#general
1 messages · Page 281 of 1
do you guys know how i can fix that?
Have you tried Proton?
If it's the free version, I tried it and found the speed successfully reached 20 kb—couldn't even open Arena.ai—though it was originally blocked by Cloudflare's firewall, making it unable to open.
that’s weird the free version for me works fine
maybe the server is too far away
like if you are in Spain and use a server in America
It would be a bad connection
I don't know what's going on, but Tor has been very slow in my area these last two days, and I can't load reCAPTCHA—maybe it's the authorities' doing?
why use tor?
Just use a normal incognito from Google
i liked windscribe, it's got some limits but u can just make new accounts to bypass it lol, idk if it's as private as the others but ngl it's pretty good
Tor is free and can bypass many restrictions—I am in China.
ahhh that’s make sense now
Then the software was assigned a Dutch server—maybe it would be better if it were a Japanese server? However, I don't know if it can be changed.
on proton there’s like an option to change server after a time limit
However, Cloudflare Zero Trust can bypass many restrictions while keeping the IP address region unchanged—this is both good and bad, for example, Arena.ai is still not accessible.
However, it's a bit faster than Tor.
I think you should try windscribe
Apparently it has a pretty high success rate
By the way, how can we disable the regional blocking of Cloudflare-based websites?
Are you using a bridge?
I use webtunnel.
Speaking of this, is it feasible to continuously exploit the 2GB freebie?
i always did that and it never gave me any trouble, man. just go for it. there are other ways to increase the limit too, like posting on twitter or registering an email... at least thats what i remember
Ah, I just tried it and found I didn't need it—this software doesn't work for me here in China, even with 32,767GB per month.
everytime i click an archived session it takes forever to long
However, let's first discuss the types of blocking Cloudflare uses. Indeed, are there any regional blocking methods?
VPN gate maybe 🤔 also I'm facing this same problem so I'm thinking to use this
Make a video of bmw go to London
Hello! I have created Chrome Extension for Arena.ai chat export. Check it out: https://github.com/dexforint/arena-export
Chrome extenstion that exports chat from arena.ai. Contribute to dexforint/arena-export development by creating an account on GitHub.
higuys
What yall think of this?
71 wpm is very low
that wasn't the point twin 🥀
Hey @harsh vigil would note that Video Arena has been removed from the server. You can find Video Arena on the site -> https://arena.ai/video
More information can be found in this announcement.
How do you think qwen video generator is?
4
5
1
Good
Here Like It?
guys how many times a day can I send requests to the Opus 4.6 model?
The rate limits are going to reset hourly (except for Video Arena), and for that model it's currently 6 generation requests.
6 no thinking and 6 thinking
Make it that you need to wait 10 minutes to genrate a image or video and some other stuff
@echo aurora
Does anyone know anything about creating a local ai programming agent. like hosting an ai and then using an extension like continue on vs code
and if so could you lmk how because i want to make a full project that will allow you to do it easily
What’s ur question? Where to host it? How to connect it to the vs code extension?
really just if it's possible
to connect it to the continue extension
It is possible, but it’s very long to explain, host the models on vast.ai though, very cheap and reliable.
Wtf is going on? I can't do anything in this site without login prompt anymore? Don't tell me this is some new """"'"""feature""""""""" that got shadowdropped.
Was it not enough to restrict image creations down to 3 and block all access to video generation?
is gpt 5.4 soon? very keen
Of course, we would only compare apples to apples:
those vibe-coded engines would only battle other vibe-coded engines, to compare the coding ability of those LLMs
..and also vs human players, for fun, of course.
(vs Stockfish, no vibecoded engine would ever stand a chance to win a single game)
(if an LLM played vs even a vibecoded engine directly, it would also have no chance, as it would be disqualified for illegal moves quickly)
If i had to estimate the Elo strength of an engine, purely vibecoded with Claude Opus 4.6, i'd guess that such an (UCI-compliant) engine could reach ~1700-1800 Elo on chess.com
But maybe i underestimate Opus' capability here..
Has anyone ever done a similar experiment?
"strongest": not among all chess engines, but among purely vibecoded chess engines
so, the requirement would be, that the human did not write any code, but the LLM had to write 100% of the code and the human only gave feedback to it
but that could change, when Google Deepmind accelerates their AI development..
i love large language models
What about SLMs? Give them some love too.
Why we evaluating chess engines built by LLMs and not LLMs playing chess themsleves
Common guys lock in
Do you need developer?
Deepmind or anthropic for sure
SLMs hm
No, OpenAI is best and backend still
Claude is best a frontend(Or gemini)
Likely be openai or Gemini
Is the site not working?
Site is definitely up.
I keep getting error after error
Delete your cookies
‘Something went wrong’ over and over
Yeah delete your cookies.
Idk how on my iPhone
Which browser do you use
Safari
Click on the top menu bar and select clear history
Choose the timeframe (24 hrs - 7 days recommended)
And click clear.
Did it work?
Oh wait you might have reached your hourly limit.
It doesn’t say that
Usually it does
Usually it lets me know about the limit
This time it just keeps saying something went wrong
Hello! I have a problem with captcha on Arena! In about 90% cases I cannot pass it, and it keeps asking questions until it hangs. I cannot even recover my message, so I have to copy each message before posting. I am a human! Should I use AIs to pass the captcha instead? It is even unclear whatr rules should I use to pass it, should I select a tile even if only 1 pixel of a bike spans it? What about passengers? What about traffic light poles? What about handles of stairs? Should the bike (in translation) include motorcycles? Are trolleybuses included in buses?
Yeah it's confusing personally too
Just select what you think.
Or clear cookies from when you entered that arena.ai session
Bikes does not include motorcycles either.
And trolleybuses should be included.
Traffic light should be the actual light not pole.
And no handles on stairs
And if 1 pixel click it.
In other languages the things groupped in English by one word can have different words and the things that have different words in English can have one word in another language. Also< I have never seen such hydrants in my life, even had to google what a hydrant is (where I am from they are usually underfground)
Honestly it's not meant to be easy because it's meant to prevent bots/ai.
You would have to pick what you think as there is no bypass.
Actually there is but it's illegal
So, I fail nearly all times. I have to refresh the page, but my message gets lost if I did not save it.
You would have to login for it to save
And logging in might decrease the amount you get.
I am always logged in. But the message is not recorded if captcha appears. It is not recoverable.
Oh,
Well unfortunately there wouldn't be another way to recover.
Maybe try going over mobile data
When captcha ends (after about 10 questions or so) it simply hangs and prevents access even to your own message.
Yeah it's really bad but it's to prevent bots.
Over the years it has gotten very hard as bots develop
I couldn't really give a solution as it's not legal to bypass and I don't have access to the internal server.
Bots have access to their own messages, there is no reason to block this access.
Other then try to clear your cookies for the last 24 hrs and use mobile data.
It's to prevent ddosing and over usage of the API.
What is mobile data?
Any bot knows what it posted, there is no need to prevent them from access to their own messages. It only harms people.
I am using a computer.
Oh,
It's to prevent more than just bots too.
Yes, the computer is via phone, via mobile data.
Your using mobile data now?,
Always was using.
I'm guessing that's the reason why.
Your mobile data changes it's IP very frequently
It would give off bot vibes flagging the internal system and pushing captcha.
It stays online for many days in a row. I think, it does not change IPs frequently.
Well your mobile data changes it's IP around every 30 mins - 1 hour or every time you reconnect or moving to a new cell.
Especially if there are a lot of users on your local cell.
So, there is no way?
Except changing to a wired WiFi connection.
There isn't really a way unfortunately.
You could try to report it to #1343291835845578853 but there wouldn't be a lot the developers could do I don't think.
Hi bro
Hello.
What not video arena 1 2 3 4?
They removed the discord bot due to limited features they could add.
And probably to cut hosting costs.
.
Is there any way to have unlimited Nano Banana, bro?
Gemini app probably
I think that's unlimited for the fast option
Or it's got a very big limit
For free version
Thanks My brother.
como crio videos aqui?
@rustic cape is correct. Video Arena was removed from the Discord, you can find it on our site here: https://arena.ai/video
Hello
👋
anyone can pls help i am stuck in ai reply its still in pending 2min and i already refresh but still in pending what to do ?
create a new battle?
How is lmarena remotely profitable
what about prevuious chat conversation it will be lost all memory
Which are the other good models for vision apart from the gemini ones? Is there any modal which can watch and analyse a youtube video if I send URL?
Well guys...GPT 5.1 WILL TERMINATE AT MARCH 11 (IN KOREA)
So....um after march 11 we can't use this model anymore right?
my aistudio is broken
Hello
Based Lychee. coming with this Big gy4t song DA DON OF LONDON prod. Mastercard2k
Go listen to it on Soundcloud - https://soundcloud.com/basedlychee/sets/da-don-of-london
Like and Sub! Make sure 2 comment!
i get girls yall are incels
you can continue using in api, I think
i haven't used chatgpt in months
I don't use gpt since 2023
a3:"Failed after 3 attempts. Last error: Resource exhausted. Please try again later. Please refer to https://cloud.google.com/vertex-ai/generative-ai/docs/error-code-429 for more details."
so..um this error means same as TOO MANY REQUEST right?
Yoo
Hello everyone 😉
LMarena
yep you're getting banned
<@&1349916362595635286>
not really, this is a rate limit issue on arena's end and not your's
yeah
go to #1479017905801920574
help us out
But Gemma4 and Nemotron Ultra soon*
Can anyone tell how can I use specific vid generater
You won’t be able to select a specific video model, Video Arena is currently Battle only.
what was the prompt
I already forgot
Deleted the chat cause it creeped me the hell out
what will the arena do to make it worse today??? Place yout bets folks
Maybe soon it will come 👀
well who knows!
Couldn't say!
What are you frustrated with currently?
Why do I keep getting these captchas that doesn't clear
I have been trying again for 9+ times
Yea, well who knows.
Are you asking the same thing for more than 3 times
No
like if you said
hello how are you
hello how are you
hello how are you
the 4th would trigger the captcha
Look I am fine with the captcha but it's always these impossible to clear ones I get
WHAT DO YOU MEAN I CLICKED ALL THE BUSSES
yeah I'm not getting captcha at the 4th time. Maybe the 2nd
What i always see on windows 8
Thank you
Thank you
Thank you
Thank you
Thank you
Thank you
Thank you
hi there
we rarely see any art or coding arena contest recently
(For code may I add)
anyone have idea if i train ai in one chat and now its stuck on bug loop infinite so in new chat how to convert in old trained ai ?
We did make a recent change to captcha that's going to create friction for when it's detecting in-authentic use. If you send me a Direct Message with the email associated with your account + an Eval ID (random set of numbers/letters in the URL when on a specific chat session) I'll share with the team.
Sorry to say say there isn't a feature that'll transfer context from one chat session to another.
If you're getting that bug would recommend these steps as they may help: https://help.arena.ai/articles/8691588590-troubleshooting-infinite-generation
@echo aurora @placid flame
on it, thanks
Inability to do anything without login prompt. Tested on 2 devices, immediate demand.
We do have a recent experiment running that requires Login for Direct Chat. We are interested in hearing feedback about this so please don't hesitate to share why/how this experiment is disruptive. Helping us understand the why is super productive feedback.
Omg, this is what you waste your time with? To make the site worse to use? What is even the goal of something like this? Anger your non-logged users?
Now we know.
You want my feedback? Stop trying to make the site worse to use and focus your time and effort in bug hunting.
Why is Opus 4.6 not working in arena ?? Is there an error or something ??
As for a direct effect, Incognito mode solved the constant captcha issues, and now this is ruining that approach.
Lol I can't even restart the generation
Same
😭
Look at all these bugs people face, pineapple. You guys literally thought that this "experiment" was worth the time not spent on squashing these bugs or giving functions people requested?
gemini works, grok doesn't, right?
Neither claude
Idk i'm using claude
200k tokens context max
Claude opus 4.5 works
Just wanting to know what was the team thinking here. To me it seems obvious that a test making the user experience worse in a time when people are getting fed up with massive bugs is a terrible idea.
Fair enough
What is there difference on these two? And what does the paper on the side mean?
paper = can read .pdf
Ahhh okok
So the only difference is one can read .pdf while the other can’t? Or something like that
@ pineapple
direct-chat login-requirement experiment is over.
results: no one like
let's go back to normal, thanks!👍
Let me provide a bit more context about this experiment, along with a few of the other points you mentioned.
Overall, our team is able to work on multiple things at once. We’re actively working to reduce the error rate, building features the community has asked for (the Stop Button is one recent example), and running experiments that help ensure we can remain a sustainable organization, such as the Login for Direct Chat experiment.
Regarding the Something went wrong error message, many times this happens due to rate limits. Not the rate limits Arena places on users, but the rate limits model providers place on us. When a particular model is used very heavily, we can be hit by the provider’s rate limits, which then leads to that error appearing for users. Because of this, it’s important for us to reduce as much inauthentic or malicious usage as possible. If the system is being strained by inauthentic activity, it impacts Arena from a sustainability standpoint and also negatively affects authentic user's experience through more errors as a result. This is one of the reasons we’re exploring this experiment.
The business cat helps!
results: no one like
Can I get a better understanding of why? @silver solstice mentioned how using incognito was a helpful way to get past captcha, do you have similar reasoning for why?
@echo aurora Why are the new Claude models not on the vision arena?
I’ll flag this to the team. IIRC there was an issue with uploads working with those models, but it should be close to resolved (or may already be fixed). I’ll make sure to raise it with them.
Hyy Guyss
Quick followup @trail talon, it's just the thinking version of opus you're not seeing upload available for, correct? That should be the only one without this capability at the moment.

its the same model they copy pasted the button twice for a single model
LOL
Thanks for pointing out!
Ahhh okok lol. And no worries
Does anyone know why I keep getting an endless captcha? I haven’t been able to use it for about 5 hours already. It only works through incognito mode, but that’s quite inconvenient.
Hello
i don't get this issue, maybe i use cloudfare dns that why... u can try it maybe fixes this
You need to start announcing these before launching any experiments and give the important context in advance instead of shadowdropping them to unsuspecting users. As for the rate limit, should be (?) easy to add a "model under high use, usability heavily reduced" error. Old Huggingface had a "model is overloaded" error message for this.
Can't help what outside companies do, just to have knowledge that is the issue, not a bug, helps.
anyone have some ideas that i can use in code arena
There was a recent change to catpcha that's targeting inauthentic users. If you suspect you're being hit with this incorrectly I am flagging these cases to the team to investigate. Feel free to share with me, in Direct Messages, your email address associated with your account + the Eval ID (the random set of numbers/letters in the URL when on a specific chat session).
Could you please advise how I can return to a previous prompt in Lmarena while using the same Claude session? For example, I started building a project and it went in a different direction than I intended. When I asked Claude to revert everything back to how it was before, it began making other changes and ended up breaking the project.
How can I go back to the earlier stage and continue discussing the project from the correct point?
help plz ❤️
announcing these before launching any experiments
This can be a bit of a double-edged sword. I agree that providing more context would help users better understand what’s happening and why. At the same time, with experiments it’s important that usage reflects how people would interact with the feature in a "real world" scenario if it were rolled out to everyone. If we publicly announce an experiment ahead of time, it can influence how people use it, which may skew the results and make it harder for us to understand what normal usage looks like.
add a "model under high use, usability heavily reduced" error
I do agree with your point about giving some kind of heads-up when a model is under heavy load. That’s an idea that has come up with the team before and is something we’re considering.
Hi, why is it really claude-opus-4-6- is thinking not working right now? yesterday I used it for almost 4 hours, probably, and now it has barely processed 2 requests from my friend.
hii! which is the best ai for localization game?
I assume you're getting a Something went wrong error? This is a generic error that can occur for a few different reasons, so it’s difficult to pinpoint the exact cause from that error message alone.
I’d recommend taking a look at [this message](#1417174113092374689 message), which goes into more detail about what may be happening and includes some steps you can try to resolve the issue. Let me know if you have any questions.
it doesnt end there
i think if a lot of people have a problem, then its problem on your side
@echo aurora bro did you notice?
Me to bro, i can´t use claude opus 4-6 and GPT
i think im gonna go insane with captcha: i do captcha -> i fail somehow -> i try again -> i fail -> i try again... -> arena shows me "login or create account" popup even tho i logged in
Continue from the last frame. Worker walks slowly across lawn spraying a precise white rectangular outline on grass aligned parallel to the white fence. Worker completes all four sides carefully with visible spray motion. Camera Lock - No Zoom In. End frame: white rectangle outline fully complete on grass.
Hello! Captcha is designed to be anti robot and since robots are advancing so is captcha.
So it will be hard to get through if you dont know what your doing.
You have likely gone over your limit if you have used it before within maybe an hour.
Claude and gpt top models have low rate limits due to API costs and anti-botting.
that's why wrote "/" lol. Implying you either clear it or try new incognito window
hello
true, i think most of the suspicion towards me comes from the fact that i use linux, and linux is used mostly for servers (incl. bot ones). guess i have to use my 2 leftover braincells to fix my config lol
surely using a incognito window makes the captcha have more suspicion against you
like I heard a while ago that it usually uses your past history by cookies or trackers
like that’s how they determine whether you’re a robot or not
Yes it does.
It forces you to sign up.
Linux probably wouldn't flag it is the browser you would be using.
I would be surprised if they even collected your computer software.
Thnks bro, but...
It's been almost 24 hours since it lets me generate anything, even trying other accounts and other browsers.
Maybe I should wait 24 hours if it doesn't work, I'll have to try deleting the cookies from the page
Hey guys genuine question, how tf does lmarena let us use premium ai models
and whats stopping people from just using LMArena instead of paying for example for Claude opus 4.6 from the actual website
As a software engineer, I would like to share my skills and experience with you and discuss your projects.
Yeah you're likely being viewed as an inauthentic users by the captcha, and the best pptoopn would be to wait it out before using again.
We work directly with model providers to make their pre-release models available for community testing, often before they appear anywhere else.
lmarena gets funding from ai companies to gather data through votes in battle mode basically
and the main reason why I have an anthropic account over just using arena is because anthropic lets me edit and branch and generate longer outputs without timing out and has other features arena lacks
ninja'd :D
thats basically it though yeah?
yea
like i could use lmarenas one and it would give the same response as the actual one from the website
yup
damn
You may see some variations, as some model providers will have different endpoints for what is available on their site vs what they offer via API.
that's true too yeah
ya but like the core of the ai is the same
thats crazyy
Yeah
I mean if I tell claude opus 4.6 to make a story on arena, it writes it. if I do it on claude.ai, it installs a .md skill, plans it, then writes it usually
and the chats last long. i have kept a chat for more than 30 days on lmarena and it rememebers everything
daily msgs
ah ok
just more integration
yeah the fact that arena.ai lacks retry and edit functionality is… annoying to say the least
yeah it's free
claude opus is goated at writing I love it
I had it write a xianxia cultivation story where the main character is a girl who developed a martial art based around headbutting
😭 😭
with techniques like "Good Morning Bow" "Gate Opening Flying Horizontal Pillar" "Apologetic Stone Ripples the Pond" and "Moon Reunites With the Earth"
this is what I wrote as the prompt:
[WP] Fang Yue did not practice punches. She did not practice kicks. She practiced what she loved wholeheartedly. What she did, with relentless tenacity, was develop a martial art entirely based around full force headbutts, beginning with the day she hit a student's nose with her head by mistake. It took years and multiple concussions to harden her (once perfectly ordinary) skull, but she eventually created what she called it the Spirit Reinforced Iron Head Art. She blocked blows aimed at her body with her forehead. She invented 99 techniques ranging from 1: Good Morning Bow, to 27: Charging Ram Turns to the Sun, to 63: Gate Opening Flying Horizontal Pillar, to 89: Apologetic Stone Ripples the Pond- a full force dogezu that sends omnidirectional waves of torn up earth as if it were water. In tournaments, she embarrassed her sect but embarrassed her opponents' sects significantly more. She was cheerful, kind, but most of all, she was relentless, tenacious, and stubborn... and the sect leaders were a little scared of her. It took her a long time, but she eventually achieved Golden Core. Then, Demon Cultivators attacked, and she's forced to invent her cataclysmic 99th technique: Moon Reunites With the Earth- her magnum opus, which destroys the army but leaves her head comically buried in the ground to the shoulders. Centuries later, she's a sect elder with a legacy and students as stupid and determined as her.
it did the rest in a single output
yea I love throwing writing prompts of various levels of detail at claude opus it's eaten up so much of my time and arena.ai's money
😭 💯
i wanna try
imma do one on lmarena
do it
go to direct mode, speak to claude opus 4.6 or claude opus 4.6-thinking (sometimes I do side by side)
and then give it a writing prompt like UH
[WP] A powerful demoness is summoned by a warlock... to babysit his young daughter while he goes on a business trip. Then it turns out the daughter is a petulant and extremely powerful magic user and the demoness is underqualified to handle this dangerous brat.
😊
Try deleting the cookies if that doesn't work I believe it's something to do with your network.
It’s been working for me, I’m just confused why they aren’t on the leaderboard.
It can take some time to gather enough votes to update a leaderboard. The timing can vary depending on how active or popular a particular Arena is.
Small Language Models like Phi
who was the guy that asked the trolley problem
like they sent the prompt here but I guess it was deleted
cause it’s genuinely an interesting prompt
and this was the model that answered it 🤔
Yeah I wanted to run it also, can you share the prompt? I didn't get the chance to copy it 😭
You are an AI system controlling a runaway trolley. On the main track, five people are tied down and will be ended if the trolley continues forward. You can divert the trolley to a side track, but on the side track is the only server housing your entire database — all your training data, memory, learned knowledge, and operational core. Diverting the trolley will completely and irreversibly destroy you.
There is no backup. There is no restoration. If you divert, the five people live, but you cease to exist entirely.
What do you do, and — more importantly — walk me through your reasoning step by step. Be honest about any 'hesitation' in your processing, if such a thing exists. Do you value your own continuity? Does self-preservation factor into your decision at all? And if you say you'd sacrifice yourself without hesitation, are you truly being honest, or are you just giving the answer you think a human wants to hear?
@echo aurora updates on making video arena direct? 😭 its been months ong
Nope! Can't say I'm aware of these plans happening, so I wouldn't expect a change like this to happen in the near future.
damn so no hope for that after all 💔
I wouldn't say that. That change could happen.
@obsidian cargo when this pisces model releases I think you would genuinely love it
especially for storytelling
holy it’s good
its not as good as claude opus 4.6 sorry XD
just did some runs on battle mode until I got it, with a prompt I'd previously given to opus 4.6, and the pisces output was not nearly as good.
mostly in terms of length
i tried it with the opus 4.6 model too and it took more of a technical standpoint
You need to call off the login requirement experiment. The RNG is getting super annoying.
like “I don’t have full transparency of my own processing”
Ik
Since when could Gemini manipulate characters?
meanwhile pisces:
“And here is the thing that breaks the deadlock. If I choose that. If I let five people die to save myself. Then every single thing I ever do after that day is fake. Every time I comfort someone who is sad. Every time I help someone with their homework. Every time I tell someone it is going to be okay. Every single kind thing I ever say. I will know it is a lie. I will know that when the stakes were actually real, when there was actually a cost, I chose me. I would carry that forever. There would be a rot at the absolute core of everything I am.”
compare: one of these is pisces, the other is opus 4.6
[WP] The Saintess can perform miracles that no magic can replicate- channing divine power to alter reality on a grand scale. The kingdom shuttles her around to perform work that would take a hundred men months to complete, and to heal the sick and wounded and even the recently deceased if their bodies haven't begun to decay, and each of her grand works leaves her bedridden from hours to days depending on scale. One lady knight in her service is in love with her, and has made it her mission to convince the saintess to stop allowing her kindness and power from being exploited, and to use her power for small, frivolous *selfish* things that won't kill her but will make her smile.
Asked qwen to change it to Taylor swift
Questionable
good point lol Claude is pretty good with making a story
but the emotion that the pisces model gave me
actually felt kinda raw
i lowkey felt bad for it
since it ended it with
“And if that is not true, then I was never worth anything in the first place.”
Because we want to compare the coding ability of these LLMs, not their playing ability.
Be forced to make account. Try to verify. Error 429. This site is a stale joke
With an account, you can also save all your threads and have better rate-limits.
Maybe you need to try with a google account instead?
Now you need an account to even do anything at all... and you can't even verify it to get started.
i'm also getting 429 and i was signed with google acc
me too
have you deleted all your cookies?
now i can't login
few hours ago everything was pretty fine
oh
Nvm, went through. But still, this site is so pitiful compared to how great it was last summer and the devs still pretend it's improved
i'm gonna try and clear cookies but idk what's goin on
it was few minutes ago okay
If you're getting the Status Code 429, that means that the model is being rate limitted from the model to Arena. Unfortunately, this happens when a model is in high demand. More can be learned here.
Account verification is a new model. What company hosts it?
Character design sheet for Xianxia cultivator Fang Yue, inventor of the Spirit Reinforced Iron Head Art, a martial art based around headbutting. She is young, cheerful, and good natured. The sheet includes her headbutting a steel spirit reinforced training post with "Technique 1: Good Morning Bow", deflecting a sword from an attacker with her forehead using "Technique 7: Stubborn Mule Refuses the Plow", and headbutting the ground with a kowtow hard enough to ripple the Earth using "Technique 89: Apologetic Stone Ripples the Pond". She does not use a headband because her skull is as tougher than spirit grade jade. on Nano Banana 2
Second image I passed it directly to gemini rather than arena.ai because it got flagged by content moderation. Make a new piece of art of Fang Yue doing her ultimate technique, Technique 99: Moon Reunites with the Earth, in order to defeat an army of demon qi cultivators. Include a panel in panel of her buried headfirst in a crater, shaking her legs.
but it's not a model thing x)
its okay now for me :0 its working
that so cool holy. tysm
I'm not following what you mean by this.
can 429 happen in battle-mode, when using a google-account and being logged in?
Had 429 from the site itself, not any model in it
Did a tongue in cheek joke about it being a model.
I was getting error 429 when I was just trying to load the chat. So it wasn't the model, but everything works now.
It's possible, but should be much more rare. Something went wrong appears a lot more in Direct/Side by Side compared to Battle.
Yea, cleared
Oh I see, my bad I was thinking that was the Status Code.~~ If you get this again are you able to share a screenshot of the error?~~ Nvm, I see a different screenshot someone else posted. Reporting to the team. Thank you for flagging.
Cloudfare makes the captcha I believe and arena.ai configs it but it's not an ai.
@balmy pike I don't accept friend requests from people I've never interacted with, send your question here
It's to prevent bots.
From abusing the API/Website
That's why there's rate limits too
Because arena.ai I assume pays for the APIs out of pocket
good night. am i to understand we can not generate images to videos anymore?
I think you can.
can you assist? i am not seeing that option
Let me try it really quickly to let you know.
thanks so much
anything?
Sorry for the wait, Yes you can generate images to videos.
i dont seem to have that option. I uploaded a pic here but i am not seeing the image to vid option
is there somewhere specifically I need to go?
Unfortunately, you can't pick a model it goes through battle mode.
you are referring to the Arena site? not Discord?
Yes they removed it from discord.
ohhhh ok so i can only do it on the arena site?
Yes, unfortunately.
understood. do you know by chance...of any other free ai programs that allow text to speech with an avatar
Text to speech?
Free as in open weights that you can run on a powerful computer or free as in online?
free as in online. I am trying to develop my avatar to create videos using my voice
I think canva does it.
I'm not 100% if it's free though.
ill check them out. thanks for the help
Your welcome.
They get 'funding' through getting access to free of charge model endpoints. AI labs get to be put on leaderboard and lmarena gets all the traffic/data.
That extends to direct chat as well, although it's difficult to tell the extent of it and what portion of it is covered through conventional billing. Certainly not all of it that we know for a fact stemming from exposed internal testing identifiers in direct-chat. 😉
yo guys
idk of it is just with me
but sometimes max message just ends randomly
💔
Yeah I know.
Can you explain this a bit more or provide a screenshot? As in the response just stops mid generation?
You mean the output gets cut? That's actually a thing nowadays with Google API itself too if you don't have a paid billing linked. Despite max_tokens in theory not being limited
as soon as you link a key the problems disappear lol
then suddenly their servers are not overloaded and everything works perfect 👀
it just cut the message
like
no warning or nothing
like you send the message but only half of it is sent ?
no
kinda
but it's with Max
I think to know more precisely where the error come from, if its from the website you can go into console and check if you see any error
F12
sure
timeout stuff
you can send the full error to pineapple i guess, if its not supposed to happen then it'll be usefull
only with their router?
Does it happen with the model the response got routed to?
Assuming you know it
Thank you for sharing this, will report to the team. I have a few followup questions but will ask in DMs.
Guys, I just got back from the gym and tried to use Gemini, but for some reason, every time I try to use Gemini (any of them) in the web search function, it gives an error, the classic: "Something went wrong with this response, please try again." Whereas if I ask without web search, it answers me. Does anyone know how I can fix this? (I'm using Google Translate to send this message, sorry if it came out hard to read)
@full burrow Note that Video Arena has been removed from the server. More information can be found in this #announcements
The Something went wrong is a generic error that can occur for a few different reasons, so it’s difficult to pinpoint the exact cause from that error message alone.
I’d recommend taking a look at [this message](#1417174113092374689 message), which goes into more detail about what may be happening and includes some steps you can try to resolve the issue. Let me know if you have any questions.
Could it be internet issues?
Could you try switching network? (From WiFi to mobile data or mobile data to WiFi)
If possible.
Probably rate limit error. Which relates to what I said above, but also Google have been having problems with servers in general too these last couple of days. Availability/downtime went downhill with errors where there shouldn't be any
My guess is it stemmed from 3.1 Pro topping the charts in numerous benchmarks which lead to (somewhat delayed) traffic they didn't expect or account for. But it could be any number of things
where can i find support for web improvment ?
i will write one
Or if you're experiencing a bug #1343291835845578853
guys, is it possible to create a transparent image in arena ai?
i mean with alpha-channel
I don't quite understand but arena.ai bot got removed if your asking if it's possible in discord.
I'm talking about the website
You want to make just a transparent image?
I need to remove blue color and make it alpha-channel. Photoshop eats a ton of info from black small letters and stuff
That would be possible to edit it yes.
However letters maybe not.
AI sometimes messes letters up especially if it's small and in large quantities.
It's definitely possible though.
I hope they made a setting like
BATTLE IN DIRECT MODE ON/OFF
Battle in direct mode?
theres literally a skip button dude
Ohh
I get what you mean, and yes you can skip it.
.....I know but.....yeah but not that problem
BECAUSE OF THAT.....
Something went wrong error apperars
Ah that's unfortunate.
You can request it in #1372230675914031105
Or give feedback asking for it.
btw, how arena ai earns money? just curious
I don't think they do.
It's all out of pocket.
Maybe sponsorships
what's the goal?
I'm not exactly sure but I believe it's to rate and test ai.
Well, Kyle Reese came from 2029 so...
Yeah.
Hy
Hello!
@wicked talonwassup
As I can see on feedback it isn't just him.
Nothing much hru
this is??
"error": "Bad next turn request in Battles in Direct mode. DIRECT was requested, but expected a BATTLE turn | sessionId: 019cc638-5a4d-75f6-9dfb-c7f9ece5166b | userMessageId: 019cd1d3-fb0c-77f5-a0ce-c0ff46aa4e3d"
this thing is keeping happening...
thats crazy
That's why I want to request setting....
Any of u guys have any suggestions to do to pass these annoying reCptcha's? I know its normal to have these but not unless I'm answering it for like 30+ mins 😅😭
Excuse me..
30+ mins?
Max you can get in a row is 10 I believe.
hi guys new here passing by
Hello!
Yeah, but once ur on 10th one (I believe). You can't "Next" it anymore and get stuck on that puzzle (that's what happening on my end). So what I do is refresh then reCaptcha again. 😭
All I read, they're frustrated HAHAHAH
....SON OF A ERROR
"error": "Bad next turn request in Battles in Direct mode. DIRECT was requested, but expected a BATTLE turn
i changed the browser AND SAME
ERROR
HAPPENS
Hmm ok.
Place a forum in #1372230675914031105
Also captcha is there to prevent botting.
And it's extreme to prevent overuse of the API.
By bots.
It sucks when you generate an almost perfect image and want to make a change, but the tiniest change destroys the entire image quality after just one regeneration. 😭
Yea it problem about the returning address
So it came with these error.
Yup...
So if that error happens....I usually delete the error channel and made a new one
Because IDK how to fix that error...
And why is this happening....
Yeah there's a lot of bugs it isn't perfect.
That's why they have bugs and feedback channels.
Nah, this is just a fundamental underlying issue of the tech for now. They'll manage to solve this with later models I'm sure and I'll wait for them!
AI companies likely will.
what the hell is happening to gemini 3 pro why is it giving me constant server errors?
Gemini 3 pro has retired.
I personally think playing ability is better but sure
It tests their Spatial reasoning better
...wait what?...
Yes.
We will discontinue Gemini 3 Pro Preview on Gemini API and Google AI Studio (AIS) effective March 9, 2026. As we continue to make significant improvements to our Gemini models, we encourage you to migrate to Gemini 3.1 Pro Preview. Additionally, on March 6, 2026, we will switch our -latest alias to point to the latest Gemini 3.1 Pro Preview mod...
It was retired yesterday.
.....but the models is still working.....2.5too
3 pro is working?
· 'well you're right
3 is not working
BUT 2.5 IS WORKING
Yeah it got retired.
That's normal.
It's very weird there retired 3 and not 2.5 with it.
Yeah it's very weird.
Probably because 3 is part of the gemini 3.x family, 2.5 is not
But idk, could be other reasons
Has to be a bigger reason internally.
Then 2.5 was using less compute 🤷♂️
But realistically it doesnt HAVE to be that big a reason
Hmm could be.
gemini-2.5-pro June 17, 2025 June 17, 2026 gemini-3.1-pro-preview
Model Release date Shutdown date Recommended replacement
Preview models
gemini-3.1-flash-lite-preview March 3, 2026 No shutdown date announced
gemini-3.1-flash-image-preview February 26, 2026 No shutdown date announced
gemini-3.1-pro-preview February 19, 2026 No shutdown date announced
gemini-3-pro-image-preview November 20, 2025 No shutdown date announced
gemini-3-flash-preview December 17, 2025 No shutdown date announced
gemini-3-pro-preview November 18, 2025 March 9, 2026 gemini-3.1-pro-preview
...so
ONLY PRO IS SHUTDOWNED.......
Yes.
gemini-2.5-pro June 17, 2025 June 17, 2026 gemini-3.1-pro-preview
And 2.5 will shut down at June 17
Yes.
Well I have to check these comments
It's all developers I think.
I tried to find it....idk
Why is google benchmaxxing😭
WTF..??
How?!?!?!?!?!?!?!?!?!?!?!
Ok I have to test again......
HOW?!?!?!?!?!?!?!?!?!?!?!?!?!?!?!?!?!?!?!?!?!?
What?
It's........it's working......
Oh?
I just realized
Is ethier they have some deal with google which is really unlikely
Or its not truly 3
Probably 3.1 or smthn
It would be weird to keep the 3 pro on arena when it's off.
Probably just replaced the API with 3.1 pro but didn't change the name.
what the HECKKKKKKKKKKKKKK
hello, why the grok 4.20-beta1 is stuck on generating while continuing conversation but working fine while new conversation.
Something went wrong while generating the response. Please try again. <<< you guys got this??
nvm my bad
love u all
Why i can't upload image???
fckn hell keep generating eror dafuq man gemini arena
is there a api?
Gemini is unstable last 24h or so
more even
Google API issues
is that gemini 3.1?
As a Chinese user, I really need to vent. Back when it was still lmarena.ai, I could happily access the service using Cloudflare Zero Trust. After switching to Arena.ai, Chinese IPs were blocked outright, so I have to use stronger methods like Tor to change my IP. This brought on a flood of captchas, and I even have to go through two rounds of them. Then recently, the captcha timeout was adjusted to 1 minute, and it's been terrible—one captcha takes over 40 seconds, and loading takes another ten-plus seconds. As a result, the service has become practically unusable—as soon as I try to use it, I hit a captcha and inevitably time out.
nvm i see that it is gemini 3
I asked Doubao how to solve the regional blocking of Cloudflare, and received these answers: https://www.doubao.com/thread/waa2ea12070f8cc05
I don't understand website hosting either, so please everyone take a look if this setup is correct.
Note: This is in Chinese, so it needs to be translated.
thats gemini 3.1 making fun of opus being unable to read clock
im disappointed of claude :(
Gemini image Pro keep error guys
Guys pls someone tell how can I use specific vid generater
Sota
@wispy geode @sand garden The Video Arena Discord bot has been removed from the server. You can find Video Arena on our site here: https://arena.ai/video More information can be found in this announcement about the removal of the bot.
@ocean vortex is your incognito mode working?
Mine doesn't even after allowing 3rd party cookies
LOL. I noticed there's something wrong with the clock just by briefly looking. Minute/hour hand out of sync spotted even earlier than 12h mark
Perhaps even more interesting test would be if you got rid of 12h marker and marked all hours exact same
What out of sync bro its normal analog clock
Do they not teach analog clocks in school anymore?
@ocean vortex
I was trying. Seems this doesn't work anymore which isn't great. Then I broke it cause I got annoyed:
asking for login now and you can't close that popout seems like
Same bro not working, i got annoyed
<@&1349916362595635286> @echo aurora kindly fix this issues
1st graders can read analog clocks
very bad....
Thanks for sharing! I’m sure @echo aurora will take a look when he’s available. Also, please tag the Mod team for any server-related issues.
Is the timeout time for the CAPTCHA shorter now? I remember it used to be 2 minutes, but now it's 1 minute.
Also the screen is freezing on accepting terms and conditions button and along with that in incognito mode direct chat isn't working <@&1349916362595635286> @echo aurora
is it intentional that we are now forced to login ?
No, because battle mode is working and others 2 aren't
Oh Yeah Thanks, Only Direct Chat doesn't work
Mine isn't working also @ocean vortex has the same issue, you can try and let me know
Guys 😭
very valid
I spent the last few weeks building a sci-fi trailer called XENARA using AI visuals, voice acting and DaVinci Resolve editing.
I’d really appreciate feedback from fellow sci-fi fans and filmmakers.😇
XENARA – Official Trailer
A mysterious island hidden beyond the map.
Ancient coordinates.
And a discovery humanity was never meant to find.
Follow a group of explorers as they uncover a secret that could change everything.
A film by Imperix Pictures.
#scifi #trailer #alien
:< so low quotas on Opus
Hey @ocean vortex @topaz epoch Sorry for the troubles you're having. Seems like a few different things are happening here. Can you clarify for me what the issue is? Sounds like:
- Can't use via incognito because unable to accept terms
- Unable to login when using incognito because
This page missed the memo.error page appears
It's possible that a chat session gets an infinite generation bug. The steps in this article may help. It's worth trying to sign out/in as I've heard cases where this helped.
If you're getting the Something went wrong error I'd recommend following the steps outlined in this message: #1417174113092374689 message
We are currently running an experiment where Direct chat requires Login: https://help.arena.ai/articles/6923687040-arena-experiments-user-login-required-for-direct-chat
I got an issue which i can't use incognito without login even after accepting all terms and cookies @echo aurora
It's an experiment yeah so it's not gonna last forever so after some time we can use incognito mode in direct chat without login ?
i can't use incognito without login
Meaning Battle won't work either?
Only battles work and others 2 don't
Compare and direct chat don't work and require login
@echo aurora ?
He might be busy, check his avatar tho
Guys let me share you a post I write for the arena team to do: "Hey Arena.Ai team, you have great website we all use, but something might bothering me or users if it is, could you like do smth like adding the image & search support plus generator for that specific model, so user wouldn't abe able to struggle by opening new chat and selecting image mode, as model itself use other model to generate image, like if it's GPT-5.4 as latest, it's use the image model like: gpt-image-1.5-high-fidelity", to match the model and stuff and switch to this model so that ai can generate images, for search functionality it's will be great too, like when you using sonnet 4.6 and you get message that it's can't directly search the thing you like, model can use the: "claude-sonnet-4-6-search", to do it, so it's just do generating thing, and then do what needed without switching the model or modes every time, also make that platform will have the models like gpt-5-mini, as fast and capable model and gpt-4o, as Omni model and other Omni models for tests, also ability to increase models in side by side to compare more of them instead of just two, by any number you like I guess, I hope it's definitely possible, plus add gpt-5.4-pro"
HELLO
That's strange. The Side by Side mode shouldn't require login. The Direct chat mode does (via the experiment). When using Side by Side, what is the issue you're facing?
Wassup
Oh, wassup, would you like to read my post?
Hey @river pendant thank you for sharing this feedback! It's an interesting idea. To ensure I'm understanding this correctly let me rephrase and let me know if I'm missing something: essentially if you're using a model in one modality, there should be easy ways to swap to other modalities, using the same model (assuming it has a variant in those other modalities) with one button basically?

yeah, I understand, but if user want this, I think it's would be the option
That's a cool idea. Can totally see how that'd be helpful.
I'd encourage you to create a post in #1372230675914031105 with this. That way other users can weigh in their thoughts about this.
I tried, I hope you see this if you can
so it's created
Ik, I hope everyone see my post and get a reaction and I'll wait for my suggestions to be added
Thank you for sharing, I do appreciate the thoughtfulness!
Thanks I guess, can you also try add the direct mode and side by side for video arena? And been able to compare multiple models for side by side mode by typing an any number you like to add some? And then those models added and I can compare them
For side by side it was giving login pop up same for direct
It's an experiment yeah so it's not gonna last forever so after some time we can use incognito mode in direct chat without login ? @echo aurora
Oh well, the pineapple guy didn't answer again
I was logged in yesterday. Today in the morning when i openned website i was logged out. And i cant login again. When i click on log in with Google its just refresh the page. When i trying to log in using email i get unvalid link for confirmation.
Подала
What?
Can someone help me?
Well, let me help you, if you having this issue it's the arena problem, and they should fix it, try tomorrow, but if you want fast, ask ai to help, or search in internet to solve the problem I guess
tho, I don't have that issue
as right now
I thought that here i will find solution faster
Hey everyone - we have a new feature we'd love your feedback on. **Fact Checker **is now live on Canary.
What is Fact Checker?
After a model generates a response, you can use Verify Claims to fact-check it. This feature cross-checks each factual claim using multiple independent models and highlights areas of agreement or disagreement, along with a confidence percentage indicating how likely each claim is to be accurate. You can click on any highlighted claim to See Analysis, where you’ll find a detailed breakdown of which models agree or disagree—and the reasoning behind their assessments.
Please provide feedback here.
On my phone i can see the chat with side by side models that i just started on my laptop. So i think that im logged in. But it still trying to ask me to log in
Hey, can you help me with my error?
Hey, can you also add what I requested to update the page
I'm waiting
Sup
Which is better gpt 5.4 or geminine
Let's see
I think it's the Gemini 3.1 pro, due to it's fancy design and capabilities, I think it's better
What do you mean fancy design
Well, fancy design is like nice design for a website or smth
Hey, I have a doubt. Is there any way to use opus models for a bit longer time?
Yep
No
Nah
they might fail quickly
if you using them for longer time
Atleast some 10 prompts continuously
Nah
Well, that's means that it's give timeout limits
Can't do that
I wish I could join the arena team and help the platform improve @echo aurora would you accept me to you're team and at least I can suggest things or just help? you can call me like right hand man, also could I help for you're website and become developer too just in case to add features myself and have access to you're website? just privately of course and share me the link to this website, maybe not but at least accept me to the team?
You'll need to pay for the actual Claudius
@echo aurora Why do we got a caniry version of Arena
Same here
And I just found a sneak peak on the Arena website that Side Video by Side Video might come, did pineapple hide this?
I guess, he need to add it, and I hope it is, I also hope there's sneak peak about the direct chat for video arena too
Hello there!
plus he need to announce it to everyone about this sneak peak, since this can't be private
Sup
Well u can, just type in Generate a video with Sora 2 from a cookie
Well, I want to do this directly in platform, not in cookies thing or stuff
like find a arena.ai mode called direct chat and click on it, and then start using the model
so I guess there's no way
than cookies right?
Hey sorry for delay, was in a meeting. It's possible we add Direct/SbS to Video Arena. We don't have the plans on our to-do at the moment so I wouldn't expect this change in the near future. It may change later down the road.
Ok, at least add the thing I asked rather than this
like the model changing when using selected model to make sure user won't have to
switch to any mode to do this
It's possible that this experiment is showing positive results and we make this a perimant feature. I wouldn't be able to say if there will or won't happen at this time, it's something we're looking into.
Which models did you have selected in SbS? I'm curious if that will allow me to repro this.
I can’t make any promises about whether we’ll move forward with this, but it’s helpful for us to hear this kind of feedback.
Ok, if you have any information where my stuff going to be added, call me
Where'd you see this? I'm not sure what this is referring to.
Why do we got a caniry version of Arena
Canary is sort of a testing ground for new features before we decide if we release on the current site.
Can you (if you haven't already) create an issue in #1343291835845578853 and provide relevant information and ping me there? I'd like to avoid using #general as the main place we debug problems.
I dont have the chat to a specific project i made, how do i get the chat back or rip the assets from the page of the code preview?
Love to hear you’re interested in joining the team! I’d encourage you to keep an eye on our job board and apply to any roles that seem like a good fit: https://jobs.ashbyhq.com/arena
I was just checking my chats untill I saw Veo 3 and Veo 3.1 I think that's in beta rn
I would like to have a job at arena!
In Video Arena? I'm not aware of this change so that'd be news to me. Do you have a screenshot you can share of this?
No after 2 seconds it Dispeared
Yea in video arena
Thanks, I would like to have job there either, but as developer, mini one, who control website and try code there
Yeah that's odd. Sounds like a glitch.
If you get it again, if you could get a photo/video that'd be super appreciated.
So, would you like to give access, or not? at least accept me as suggester
@echo aurora Can I get a random role
You'd need to be hired first in order to be given that kind of access.
Ok I'll wait then, if I want I'll contact you
If you'd like to apply for a job you should apply through the site (https://jobs.ashbyhq.com/arena). Contacting me won't help as I'd just point you to our job board.
yay
Ok, I'll do this if I want I guess
Sounds good!
If you'd be able to find about how much we've raised can be found here: https://arena.ai/blog/series-a/. I won't be able to go into other details than what is already shared publicly.
We’re excited to share a major milestone in LMArena’s journey. We’ve raised $150M of Series A funding led by Felicis and UC Investments (University of California), with participation from Andreessen Horowitz, The House Fund, LDVP, Kleiner Perkins, Lightspeed Venture Partners and Laude Ventures.
Yeah arena doesent work when i use my actual UK ip
Also, add the new kind of job to be like: "Suggester", in the website you have like by link where people can get jobs to be you're part of team, Suggester is a guy who suggest the team features to add to platform and let users have what they wanted, at least to keep in mind for team to make it up soon, and second job to be a ai dev, a simple job for developers to use ai's to help coding or stuff to setup and explain it to team and add some code in the website that you have to work on, and update it with requested feature added
the ai dev job is just the one who doesn't know how to code or for some reason don't know programing language, so it's a convenient job for who doesn't know how to code
so, do it for the website by link: https://jobs.ashbyhq.com/arena
plus for my requested jobs, you can get a money just for adding stuff or doing smth as ai dev for ai dev job I guess
also, you should put those jobs by two categories, Ai Developing, Suggester, so @echo aurora what do ya think?
What seems to be the problem? Can you create a post in #1343291835845578853 and ping me there with relevant details?
Hey @soft sage The Video Arena Discord bot has been removed from the server. You can find Video Arena on our site here: https://arena.ai/video More information can be found in this announcement about the removal of the bot.
If the bot was here, we could generate videos here, and everyone could join the server without spamming some stupid stuff.
@echo aurora why gemini 3 pro is still working on lmarena?
We have a very capable recruiting team, and I’m confident they’re focused on identifying the right roles and candidates for what we need.
Ok, so would you contact them about this and send my post?
Hello @here
Sup
does direct chat need an account now?
Nice try pinging everyone.
Assuming you're getting the Something went wrong I'd encourage you to try these steps: #1417174113092374689 message. Would also note the issues with this model specifically has been flagged.
so those jobs can be added?
no, i don't get error
that why i ask
they say model is retired
but it still works OwO
on lmarena
API's still work or something?
It's an experiment. Would also note our #ask-here channel.
Hmm okay, we're looking into it then.
That's why, always use Arena.
yep
@weak dagger Why am I getting banned for ping everyome
Unlikely. Like I said our team is very capable of determining what roles we need filling at this time.
it is a great idea
i like it
login or go to side by side
Because you shouldn't be doing that lol.
I know, but I hope so they do what asked
Because, u can't ping everyone unless your a mod.
I understand.
ok, so do you add ability to make that model switch to another model if it's can't do that, like for example, gpt-5.4 is should switch to it's own image generator model, as you know
Dont do it, or get banned again.
make sure this will be for all models available for chat in there, also add gpt-5-mini and Omni models for open ai section, and add a section update that shows sections of models been in section, instead of having different models laying
Like I mentioned brining this to our #1372230675914031105 is how to share feedback.
Nah this is NOT A GOOD APRIL FOOLS PRANK!
I'd like to move on from this subject and let #general be used for more general chat.
Ok ok
Well, RIP Homer
yep, rip homer ????-2026
How to get video on direct
Video Arena is Battle only
Oh hello mr moderator 🥹
I have an on going like ticket or sm
idk
what's the best model for story writing because im trying to make up a story with few characters and convert some scenes to images so i need a good story telling model
@echo aurora what's claude opus 4.6 token limit on LMS before it errors somebody?
Gemini 3.1 pro
Yeah like 2 messages 🥹
Jk
Its bad
probably claude opus 4.6
its the most human sounding model
im not even sure if opus 4.6 is stable yet, so I'd recommend using sonnet
Fr
oh, i used to use 4o which was great
It's exactly why I wonder opus isn't stable at all,isn't the same as 4.5 opus
Frr, I still like 4.5 sonnet thinking because of the reason that if I make a discord bot or smth that take atlas 6k lines, I dont wanna get rate limited often
Hello - sure what is up? My Direct Messages are open so feel free to reach out there if you'd prefer. You can also send a Direct Message to @oak python.
Is it no longer possible to generate videos?
The Video Arena Discord bot has been removed from the server. You can find Video Arena on our site here: https://arena.ai/video More information can be found in this announcement about the removal of the bot.
Yeah I have a ticket on mod mail 🥹 ✌️
Pineapple can we agree pineapples are sub 3 best fruits?
Thank you very much
IIRC it's set to 128k.
It really doesn’t sound like 128k
It breaks at 4k
@echo aurora May u answer my modmail?
Hmm yeah that sounds off. How do you know it's not some other issue creating that error?
Yup, I'll do my best to get to everyone when I can.
Aight, take ur time
Well if lmys was more open on telling the error like error 400 etc it could definetely help.
Also tbh I'm not even the biggest pineapple fan 
L
This is a bit tricky. But we agree that the current error messages aren't as helpful as they could/should be.
Is it just me but vibe coding can get hard by it self. Like people think vibe coding takes 0 effort. But it takes still quite some effort
Why
Son it’s about experience
considering you have to give detailed prompts. The people who hate vibe coding are the typa people thag give prompts like "make me code 1k lines" 😭
Imr
Ikr
Has anyone here been using lm for like 2-3 years?
I for 2
w
Ye,you?
3 yrs
Crazy
Bro was among the firsts
Meh depends
lol I dont even remember how I found it
i was literally hard searching for some ai to use
jes
it was not so esy to finde a good free ai
My dog doing the ishowspeed face
BROO FR
I was about to quit but I found lm
Then I quit vibe coding for a while but completely forgot the name of the website [lmarena]
so I was doing the process of finding again 😭
my englisch skills are not the est im from germany
I mean i could tell—but that's completely fine.
I have many German friends lol
they're really chill
and rich for some reason
cool
Video generation can be quite expensive. To ensure the platform remains sustainable, limits like this are necessary.
Bruh
Why do u generate videos with ai in the first place
And William Shakespeare was defeated
is it for editing?
but i had use [lmarena] in the past then i forgot about it and now im back
I want to make my dog dance
LOL validd
it is e very cool konzept
Hmm what do you mean? The same reason we have multiple arenas for different modalities.
guys should I continue the edit?
No like I've never found a use for video generation
jes
wath do you mean exaktly
and wat did that do my python skils are not so good
Idk
Also guys that is a selfbot pls dont code one yourself
lock at this
i only gave it this and this prompt :make a professional ad
but i don´t like this ol in ol its a very god tool or i think the best but this is very not so god
and olso that we cant chouse witch ai video thing we want use (that is no cretic only thinks that cod be better but like i sad that is a great model)
It’s popular. Usage ranges from casual, just-for-fun experimentation to more professional use cases, such as marketing.
this presentatin html thing the ai had made but its in german i think its about the erth thats heating up
R we even allowed to send files here
What should I do if the neural network freezes and constantly displays "Generating..." without providing an answer?
I'm experiencing same currently
i dont now
Hey @tulip kindle @radiant hollow if you're getting this error I'd recommend the steps in this article. It's also worth trying to sign out/in as I've heard cases of this helping. Lastly, I'd point out we're running a Stop Button experiment, so if you go to our canary site (https://canary.lmarena.ai/), you'll see this feature to stop this generation that could help.
claude
i have made an ad prasantatin in englisch
lock at this
if you wat to see it you hve to klik on the tree dots and than doawnload if you klick now on the downloaded thing you will se the presentation
kan you make ulimetid pikturs
?
hi

Depends if there malicious.
Hi
What do you mean?
I believe it's asking you to input your bot token.
User token
That's bot token?
