using a paid opus 4.6 router
3.1 pro is free in ai studio
is there a way to do ploting inside arena like how claude does the visual and gemini
i'm loving itt
Gpt image 2?
How r yall getting gpt image 2
what happen
@echo aurora I can't create images in LMArena. Every time I click the generate button, an endless Captcha appears and then it gives an error. They need to remove the Captchas; they're useless now with so much AI on the internet
lol, facepalm
i totally forgot about good ole' GPT ^^
is gpt-5.4 good for roleplaying and interactive fiction, etc?
gpt kinda fell behind gemini and claude nowadays
... ok, so if you guys want to vote for GPT, please use the last option (other)
Option #10 is for ChatGPT
just go for gemini 3.1 pro or opus 4.6
i hope that will change with gpt-5.5 (aka "Spud")
that's claude 😭
session limit (in my own theory)
since i'm experiencing that problem
whenever it's complex task]
#Video Arena
ai models in arenaai are getting deleted one by one is the website closing ?
I need to do my masters thesis and I need literatures to review. Help me to find a way out
idk i gave him not hard task
random but I noticed the pattern of ur reviewers, are u a med student?
Depends on how big the prompt is really
GLM Certainly took its time
@echo aurora can you check dm?
I think they don't want those model reviews or the cost has become high because of the abuser that's why they are removing frontier models
Feedback*
Did opus peak or not
10
12
1
yes
actually they explained it in #announcements its just expensive but they're still available in battle mode to collect leaderboard data
Do u use third party providers?
Do you have access to the mod mail?
no access to mod mail. I do have a claude pro sub though.
just the 24$ a month tier
Ah yes the trial tier
Free is 1 prompt tier
Where did you get it
Fireworks AI + OpenCode
Free GLM
5.1
And Qwen 3.6
Plus
Bruh Ive seen that but how
I have made a temp account with credit card linked in there I don't see free GLM 5.1
You gotta generate an API key in fireworks AI
Then put that API key into opencode
That's it?
Yes
And this means I already have this pass right?
Now it has a $50 limit so you just switch accounts when thats hit
I dont see that ngl
Dawg I thought you need to get credits for that holy I would've ignored this
It said all passes claimed
LMAO
Bro what are u talking about
you asked if I had access to mod mail and if I used third party providers.
I'm not a mod so
What's the base url and GLM 5.1 model id?
Also the qwen 3.6 plus id too
accounts/fireworks/models/glm-5p1 idk qwen
I literally just use /models
Why the Kirk has arenas usage for sonnet decreased so much
And base url?
What is "Math" and "Mathematics"
Why the Kirk has arenas usage for sonnet decreased so much
No clue
For API
what doe this mean for the leaderboard?!
what categories even mean? how they form
Oh you can already auth login using fireworks
Yes
Uhh I think all passes are claimed
So basically I make a account?
Then make API key that's it?
Account add fake credit card to it then api yes
Well I guess the method got patched
Nope because I'm still using it
And a friend just got it
My new made accounts always get suspended
Stfu
<@&1349916362595635286>
mines exceed 80k characters (more than 1000 words) for each of my prompt
not really a med student
Thats why then
so how leaderboard categories form?
although Sometimes it's only 200 words, but Sonnet crashes all the time
Per usual it crashes whenever it's latex or html ones
Yea anthropic API sucks with this site
guys whats the best ai model for roblox lua coding
Over the weekend I saw Gemini 3.1 pro come back for a blink but now it's gone again and 2.5 pro also doesn't seem to be working. Anyone else experiencing? Any ideas of what's going on?
i have wished Ph banned Roblox, a dream about to be come true
brudda what
And it didn't happen.. For 12 Years, of playing it, in its current state, It's nuts
captchas suck
bro i just wanna know the best ai model for the coding
Claude and Codex
I'm sick of this damn CAPTCHA every second
although some use Gemini's output
I can't even get through it anymore.
does arena still offer the best ones or did they remove them becouse they costed to much
only sonnet
and codex api currently sucks
@echo aurora Must explain what happen...but he wasn't online whole day
so i should lowk buy claude jusst from their website?
Help me, people, I have a captcha for every request, and now I can't pass it at all. It says Something went wrong while generating the response. Please try again.
Trace ID: b7917894-e081
and then would the pro still offer the best coding model
Me too, gemini 2.5 is giving me errors all the time
i try to generate an image, pops a captcha, i solve it correctly after 10 "failed" attempts and asks me for login WHEREAS I AM ALREADY LOGGED IN
Also. I had noticed, today gemini models(gemini 2.5 pro and gemini 3 flash) generate EXTREMELY slowly...
And them, this 'something wromg' thing happened, for about 2 hours already
you have big difference beetween opus 4.5 and opus 4.6 ?
Is the arena down? I haven't been able to access it for over half an hour.
Trace id :c1c09d10-28d3
Gemini is so ass lately
Seem so
i cleared cookies logged in, turned out no captcha asked
best model
10
19
3
sonnet 4,6
Which models not working for you?
I cant access the home / landpage
what’s going on with Arena… ToT I keep running into errors today, it’s so bad……….
OMG!! PINEAPPLE HERE!!:3
There is a visual glitch where some of the removed models reappear, hard refreshing the page should fix this. Looking into issues with 2.5 
Hello 
Thx so much, can you confirm that Gem 3.1 pro should be visible?
2.5 appears to be working for me, do you have a Trace ID you'd be able to share?
No, gemini-3.1-pro-preview should be a removed model from Direct & Side by Side atm
Is lmarena having problems for you? I can't access the site
I just tried and the last request seemed to work, but I'll pipe back if the direct chat spits out an error (it was erroring constantly earlier today)
What issues are you having? cc @limber dock ?
Sounds good, don't hesitate to share the Trace ID, ideally in this channel: #1417174113092374689
I can get in, but not all the models work. I tried switching to others and they all behave the same ToT
It's been like this for hours