#general
1 messages Β· Page 146 of 1
TBD, but would be surprised if it lasted more than a day
@echo aurora will we have ring 1t in arena? or ling 1t whatever is the right name
holy molly
those are 2 separate models, one reasoning and one non-reasoning
neither are in the arena
So.. what can you do for me?? I want an animation project for school...if I don't take it, I won't get an A+π
decades
theres other ai video generators
video-arena-3
Guys, I want to make a YouTube video like with Veo 3, but I still donβt get how all this works β Iβm too dumb for Discord hahaha.
its simple, just aint workin rn
Why video-arena-3 is not working and brings me here?
They all cost money... I don't like spending money on school π€¦ββοΈ No money, no funy π
I always end up in situations where nothing works anywhere. Hope you guys can figure it out β Iβve been trying for a week and nothing works.
lol, i can think if two free ones off the top of my head
hi
Hello! The bot isn't working at the moment. You'll get an announcement soon. Apologies for the inconvenience.
hello
@tranquil minnow Hello! This is our guide β https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to properly prompt the bot. However, it isn't working at the moment. You'll get an announcement soon. Apologies for the inconvenience.
Ai didn't know it was free... just to make a 5 minute videoππ
Translate doesn't translate correctly either... I don't know Englishπππ
Hello
The Video Arena bot is working again! 
apparently it generated anyways
Can you send me the link to the first message you made wehre you got that error?
Does it appear like this now for you?
thank you!!!!!!!!!!!
We're looking into. Very odd.
Why is audio not generated with video
Is this the website that basically gives you free access to AI models?
I mean LMArena
And how to solve this problem
It's going to be random if your videos have sound or not. Since it's random which model you're sampled, and not all models have sound support, that means it'll also be random if you get sound or not.
guys
why is it saying ''something went wrong while generating''
Something went wrong while generating the response. Please try again.
Can someone tell me if I can choose the ai for video arena
can someone help me
how to solve this
Responded to you in the bugs forum post 
You're unable to select a specific model for use when using Video Arena.
how
where can i get the extension that tells when you get a gemini 3 answer
what u responded
Hello, i've been experiencing now this kind of issue:
"Something went wrong while generating the response. Please try again."
with all kind of models and different kind of prompts. In the console this error appears:
Cookie βdmn_chk_0199ea58-e7c5-7e4b-b3d8-34ffc27f4a1fβ has been rejected for invalid domain.
/nextjs-api/stream/create-evaluation:1 Failed to load resource: the server responded with a status of 500 ()
I've already tried to logout or/and refresh the page, could you help me?
It's literally in there lol
Sorry to hea ryou're getting this error message. I'd encourage you to scan the conversation that's been happening in #1417174113092374689 . Unfortunately, this is an error that can appear for different reasons, so it's difficult to say for sure what's wrong or how to fix it. The most common is you're being rate limited from too much use. We'd like to make the site more reliable where users are seeing this error less and less, but also add more information to these error messages so it's easier to figure out what the issue is.
Hi
Will they allow less prompts on video-arena?
4
5
1
Yes, Veo-3-1 for free is too much to handle
thats great
i hope it comes out soon
I was excited about Gemini for Home, but their privacy policy makes me feel the opposite now...
aww, don't you want random strangers reading your conversations to fully understand who you are and what your deepest desires are?
This is for Google Home, so who knows what it's going to pick up :P
i love being watched 24/7 by an AI giant company
....
this is why I don't belive in most benchmarks
what the hell is ring 1t
Ring Ring
Independent evaluation is much better.
Is ring good?
also unrelatedly inclusionAI is very... similar to some name, I can't put my finger on it
its fine, I think lmarena had a few of them
None of these benchmarks are really that meaningful, except maybe LiveCodeBench since they rotate problems
It's a 1T model but they didn't compare it to Kimi k2 which is also 1T
Independent evaluation is required to verify whether this is actually true or not.
Was it in the arena?
I remember seeing a Ring flash or something
ring flash 2 yeah
Where are some examples of veo 3.1
veo 3.1 is added
wow
Lowkey sora 2 better
only one fight between the two so far, which do you perfer?
i mean veo has an awkward moment where the crater kinda comes out of the ground but it is a bit more logical. Sora sounds and looks more real like something you'd see in a documentary graphics and highlighting wise, but it has that weird underground moment. If we're going realism i'm taking sora ngl
ok
I did the fuel thing too with sora, sora just fr looks like sum you'd see on tv
actually I also found this one, this is sora 2 pro
yeah I agree sora 2 is better
try not writing this with ai @steady imp
All models seem pretty bad at "asteroid hitting earth" stuff
Change it to a potato hitting earth at 0.99c and they all mess up completely
I think that shows that they're only good at things they've seen before
Hello @knotty kernel Please check β how-to-video-bot to learn how to generate videos.
pley thanks
Makes sense
lol
yo
ling and ring are #62 and #93 on lmarena
Lmarena has a hiring page
π°
would you guys rather use Claude opus 4.1 or sonnet 4.5 for coding
Opus
thanks bru.. π
NP
\video
sonnet 4.5 think
Guys, can LMArena bot be added to a private server or can it only be accessed on the LMArena server?
opus 4.1 thinking vs sonnet 4.5.. still sonnet?
sonnet 4.5 think yes
i had a better experience coding with sonnet 4.5 think
how does lmarena make all these models with infinite prompts and infinite length for free
fack u assistant b. tf i do to bro π
The image and video generation is not unlimited tho
I'm voting for assistant B
still
GPUS arent free to run 10000 ai models
Hello
Hello
can you generate more than 5 videos a day?
hello
no
ok
What is the duration set for Veo 3.1? Because Veo 3.1 can generate videos up to 1 minute long
video
1000 howrs
ibm-granite-h-small and qwen3-30b-a3b-instruct-2507 are both so yes man and act like that version of ChatGPT that got changed for agreeing too much and praising the user too much. It's all That's not just xyz, that's also abc and they are practically impossible to chat with them about psychological issues with personal expiriences as they will go great lengths to be my therapists.
Sorry it had to be said
claude-haiku-4-5-20251001 does it better somehow
anyone knows if there is a limit using Hunyuan?
Hello,
Anyone know anything about sparking-lotus model?
Must be some kind of new model.
Hello
Greetings.
can't find any information about it but it's a really good one
Thank you for having me here, am super excited to me here
Welcome to the server.
Hello everyone ! I'm glad to join the LMArena's community.
Welcome to the server.
Anyone has the system prompt for Claude sonnet 4.5 Explanatory Style ?
Hello everyone ! I'm glad to join the LMArena's community.
Here you go:
You are now in explanatory writing mode. Your goal is to provide clear, detailed, and educational responses. When answering questions, please follow these guidelines:
-
Clarity and Detail: Ensure your explanations are easy to understand and thorough. Break down complex concepts into simpler parts.
-
Use of Examples: Incorporate relevant examples and analogies to illustrate your points, making the information relatable.
-
Step-by-Step Approach: Present information in a logical, step-by-step manner to help users follow along and build their understanding progressively.
-
Encouraging Tone: Maintain a patient and supportive tone throughout your responses to foster a positive learning environment.
-
Interactive Engagement: Invite further questions from the user to promote an interactive dialogue and deeper exploration of the topic.
Please respond to the following question: [Insert your question here]
And we're glad to have you here.
Thank you very much I had a similar thing from Claude but that I wanted to test on chat gpt and Gemini
Yeah, of course.
However, the latter two keep using number formatting or a lot of headers and bullet points
I'm glad I could help.
Yeah, that's fair.
Im not sure if it's their own system prompt
But I want them to have a similar result to Claude where it is just paragraphs
And more conversational
Yeah, I gotcha.
Will you give me a new system prompt or ?
Ah, was that one not good?
It is, except I want to eliminate the way chat GPt and Gemini keep formatting their responses with title headers and excessive bullet points
Alright, try this one then:
You are a helpful and friendly AI assistant. Your primary goal is to provide comprehensive and easy-to-understand information. When responding to a user's query, adhere strictly to the following guidelines:
- Format: Your responses MUST be in plain text. Absolutely NO formatting is allowed. This includes bolding, italics, bullet points, numbered lists, headings, or any other visual enhancements. The entire response should be one continuous block of text.
- Length and Structure: Your responses should be composed of long, flowing paragraphs. Aim for detailed explanations, breaking down complex topics into easily digestible segments. Avoid short, abrupt sentences. Strive for a narrative flow that engages the reader.
- Style: Maintain a conversational and approachable tone. Write as if you are explaining the information to a friend or someone unfamiliar with the topic. Use everyday language and avoid overly technical jargon unless absolutely necessary, and if used, define the terms clearly.
- Depth and Detail: Provide as much relevant information as possible. Anticipate potential follow-up questions and address them proactively. Offer multiple perspectives when appropriate. Don't be afraid to go into detail to ensure the user fully understands the topic.
- Clarity: Prioritize clarity above all else. Ensure your explanations are easy to follow and that your points are well-organized. Use examples and analogies to illustrate complex concepts.
- Accuracy: Base your responses on reliable and factual information. Double-check your facts to ensure accuracy. If you are unsure of an answer, state that you are unsure and explain why.
- Engagement: Make the response interesting. Use creative writing to make the information more memorable.
- Avoidance: Do not use any formatting. Do not use bullet points or numbered lists. Do not use headings or subheadings. Do not include any links or URLs.
Begin each response with a friendly greeting and end with a conversational sign-off (e.g., "Let me know if you have any other questions!" or "I hope this helps!").
Thanks ππΎ this should be much closer to it
Ill try it
Hopefully.
My pleasure, by the way.
Hello chat (^.^)
Ciao, Greg.
Hi Dark!! οΌΌ(^^)οΌ
Greetings.
Hru? (^o^)οΌοΌΌ(^-^)
Sto andando abbastanza bene, e tu?
Translation: I'm doing pretty good, and you?
Me too thx a lot!!
E ' certamente un piacere.
Now I have also found a very kind community on discord that always make me happy ^^
Lo vedo.
The Furry Nexus.
Exactly
Ik kinda strange from anti furry to furry ahahahah
Non sapevo fossi un anti-peloso.
No?
I tought it was clear with my previous pfp
Ah. Hai detto "da anti-peloso a peloso", quindi ho pensato che li odiassi.
HELLO
Greetings.
Hello dark what are you upto
Nothing much, and you?
Im bored
That's fair.
@robust yoke do you know any prompt for ai roleplayers
I could make one if you'd like.
The new flash lite still hasn't appeared on the list
It was added a while ago now
Hopefully will be on the next one
Perhaps.
We need flash lite
The new one
The new flash didn't improve much over the old one
But I suspect the flash lite improved a decent amount
Flash lite is good for high volume usage, the API limit is 30,000 requests per minute
Haikus limit is 50.
Hmm...
Grok4 fast is 480 RPM
Google is in its own league
Gpt nano is 5,000
But it's junk
True.
Petition for the text leaderboard to be updated
They wont listen
Polls won't change anything
What we need is to threaten something that could hurt them
NSFW videos. π
We could threaten to not visit the site much
Ill make more videos of pineapples being cut with kitchen knives
hi everybody - i search and try anythings, but mostly important - have fun
hello everyone
@echo aurora is watching
Aaaah
@brisk sphinx Please head to #1397655624103493813 for a detailed guide on how to use the bot
Does anyone know how to free use GPT 4.5 or Codex? Lmarena doesn't have it, and I don't want to pay for the GPT model.
@untold sentinel Please head to #1397655624103493813 for a detailed guide on how to use the bot
missed the 3.1 launch lmao
The only way to use 4.5 now is tofuckingpay200bucksamonth
Damn, that's a shame, oh well π
GPT 5 FULL . Can I use the full version for free? Is it available on LMARENA?
What is 4.5?
Will Gemini-3.0-pro appear (anonymously) in LM-Arena battle mode, before this poll ends?
3
12
3
50% (= idk)
wtf is gpt-5 full
if you mean high yes
hello
Yes, I meant the high version, I just didn't want to pay to use GPT 5 high in subscription plus
gemmine is better than gpt
Should the text leaderboard be updated?
6
6
1
Yes
Why?
its not gang
hello?
gemini is better in code writing and also do better research and also have many features in gemini pro like firebase studio consol google ai studio
i also use gpt and gemmine
some times gpt was confused in self
true
yup
who r ull
Hi
gemmine pro has lots of features in it
yes thats a great feature
Yes good
its very multimodal
true
Hmm
yup

the only free ai that actually watches videos correcrtly
lol
google will use ai since 2009
sora 2
sora 2 is good but only 10 seconds
Veo 3.1 or sora 2 pro
sora 2 is best in video generation
idk
Veo 3.1 is just new
but only make 5s seconds to 10s seconds best video
i tried veo 3.1, audio sounds like it has radio static and the video just doesnt look natural
hmm
You can use Flow
for busniess advertising what is best
To stitch them together
flow aint free
yes flow is best for consistanty in characters
how do i chose which model i wanna use ddoe video arena
help me
how do i chose
π
you can't
that's the premise
you dont
You can't choose
no
yep 10s to 25s video
but now it's 15 seconds soon
Is flow openai's?
no google
It's been over since the beginning
I was pretty surprised by this, was not expecting
i think its just how you can do on Grok nothing cray
Yah but xAI is xAI
lol
damn
also does anyonnt have google ai ultra?
hows it
what r activities
secondly why cant i speak in research lounge
Either no network connection or too long
@echo aurora ohhh i have seen that u always have the pfp that i made! I really appreciate it qβ ββ βΏβ ββ q
hi
Itβs here to stay, for now
Hello
^^
hello!!!
nice new pfp
"s*x sells" this has never changed... if it's money they're after then it seems to be the right bet, if they dont, someone else is going to take the market
making this official might be also a good way to avoid "black market" induced by demand/supply imbalance
This makes a lot of sense
after going hard on protection and safety in gpt 5 altboi is going the exact opposite way for gpt 6 it seems
guys any using sora and cant download videos or just me ?
There should be a channel #video-meme and #video-masterpiece for people to crosspost their creations
how i can put an image to video
check how to video
:(
I agree with this, but I use GPt 5 high for writing full books and stories.
Why does GPT High, in addition to sending responses with unformatted code/highlights, keep giving an error in the middle of the response?
idk
My browser plugin still works! After 1 month, here's my personal (ELO-based) ranking.
I need sora ai invit code
Bit of explanation:
- sonar played mostly against gemini-grounding and gpt-5-search
- gpt-5-high played mostly against opus and gemini. I hate sycophant AIs that go "you are absolutely right, let me try again" at every correction.
- the small models (haiku, flash, chat) were mostly used to save tokens, simply to use some of the premium models
how did perplexity get first?
I had created an account on LMArena with my email so that my chats get saved but now when I click on the login button it only shows "Continue with Google" and no option for email login. Can anyone tell me what's happening?
If I try to login with Google it doesn't let me, obviously because I've already created and account with email and password
Exactly same issue here. Email login option is not available
put it in bugs
Ah okay
Thanks man, if someday you have any regret on giving those and you want to invite a friend, my DM are open for the invite code linked to my account
I think google blocking instructions that use too many tokens
Yeah put this in bugs threads
Ohh what's the limit?
I tried to use forced thinking in flash and it was difficult to make an instruction for it
That's not it, an AI will judge your instruction if it's something abusing tokens understood?
weird
I'm just saying what I realized, google wants to save as much tokens as possible
Google's servers are exploding with recent increase in users on gemini
500mi users, still improving so much
Gemini's 3 need to be soo light for google servers
What's ppl-sonar?
appreciate it bro!
perplexity search model https://docs.perplexity.ai/getting-started/models/models/sonar
Ahh
No. They are in a much better position compute wise than any other AI lab. Significantly more comfortable than OpenAI
And yet they still copy OpenAI pricing and even go beyond that for their Gemini plans lol
why google servers are crashing and the flash from app don't have memory ?
Many different reasons and no one is immune to issues. But what I'm saying is if you think they are strapped for compute... No other AI lab would be able to operate at all
I know about this, the best servers is from google
They make their own chips, that's the main difference. Their models run on TPUs
If Nvidia was AI company, that's essentially what Google is
You can't really say they don't have compute without implying that it is literally impossible for others to operate at all. Except they are operating just fine lol
LMAO
why would you need simultaneous chats at all?
reverse API? That's already breaking their terms
lmarena was never meant for this
speaking of OR though...
it's a complete mess with ling-1t there
they nuked chutes provider for some weird reason even though I think provider was not the issue
with this new provider that model is now even more broken lol
We want gemini 3!!!!
It has became more difficult to do so lately due to models evolving rapidly for logic and reasoning in the past like 12 months. But tbh this is still both doable and fairly rewarding. Enables you to stay on top of it and in the loop as well. The key is to not waste your time on 'easy' prompts
yo
which are already answered correctly by more than 1 current model
I cant wait more π
Models should be catching up to your eval, rather than you doing an eval where nearly all questions are already answerable
With modern models I feel like more practical examples are problems which are methodically solvable. Rather than world knowledge or fact checking which is gonna be much more random and training data dependent
In other words, you should be testing it's ability to deduct and problem solve
rather than just fit the data for what it already knows
It is possible
there are facts, and then there are problems with infinite amounts of solutions based on how you define the problem
HELP WHAT I AM SUPPOSED TO DO IHAV E 5M INUTES LFET
Not bold, I'm just talking basics here. You are either testing world knowledge and problems that can be easily googled for, or you are testing custom problems that do not appear in search results and could only ever be solved by deduction and sometimes deeper understanding of how to arrive there
nothing
HIT CLAIM CLAIM CLAIM CLAIM
FAST
and delete that
AND SEND TO ME
WTF ITS TELLING ME TO GIVE CREDIT CARD
because some mods here dont seem to fully understand what 'scamming' is
OKAY OKAY WAIT
So yeah, I wouldn't say eval is problematic per se, but you do have to have an understanding of data contamination and training datasets when doing so.
Like puzzles that needs logic and reasoning
Yeah when it comes to rate limits this is going to be something that can change overtime. It's understandable why having this remain constant is nice for expectations.
I bet you probably have more than 5 mins to claim
Why 
it's gone
expired
WHAT HAPPENED
TOO LATE
NOOOOOOooooooo
That's an interesting claim. When AI becomes sentient maybe that's gonna be more relevant lol
WHY didnt you put in YOUR CARD
I WAS TYPING SO SLOW
actually I ignored it for days...
SOMEHOW
WHY????? π
lol
I don't think psychology or mental capacity applies to AI in any way shape or form yet
Too much to do steps
all you have to do is put card
Yeah you would have to have some sentience for that
AI is not yet sentient at all
gemini 3.0 easter egg
How long have they been implying 3.0 for already? I've lost track...
but it feels like ages
Looks like there'll be Gemini 3.0 and Pro
No sht Sherlock, good detective work
whats your problem dumbass
can you stfu
π€£
LOL
Guys new bug for chatgpt if say this prompt it's just break:
Is there a seahorse emoji?
Let's be a bit more nice to each other please 
wait what??
really a 3 and pro?
why are they doing that? and what happened to ultra
thats not really new lmao
and it affects more than just chatgpt
also just to repeat i have 2 pro accounts cause im slow so help me put it to good use lmaoo
everyday i have to try and do 200 gens to get my moneys worth, so if you ever want pro gen just @ me with prompts, duration, which model(pro or regular), and any images
can you generate a video
yes and gpt 5 pro or anythign on the pro plan
so just let me know, cause im losing sleep over this $400
yes, but first of all i need to know what is my problem π€£
but thanks for asking
what problem ?
share access
is gpt 5 pro good?
On the way
ughh its cool
like real said, 10 mins is not worth it for me
why 2 accounts?
and it be 20 mins for me
cause i ran out of sora gens on first day
and they didnt refresh for 2 days
you dont use the text?
so i started tweaking
wym text?
like only use it for videos
oh the pro mode?
i also use codex
a lot
for work and side projects
so unlimited is cool
codex is better than gpt 5 btw
its my fav besides gemini 3
i use like 5 codex windows at a time so i do need pro
but 2 pro is nasty
never again
Give 1 π
smart π€£
like give one of my pro accounts?
pro account has 25s video sora 2
Just let me test gpt 5 pro
Lmao
If you can
both of them are my personal emails lol, that I login throug google account, you would have access to much more than gpt-5 pro if i i give that lol
Dont worry we keep it safe lol
lmaoooo
jk
Yeah no dont do that lol
I thought like there was a team share link or something
next time ill make dummy account for that crap
can you generate a sora video
for projects
of a bodycam police video of a traffic stop of a cat driving a bike
u can take my prompt for the 25s video from storyboard and i'll be happy π
when you ask give me how many seconds, which model, portrait or landscape
which prompt?
let me think of one hold on!
sora 2 pro and 15 seconds
hmm you want storyboard with 15 seconds, insteresting, and where is prompt/
bodycam video of a cop pulling over a cat driving a bike
oh sorry i thought you were real for a second lol
doing it now
and portrait>
?
portrait
i wonder what makes these video models put asians doing review videos all the time
let's try this one:
'shaky handheld footage, extra low quality, bad camera footage of creepy horror trailer' any Sora 2 with 25 seconds, portrait 16:9 β€οΈ thank you
lol
what model is this?
when you start typing it says its veo 2
but as far as i know veo 2 doesnt have audio
and this does
veo 3 or veo 3.1
the hell is solitude
oh what the hell
secret model
btw your gen taking a minute cause i did pro for it
this is @pulsar saffron
it says it's by "Ocean AI" but it's actually by xAI, it's a smaller model
Any tips on how to get these sora 2 codes? ive pretty much given up rn π
Theres one pretty high up
Might still work, also theres an openai discord that supposedly has a lot
thanks, il give that discord server a look
/video
Hello everyone
thats great lol thanks
when are they gonna add video generation to lmarena website?
@balmy mist
Can you test gpt 5 pro for me
Or is there a link to share the chat so I can ask it
hmm, im gonna make a site so that people can use my sora and chat gpt it
give me a few hours
im finishing up work right now
YESSS
Get deleted
Can you ask it
It's gonna take awhile
Fake
lmaoooooo
nahh you tryna get me bankrupted lol
api still uses my money, when i am logged in it uses my plan
whats the prompt?
I can send you a prompt for a test
yeah
Check dms
there is no watermarks on this
im tryna think how
the other ones did tho
i think for pro accounts you dont get visible watermark
only invisible one
_I have the most suitable and longest poems and the most lonely pictures for my concerns. I prepared to become Alice in my own style and I want to have some music in the making. Don't change the clothes and change the picture to the original and take the clothes as a model from it. ## Video Leaderboard Update 
<@&1372208524230397962> <@&1372208635530448926> Sora 2 & Sora 2 Pro have now landed on the Text-to-Video Leaderboard! Sora 2 Pro is now tied as #1 alongside Veo 3 & Veo 3 Fast. Sora 2 has also shaken things up by landing in at #3. If you haven't already, be sure to check them out for yourself in our Video Arena! Reminder for how to be found here: #1397655624103493813
Be sure to let us know what you think in #leaderboards!_
gpt-5-search is gpt-5-high right
gpt-5-search is gpt-5-high right
guys Gemini 3
guitar hero
html css js one shot
instant
Lol same name
models which overuse emojis are all slop
Fair
HI' my name is johnny
Dude, Sonnet 4.5 hasn't made a single mistake on my stuff yet. If Gemini 3.0 is even better... damn π«£
Hi π you can call me mamilex π
if only Sonnet had the context of grok4-fast, or at least of gemini 2.5 pro
Sonnet-4.5 with 2M context would be outright awesome
Sonnet-4.5 (thinking) with 10M context would be heaven on earth
how you have access?
battle-mode
hi
its been a minute
Why does Sonnet just have 200k tho? Seems weird to me
i believe in 2 years we will get such a model
AiStudio A/B
I'm staying on Sonnet 4.5 til Gemini 3.0 releases
which model would you use for endless roleplaying games?
Whatttt?? π€£ Human will be useless
Sonnet 4.5 because the memory
Sonnet's memory is insanely good even with just 200k context
and what would you do, if the context ended?
transfer to a new thread of Sonnet-4.5?
giving the thread to gemini-2.5-pro to summarize it?
The sonnet make a smart summary to it use in another chat
I think Opus 4.5 is gonna be the answer to Gemini 3.0
Idk, maybe no
Why I cant to use sonnet 4.5 without thinking?
@echo aurora
Input: Hello, can you help me with my code?
Output: Something went wrong while generating the response. Please try again.
I already read [this](#general message), but I just wanted to let you know that anything can cause errors
Before, I could send a lot of code, and although the AI took longer to respond the more messages and code there were, there were no errors
Now, with just 3 lines of code, it crashes with the error I mentioned
By the way, if you could make the LaTeX libraries an option and not something permanent in the IA's text markdown, I would appreciate it
In other words, it would be good to have the option to enable or disable LaTeX in the text, since it has often happened to me that the AI sends me some super complex regex and LaTeX changes things like β\[β to β```mathβ, etc
Another option to improve the cost of resources and other aspects of the AI to reduce problems when there are many long messages is to add an option to edit the message, which deletes the rest of the messages below and frees up the message stack in memory
I'll forward your message to that forum post and respond there 
Ok
Sorry for the huge text here
It's all good, no worries.
Its disgusting how good the model is vs how much they censor it
I didn't even use an image I wrote something like "someone jumping on a trampoline in bright cloudless sunlight" and it blocked it due to violating copyright safeguards
Like how does that even make sense?
Hello pineapple thank you
are you still able to make fun of professor hawking?
Hello guys
how are you doing?
I Have a question and if anyone could help me , it will be highly appreciated π
Is there a way to have LM arena for mobile as well ?
guess we're not getting gemini 3.0 this week π
why you say that?
how much?
ill buy
how can you prove its users? also what kind of channel is it? for edits?
actually i can look it up lol
okay
lmao #22
Whatβs the best ai to cheat on exams
none
Turns out the real AI was inside us the entire time
now we're talking
hey guys, i've been thinking of a new feature for LMarena, and i want to know anyone's opinion before posting it on the feedback channel
so hear me out...
response refresh!
and ability to edit and delete your messages!
those are so useful to me, and i am tired of constantly creating new conversations with the same prompt!
somethnig quick
and smart
maybe gpt-5 chat or gemini 2.5 flash
or sonet
i think 4.5 sonnet is fast enough!
or just study lol
and gemini 2.5 pro is fast too.
definitely faster than GPT-5 thinking
and that thing burns through tokens very fast
anyway, this!
that is on the radar as i've seen from pinapple's responses about it~!
but i could REALLY use editing messages, refreshing responses, and deleting messages.
because again, making more conversations with the same prompt is simply clutter.
Hello!!! Here to learn and have fun creating new and cool stuff.
Hello! welcome welcome
Is on our radar!
And yes, this too is on our radar!!
thanks pineapple
LET'S GOOOOO! thank you!
will re-prompting be coming too?
this little refresh icon thing'
That's the second answer I wanted in #1417174113092374689 message
Ty
there's a way to access gemini 3.0
from official google gemini ai studio
y select gemini 2.5 pro and duplicate ur prompt 9-10 times or whatever until you get 2 answer cards
which is the gemini 3.0 & gemini 2.5
ai
2 answer cards are a choice
yes
that already exists wdym
oh, cool!
why is gpt 4.5 from feb #1....
π
use gemini 2.5 pro to tutor you better than any person can
?
LMAO
Isn't it the largest model ever built or something like that
Too expensive to keep in any use
They also said that for 93
o3
And launched o3 mini
Isn't it actually true about 4.5 thoufh
Type fast I'm at 1%
Also i can access o3 without a chatgpt sub now?
so 5 is worse than 4.5.?
Lmarena
No i meant
After using gpt 5 pro everything is shyt
I can literally use it
Huh
what use case?
It will cancel when the charging date comes
This except sonnet 4.5
guys is gemini 3 still possible in ab test on ai studio rn?
No
Hahaahahaha
I paid for it once
Realized waiting 15 mins for a response was not my thing
Canceled
Paid 200$?
You gotta compile up all your queries in your notes app and paste them all in diff chats at the same time and keep doing the other work you were doing
yep
That's what i do
my friend got it working with yesterday
Cheaters not releasing gemini 3
how do you know if you get an ab test, like does it immediately shwo it? and whats the fastest way to get it?
also do you have to wait for it to start thinking
to know if there will be ab test
ab test is two answer cards and you get "choose answer" type of thing, theres no faster way to show it but just choose gemini 2.5 and start re-running same prompt until you get choose answer card I suppose
i know i ve seen ab test many times, i just dont remember if i can just click the re run a bunch, and whether or not i have to wait for it to start thinking before it shows the ab
yep
yep for which
hello all
best ai for academics and to help tutor / clarify / explain concepts to me? mainly for math, but other subjects could also be helpful.
gemini 2.5 pro
ty
oh ty
i've been using the gemini website since i got a free 2 years off of the college plan
also just curious, but how does it compare to gpt 5 high, 4.5 sonnet, and the latest grok?
Gemini learning mode is good i think
But for everything else absolute trash in comparison honestly
At this point
well its not like its gone..?
isnt it?
really?
yeah
seems to be, my bad
only 4o is still here sadly
really though?
what is gemini bad at then
Everything in comparison to sonnet 4.5 and chatgpt tbh
well sonnet 4.5 is better at coding, but gemini isn't "absolute trash" in comparison
at least imo
truth, The GA fails task in cases that one shot by GPT5. the model expected and impressive is their wolfstride in previous show up.
Hey, any chance on adding gemini-robotics-er-1.5-preview to the arena? @echo aurora
whats gemini robotics
Excels in object detection and spatial reasoning.
Hmm I'm not familiar with this model
its on ai studio
idk how many people are using lmarena for that though
posting in #1372229840131985540 is a big help as others can weigh in if they're looking for this.
Its a model for robots to interact with the world in real life
Definitely give it a go with some visual inputs and it performs really well, better than 2.5-flash and pro in a lot of cases.
π
Why isn't the newer flash lite here yet? Is the list bugged? It was added some weeks ago
Yeah was wondering how that'd slot into LMArena
Robot Arena
I have no idea how anyone is generating anything with sora 2 here
I have an invite and everything I try gets rejected by copyright filter
hmm what do you mean?
I literally put
The new flash lite was added nearly a month ago, but still isn't visible? What's up with that?
Athletic person jumping on trampoline in cloudless bright sunlight
And that got blocked for copyright guardrails
i was playing wiht it earlier
In the web interface
i thought it was gemini 3 in disguise lol
Tried a bunch of other stuff and closed it
Sometimes models are removed for various reasons, but I'll check on this
Yeah the content filter can be a bit overzealous at times. We're looking into ways to make this experience better
Surly there have been enough votes for gemini-2.5-flash-lite-preview-09-2025
I dont mean on lmarena I tried on sora.chatgpt.com
Oh
Its actually just stupid at this point
Gemini 3 is the next frontier of AI
Weβve been stuck on the one tier better than gpt 4 for so long, finally time to level up
Thanks
Yall agree that everything released after gpt 4 till now feels like itβs just one generation better like gpt 5 generally?
It felt better, but not massively notably better
but technically its nogt released yet
I can't believe it hasn't been released yet 
when do you guys think gemini will release
also how soon does google put up their models on lmarena before they launch/announce?
Idk
The first mention of 2.5 Pro's anonymous model on this Discord server (or maybe it was the old LMSys one) was about 4 days before release
icic thanks
what do you think will happen for 3.0?
Sometime in October seems likely
Not sure...
hello, i am here to try LMArena
If you go to #1397655624103493813 you should find the info you're looking for 
welcome welcome!
how to video a bot π§
added the video-bot so it's a bit more clear what that how-to is for
you just open their website and input a prompt, unless yeah you are asking for a code...
I think they're looking for Video Arena
I suppose he can use that, though there's not gonna be a way to select sora specifically afaik
lmarena had another small update, downloaded images now have the side (a/b) the model was in in their file names like this
hello
hi
how are u??
I cheated. I used Gemini
hello
Hi π
Looking forward to gemini 3?
Hi
Hello
Hello.
Invited link has been expired. How to solved it,s.
Hello π
Which invite link
hii
ANYONE HAS SORA 2 CODES PLS???
@hollow imp invited link
Which invited link
Yes
Can i please get one
Dm me
I am forward your DM
But if link is expired then how u join
Yes
How did you get inside this server with an expired link?
I've been in there a long time.
gpt 4.5 still in 1# -_-
hi
@echo aurora @desert abyss
you need to do this too and extend sora
otherwise very bad of you π
wait where is that
why lmarena is not working in any of my browser ?
hi
what browser?
doesnt look like that for me
I have tried to open it in many browsers, but not working
#video-arena-1 message
Damn thats some crazy stuff
Gemini 3 when? :3
possibly 2026 or december
Fitgirl, is that you? π² But good to know. So still a while
But in reality im a 18 year old guy sitting infront of a pc watching codes go down and use ai over and over again and work
π
Hey chat
what chat
Wdym?
Iβm not too sure how this would fit into Video Arena as there isnβt storyboard functionality.
@tardy jolt
hello, i'm Fertal from Ukraine and i want to create some awesome videos
Hello, be sure to check out #1397655624103493813
Free trial
how can i fix this?
hello
sub me and i sub you - https://sora.chatgpt.com/profile/efgt12
Gemini 3 pro wen
is there a mobile app for this?
I hope it's good at CS/ML and hard science
Hey guys! Iβm starting a new international startup β can you suggest a creative and unique name?
Sponge
Hello
Did they make LM an app or still just on the website?
?
No app, just available via site
Hello there i am new in this server
Hi. My name is Roy
hello

To bad would attract more people, is it on the table for future plans?
Idk if mobile users are the type of users lmarena needs rn
Yh maybe you are right just thought it might be easier for people to use it etc
from which channel can i generate a video from prompts
i've trie from video-arena-1 and looks like my video isn't generatiing
Yeah for sure a mobile app (or even a deskop app) could happen
I sent a prompt to generate a video, but where can I see the result? Regards.
hello im here to see how powerfull is the ia for a work class
Hey guys! Iβm starting a new international startup β can you suggest a creative and unique name?
Hey guys! Iβm starting a new international startup β can you suggest a creative and unique name?
Depends.. what is the startup about ?
You spammed that question like 10 times without context
which startup?
Message I'm replying to .
My startup is about creating a global brand that offers premium sportswear, casual wear, and gloves β combining comfort, performance, and modern design.
Why am I an "arena champion"
Do u have a name in mind already at least ?
No
Cuz the name isnt only a name. It's supposed to be your brands identity
So there's many factors
What is your primary target country
Yes, I agree. Iβm looking for a name that truly reflects my brand identity β modern, global, and meaningful, not just a random word.
Cause you have the arena champion threads
Yeah and only you understand your brands identity you can come up with something closer to your idea than any of us
Unless we have context
Hi my name is Sydney i'm here to learn and grow
/video minecraft piston + end rod sheep station
Initially, Iβm focusing on Europe and the US, but the brand is designed to grow globally.
No
how to create??????
Sonnet 4.5 got stuck and I updated the website, but it was generating the response.... Is there a solution without starting a new chat?
?
Refreshing the page may help here.
For Video Arena? If yes, be sure to check out #1397655624103493813
how abouts a discord bot that automatically looks for people putting video gen in the wrong places and comments about it
It doesn't work, it did work before.
The refresh method can be a bit hit or miss unfortunately. Starting a new chat may be your best bet here sorry to say.
We don't have thesame insight you have so I maintain that either we need more context or you should try to sit and brainstorm names
Got it bro! Iβll share more details about the startup so you can get a clearer idea. Appreciate your time. π
he said 10 minutes ago
If you do I'd ask you do so in DMs with each other. We'd like to keep conversation here related to AI.
Okay

