#general
1 messages · Page 153 of 1
it was written by claude 4.5 sonnet lol
GUYS I GOT GREAT NEWS
GEMINI 3 WILL BE RELEASED THIS YEAR
IMO, in coding, claude 4.5 sonnet is probably the best released model right now
damn thats surprising
i thought its gonna release with GTA 6
fr i have done research like 72 hours to find out
We got silksong before gta 6
dont remind me of silksong.. too much hype but it didn't really like it 🙁
whats silksong
not sure if you are trolling.. if you are not, we know that gemini 3 is releasing. But now sure when... my theory is Nov Ist week
game
well he's obviously trolling
it's a joke obviously
only expect trolling when you talk to craig lol
how do you get the arena champion role
lmfao
i had it until one of the mods banned me for no reason
Fill out a form
me
OMG YAY!
@fleet lintel claude 4.5 sonnet thinking 32k looks HORRIBLE, Lithiumflow is way ahead these models
also it has no assets
huge diff
A cinematic rural scene showing a golden lion standing in a wide green field during sunrise. The camera slowly pans across the open field with tall grass swaying in the wind. A farmer wearing a traditional outfit cautiously approaches the lion with a rope, trying to catch it. The lion growls softly but doesn’t attack, creating a tense and dramatic moment. Realistic lighting, 4K quality, natural camera motion, ultra-detailed environment, cinematic color grading, dust particles floating in sunlight, realistic animal and human movements.
DeepSeek r2, V4 and GLM5.0 too 🙂😂
Hi guys, I'm strugling with prompting gemini and qwen. I want to replace the background of photo A with the background of photo B, BUT I don't want a simple green screen kind of swap, I want the the camera angle and lighting to be adapted so that it matches the foreground of photo A.
chatgpt understand it but has some limitations like no 16:9 and darker tones. I can't get gemini and qwen to do it, they just don't adapt the background and sometime they adapt the foreground.
Sorryt its just too complicated for ai to do rn
where u got that
why are you doing this
30 videos of what
gimme a prompt imma do sora 2 for u
sora 2 videos
How come chatgpt can do it
Banana was designed to be photoshop
u just reply with the most stupid thing ever lol
try to prompt dota 2 gameplay, where a player is raging and calling teammates noobs and feeders in ingame voice chat
I love sydney
alr
i can prompt 3 videos at once
u want more?
prompt all 3 with it and pick the best generation
aalr
imma do the same prompt
3 times
here minecraft gameplay, remember i typed "just moving around" thats why nothing interesting happens
lol, here are the 2 generations, the 3rd one seemed a bit off
game looks completely broken, but voices and whining is spot on 😅
yeah cuz dota 2 has alot of movements
imma try some mario 64 gameplay
Sora 2 Pro?
no its the free sora 2
i think its very good
opus 4.1
damn it looks good
I KNOW BRO
Gemini 3 Pro
u made it on websim.com
❌
bro websim doesnt even have opus 4.1
only sonnet 4.5
it used https://win98icons.alexmeub.com/
hmm interesting
i wonder how good lithiumflow would perform on it
actually now that i think about it
opus 4.1 with a side of sonnet 4.5
sorry
although i hate it when it does this
THATS EVEN WORSE BRO
lol
Oof
idk why they have this error so frequent, unlike other models
DUDE
IM GONNA
SCREAM
???????????????????????????????????????????????????????????????????
you have broken lmarena dude
claude models are heavily used, thats probably why the error happens
Weave
lmao
yes
Thanks
Gemini 3.0 Release Prediction?
5
8
1
26th Oct
pineapple do you have a few minutes to spare? i wanna show you my clicker (windows 98 themed)
nvm i rewrote
ok wait
let me record
@normal abyss the gemini 3.0 will be release just in december bro
😭
stop this "toxic hope"
Preview in November
note: if you click the left option on those scam notifications you get redirected to rick roll
also dont ask why "hello!" is there
hold on
https://d.uguu.se/MqESCeia.mp4 sorry for double ping, i also havent shared winamp this time, sorry!
Gemini 2.5 Pro sessions are still consistently failing after ~30 turns. This is not happening on Google's native AI Studio, which points to an issue on the LMSYS side. It feels like your platform is either hitting its API rate limit/quota with Google or has implemented a new internal session limit for Gemini. Can you please investigate this?
thanks for sharing! Would encourage you to share in #ai-creations as well!
oh right ty
btw i used opus 4.1, sonnet 4.5 and gemini 2.5 pro when claude was sh*tting itself
which one was the one you shared?
just use the AI studio website if want to use it for long term sessions
because LMArena is not designed for long term sessions
this ver was made by gemini 2.5 pro.
btw!
of course its broken
wait
IM NOT EVEN JOKING THIS IS SO LAGGY. 😭
the site overall or models responding? both?
models responding
probably because i just pasted my code for like a billi times
btw do you understand whats going on here ❤️
ok im pretty sure the bot just stopped working what
How many chats did you do? is it all models?
I'm going to need more context
i asked the bot to optimize the code
also i have spent 15 minutes over a single problem where the progress bar was just making the game not work.. at all
idk i dont rember
I'd encourage you to create a post in #1343291835845578853 and provide the relevant info there as it'll help us keep things organized a bit better.
wait, creating a post in bugs for what?
sorry im dummy
don't play with my emotions
Looks like low level 💩 on an unifineshed model in alpha testing to me.
skill issue
That is IMG to video, where the image is not only colour corrected but of enormous size. = Sora is inferior to a dozen other AI's where exactly the same prompt were used.
Main use-case of sora2 is creating social-media like content. Not many direct alternatives for this. What you are trying to do here is not what it was made for
AND I STILL DIDNT FIND THE PROBLEM.
Just use the right tool for the job I suppose
......if I had the slightest intension for seppukum I'd go for it already. 😺
sora 2 is bad at img-video, but very good at creating videos itself
gemini 2.5 pro?
🙂↕️
ask claude 4.1 opus to find the error
Sora1 was much less specialized and more general, more in line with what you are trying to do. But it is outdated by now. With v2 they branched towards social media shorts to distinguish themselves
hmm
It's no coincidence that sora2 website is basically social media site lol
yea sora 2 is good at creating videos, not editing images to videos
i hate this
is lithiumflow gone from webdev arena?
...just to get a result that is regurgitated from things already made and seen a zillion times before? What is the point of that? None whatsoever
That is why an AI need to be able to do the simple work of animating a scene where all the work have been done in advance by providing an image according to the vision of the author, composer and visionary.
And that without downgrading the highly detailed artwork into low resolution 💩 as was shown.
I'm thinking the same thing, I can't find him anymore today and I've already found him about 5 times today
Claude 4.5 Sonnet is better, or the same at the very least
And it's a lot cheaper
Good Day All Jim here,🥰
can someone delete my account alread ybro
hello
@echo aurora the new code arena is not using react/nextjs anymore?
its generating .html files
it actually got the failed .html wrong
here is normal
wym?
yes
it got the tags wrong too
thats not the new code arena
what's new? wtf
ohhh
That's correct, multi-file react app is on the roadmap
yeah, it's actually doing it in html
hmm i see
well in webdev im asking it to generate with html too, since it got so many conflicting env issues
Hello AI
when did that release?
Got rid of direct chat?
i think these are not necessary :
Make sure the index.html file includes:
- HTML structure with proper DOCTYPE
- Tailwind CSS for styling via CDN: , styling inline, or via an inline
it will just confuse the model
idk if those tags are rendering issues or actual model output issues
It doesn't work for me, I just get errors
its on the bottom left now
unfortunately they removed gemini 3
Are the codename models available on the new dev arena?
but they're in webdev arena
are no longer
are you sure
I've been trying for over an hour
oh damn, they must removed it there too
☹️
This is on our radar, we're working on a fix asap
yup i cant even access the code
yea you can see after you vote
you can also select Direct & Side by Side too
i hope its not constrained by a specific styling as well
just let the model chose whatever its trained on
i know they are trying to build a fair battle
but thats more fair imo
i cant see the battle mode or the side by side
also the 10 max tool calling
and it only lets me choose the models
Issues with image uploading ?
are you on canary?
nvm i figured it out, u have to unselect the models for battle mode to activate
yeah
Gotcha, yeah that's also a new UI we're experimenting with at the moment.
yeah im not used to that lol
im kinda curious if the high variability from lithium is related to webdev arena sys prompt
sometimes it just do some changes that you didnt ask for
and it doesnt maintain the same UI at all
To learn
Same here
i had that happen to my generated game lol, it kept changing the character and the background even tho I told it not to change the UI/characters
yea i had to craft a prompt tailored just for lithium to stay consistent
We are experiencing some other issues with the site right now, not just Code Arena.
i really hope its not a model issue
🤔
also didn't they remove lithium and orion from the webdev too?
Hey guys!
it was impressive at UI coding
but so bad at instruct following
and its so prompt sensitive
thats why people kept zero-shotting prompts instead of improving them after 1 iteration
Hope we get Gemini 3 pro tests soon!
Is anyone else getting errors when you try to generate pictures on the web version of LM Arena?
yea a lot of people were having rendering issues with lithium
especially if you ask it to render 3D simulation
IDK, I cant upload images tho
Errors on me
it will use threejs but then the version used is incompatible with the pre-installed nextjs
Oh, their having a core issue ok, i was wondering why i couldnt generate an AI picture for an app i use. thanks for the clarification.
well i actually had a good generation for a 3D simulation of the solar system
i will send the link for it
looks good
but its not that challenging anymore to LLMs tbh
bro Claude 4.1 Opus failed horribly
im always telling it to be creative and create something from scratch
sonnet 4.5 generated a good one for me before
i even tested it on 4.5 sonnet and they all looked horrible, the planets were clipping eachother
surprising
did it generate js & html?
or did you fix some bugs with an IDE
yup the same prompt as lithiumflow, they all generated the full code but only lithium works well and looks good
did you ask it to generate code separately?
nope, i can find the prompt if you want
alright
is it me or imagen just isnt working
ohhh okayy
same like cant upload pics
We're having some problems with the site atm, team is looking into asap.
Random zoom
you know what else is not necessary?
the webdev system prompt
and yes
this is the reason it couldnt make any good games
its literally told its just a frontend engineer
and a ton of bloat
not u again
Image failed to load issue?
Yes, and other issues as well, team is working on it though.
lol
i was looking for this
just to understand whats happening bts
let me take a look\
use the new code arena for that
i dare you to find 4.1 Opus/4.5 sonnet system prompt
link?
u got claude's sys prompt?
well the lmarena code mode seems to override it
i think
how do u extract them? with a jailbreak prompt?
Video
i like project
i didnt even jailbreak
lmarena website is down, cant generate images
ty
@narrow oracle Our site isn't working at the moment. We're working on a fix. We'll have an announcement when it's back. Apologies for the inconvenience.
My body is ready for genie 3
nah its just a better extraction
Bros do I have to use arena or can I use ai of my choice?
this ones correct
arena but if you have a powerful device search up "how to host (any open source model)"
Ohhh ok
LMFAO
Would iPad Pro suffice
for small models yeah
WHY ARE HIS EYES BLUE
hello
Bings image creator is pretty good
isnt it just the gpt image
IMPORTANT: You have a maximum of 10 sequential tool-calling steps to complete the task. Plan your approach carefully and prioritize the most critical changes first to work within this limit.
i think this one is better, it wont put that strict constraints but it will at least create a minimum viable product :
- If the user's request is too complex to be completed in 10 steps, build a simplified but functional version of the core feature.
ya it is
It’s saying do not have permission? I’ve been using this for a while. What happened?
its working for me thanks
NVM
Yeah looks like it could be back.
Lol sorry just got the word
is it gonna fix both of the websites?
Yes, image upload should be working, same with Code Arena
thx
it is
Yeah just confirmed, we're back!
we're so back
they should add a special case for 3d simulation to fix the bundler/environment mismatch or a version conflict between libraries
bro the new code arena is so good
pineapple here is how many people i can fit in my game
how many programs*
why did i say people
lol
they still want to use a component based sandbox with react
they should also add "use client;" directive
for SSR incompatibility
i think thats what was causing non rendering issues in the first place
btw omnificat check #ai-creations
@echo aurora can you ask the developers to add an exception to the system prompt?
this part of the system prompt:
"ALWAYS create complete, functional web applications in a single index.html file.
Make sure the index.html file includes:
- HTML structure with proper DOCTYPE
- Tailwind CSS for styling via CDN:"
instructs to include tailwind, but its not always needed when making websites, so its likely that people are going to run into issues similar to webdev where models would often create ui based versions of whatever was prompted because the system prompt tells it that it's a react engineer and to use tailwind
Will do, thank you for sharing 
hey ananas whats code arena?
and how do i get there?
it will flash unstyled content while loading the page
the UI/UX is much better
its fast
too
yeah
webdev feels heavy
i miss lithiumflow tho
i dont miss that thing
these models feel weak
its gone?
it pissed me off
yup
its so dumb
literally how
i mean it can create some crazy stuff but the way i see is more like an accurate pattern matching from trained data
but then when you ask it something out of norm it just breaks
i dont want to blame webdev sys prompt
because if its really that prompt sensitive then we are doomed
lets wait for it's release, its probably coming out this week
maybe adding a sys prompt to it breaks it
cuz 2.5 pro does the same
google models are like that
exactly
thats what im saying
they are bad at tool calling as well
i really hope im wrong
What i don't get is how they put it up for only 4 days then take it down
Google i mean
have you seen how fast checkpoints go with google?
like
they just are like that
bro google had like 3-4 Gemini 3 checkpoints rn
all in a week
a testing model contending for gemini 3 GA
i think they're preparing for a release rn, polishing the model and choosing (hopefully the best checkpoint available)
i can see it being used to generate an initial UI frame but multi-convo with fixing bugs etc... its disastrous
it did create some cool stuff
i just didnt save the code
it made a beautiful molten core using threejs
Do you guys think gemini 3 will be free like gemini 2.5 pro
yea
but with more strict rate limits
Idc about rate limits
thats why they are still trying different checkpoints
I got 20 google accounts
oh wow
1 free prompt per month
if its on AI studio then yea
when 2.5 pro preview launched, it was completely free, with no limits
Always limits...
its not like openai
Google AI studio doesnt have that
Even now there is
on 2.5 pro?
Yes they do. They are just absurdly high
10/min
Or a bit less than a min
10 prompts a minute?
bruh thats unoticeable
I'm pretty sure
Yeah I know i said they are absurdly high
Google already has all of our data + ad money
they dont need more
thats how they train their AIs
Ublock origin 😍
or do they?
i say this week
the preview comes only on Google AI studio
the full models take like 2 weeks
to release everywhere
whats ur prediction
so you bet november
So what?
Bet you put October 15
100 bands all on October 15 yes
what the f ck
did claude just
disappear???
no it didnt
anyways im going to bed
cya
cya
You are supposed to say forton who?
Pineapple AI shows no mercy
Or midnightfade
Or mpoe
no all of them lol
sometimes they give you verbal warnings
Oh
forton ite battle royale
No, dont think so
I'm not AI 😭
Haha it's just a little joke
I hope you didn't take it to heart
How can I make it up to you
lol
Which is better for internally consistent creative writing purposes, where the AI-gamemaster must have a high intelligence?
2
4
5
Gemini 2.5 Pro
🥰
I want to write a book
Ok you will write book
How can I write a book here?
wdym
@sullen quest
.
hi
I think you're looking for https://lmarena.ai/
hello
so is lithium gone
nooooooooooo
that means it might be going live on aistudio soon tho?? im more bummed about huny 3 being down D:
sigh
how to get lithiumflow
lithiumflow was bad. hope its not gemini 3 pro
best image and llm both down at the same time rofl
it might be flash idk which is which at this point
chat should i touch virtual grass now that ai broke
nano banana at least works
Nothing is working for me.
Been in queue for 240+ seconds so far... after refreshing the page for the fifth time...
love how it added loot and swords and the village, just prompted 'dragon'
start a new chat
@echo aurora If somebody dms me advertising, can they be banned?
what is the diffrence between the three websites like there is the normal website, beta and canary
Tell them the middle finger.
dms can be faked so idoubt it kermit
i assume canary is just testing things and beta is.. beta lol
I'd block them and still report to mods as it's helpful to see if we get other reports
there seems to be zero diffrence between the beta and the canary sites
Beta and Canary versions we can use for early deployments to test with; however, those sites you can expect bugs
neet
What seems to be the issue?
agree, very sick dragon
Okay, well, I just disconnected my wifi on my phone and it seems to work with my cell data. I was gonna ask if my IP was banned or something. 😂
ah, it could be a vpn causing the issue if you're using that?
seems to be up, not seeing other reports either
That is weird because like it's the normal "Something went wrong while generating the response. Please try again." type stuff.
oh huny 3 is working again, maybe reset your modem idek
Yeah, it's working now.
You shouldn't be running into this in battle mode btw
Yeah that's the thing of how weird it gets, could be any number of things or could be something on my end... then again I tested it on a number of other places.
man id pay a dollar to get priority llm/image generation
What model would you recommend with under 2 second replies and is still very accurate
gemini flash-lite latest, maybe not under 2 seconds but is what id use if i wanted speed and awesome model
Guys which one is good for web dev
Claude 4.1 16k thinking
Claude 4.5 32k thinking
Gpt-5 High
Or a diff one
Refresh
It happens sometimes when you leave the page opens and do nothing
Ah. Thanks.
Guys, NimbleBean video model on the arena is Kling 2.5 Turbo Standard
is it good?
it's number 1
better than sora?
how do i use it?
how long are vids and how is it with copyrighted stuff?
API
ughh
0.21/5s and 0.42/10s
u like it better than sora?
It's only i2v model
Sora has t2v and i2v, even though I dont like their i2v bcs too restricted with humans
welp ai is getting to a point were we arent going to be able to tell what is or isnt real
I am surprised Kling hasn't revealed this model yet
cant wait until we have AI TV and Music
def
that why labelling it is key, like not visibly
and live shows and stuff are gonna sky rocket in price
ok but the thing is when people decide not to do that
we will have real time AI TV/Music
that is why sora has a watermark on each video
but their is AI to remove the sora watermark
its not just visible
but internal as well
its still going to be intresting though
we already have anime that is fully AI and is better then a lot of stuff out there
and some youtube videos are freaking wild with AI stuff
its just going to be how we use it and what creations we can make before it starts creating itself
live action AI Videogames that feel like one long cutseen when
I haven't seen any other AI model which produces a perfect run like that
i bet you could get away with showing people outside this server to see if it they can tell if it is AI or Not
most might say its real
it looks so real
exactly my thoughts
@ashen mauve looks at this one
if it wasnt posted here id say its real
If I send it to my cousins whom they are not in the AI community, they'd say it is real without a question
ehh with the medical one i dont know about that one
some medical nerd would probably say its real because xyzabc they hold it this way
that just looks like it is from some movie at this point
I am a doctor so I can tell
lol
my thoughts are they were not holding the forceps or whatever correctly but the way that whatever was being handed off was smooth beyond that so-so
he was holding one each hand then he took another one from his assistant, and they way it happened feels so real
this one looks like from a Netflix TV show
can someone say which channel this photo is from ?
I wonder what company its from
Since google and openai just had their major video model releases
Kling
These are Gemini 3.0 Pro models
NimbleBean is the codename for Kling 2.5 Turbo Standard model
np 🙂
What
Inspect element
Are u saying that ill fake it to get someone banned
can you say which channel this photo is from ?
first time in discord just seen for the moment
just saying its possible, someone can make 10 accounts, report fake screenshots, boom someones banned
what dose any of this mean
Higgsfield ai popcorn model aad please
Hello
hello everyone! first time here and ready start my journey as youtuber lol
Is that web os from ai or what
Hi everyone
the 🪽 🐙 has been released... (MiniMax-M2)
You are stupid stupid stupid
Is it revolutionary? 🫣 like Claude 3.5
Use AiStudio
hello i love prompting and generation, like playing with a shiny new toy
@hallow forge Please head to #1397655624103493813 for a detailed guide on how to use the bot
A bit of general advice.....
When extending a video clip many like myself extract the last frame - but that frame might be quite more blurry and have a lot of compression artifacts, so what to do?
Unsharp mask would make a few things sharper, but also make the jpg noise even worse. And sending it into one image making AI will change details.
So the best might be to use any of the many-many image enhancers that now have popped up for improving digital camera pictures.
Among those I tested I found imggen.ai to be quite better than most others, and it do indeed remove compression artifacts.
@faint charm Please head to #1397655624103493813 for a detailed guide on how to use the bot
?
hello guys
@wild jasper Please, read our guide in https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to properly prompt the bot.
i am missing a lot of context
oh nice! cant wait! thanks for sharing 😊
The angelic octopus?
i am here to create videos
@floral goblet Welcome. To learn how to create videos have a look at this guide in https://discord.com/channels/1340554757349179412/1397655624103493813
Hello
@cinder trout Please, read our guide in https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to properly prompt the bot.
How to generate video from mobile what command I should use
Hi! Please check out this guide in https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to properly prompt the bot.
@obsidian barn Submit your prompts in #video-arena-1
After pushing the prompt will it remind me once generated?
Yes. The bot should send a DM once the generation is completed.
Thanks man
No problem
claude-sonnet-4-5-20250929-thinking-32k
Something went wrong with this response, please try again.
Guys how to fix this?
Refresh page
I did that like 999 times also restart my PC
Whoever created LM......I love you bro lol
gemini 3 in 2 days:
source:
my left buttcheek
whats the best photo generating ai rn?
yey
i was right
ig
it took then 22 days to release it
i hope thats not the case for gemini
This is on the teams radar, I'll be sure to bump it
this wild
Not here 🙂
@half sonnet Please check https://discordapp.com/channels/1340554757349179412/1397655624103493813 to learn how to generate content 🙂
maybe we should act like it
geez
it lost all of that a few hours later btw
how did you take its system prompt
i think its not gonna release less than in a month
from now
Bro
gemini 3 only in december guys
Someone sent a polymarket photo
Where people are earning money just by saying no that gemini 3 won't come in october
how to do ascii art
finally I bought 2 monitors, but I'll be crazy with 2 monitors Idk
random gens
I sawing in smartphone look sooo realistc, but in an 24' monitor like playdoh
bro
what happened
which one?
the frog one
@gritty acorn Hey please check to learn how t generate 🙂 ☝️
wym?
the results so far are consistent with a random walk
is this sora
Yes. Like Midnight said, it'll notify you once finished.
ahh i was just thought it was wild that the degen llm made so much and then lost it lol
full lifecycle of trench trading
yeah
Honestly, I didn't even realize that model was gone.
I just kept trying to get it on WebDev.
I should do this sometime
Gemini 3 on November 12 according to mine
Right one though
holy tuff ts is actually fire???
Looks cool but there's some discontinuity
Seems like most of it is in the coding category (not agentic coding category though).
yeah gonna tweak it more, but im tryna get the best version of superspeed
Interesting correlation matrix, where are the music benchmarks from? @keen beacon
cause even the ones we got that are not ai kinda suck
Well its not really your fault I see it in all other videos too
But if you can prompt that issue out? Would be amazing
i think you can get lucky
Very cool, do you have a ranking breakdown?
Smaller graph. Red is mostly <0.2, dark green is mostly >0.4. Green underlined is >0.5.
I didn't expect data analysis to be so low, maybe it's fairly easy for LLMs (converting between JSON/CSV).
The agentic coding and mathematics category seems to have moderately positive correlation
I think some categories are bad (like coding), but individual categories still seem useful (like agentic coding, with a mean of correlation of >0.5 with your benchmark).
hello
Interesting how Flash does so much better than Pro here.
Is the top one Flash Lite?
Is the one that did better the reasoning or non-reasoning one?
Interesting that Pro does worse than Flash at theory too
Maybe time to test it on some chord analysis xD
I meant for my pieces. Last I tried, they were all pretty bad at identifying chord structure.
What format did you use for notes? They seemed to have a lot of trouble understanding it.
Maybe because I was feeding in individual notes, I might try again and see.
what first?
2
6
That's surprisingly close
Claude got it right:
Not sure about now, but they were pretty bad at reading music notation (vision). More or less random outputs.
Which category did they seem to struggle in?
It's a bit limiting, since it doesn't show how the notes line up or which note it actually is.
Was looking into MusicXML (too verbose though)
I WZNNA GENERATE A NEW VIDEO BUT I CANT FIGURE THIS OUT BC IM A TOTLAL NOOB ANY ONE HELP
Sonnet 4.5 got this one wrong:
Identify the type of mode or scale as specifically as possible (use the mode name only, i.e. Ionian NOT C Ionian): A-Bb-C-D-Eb-F-G-A
Hi there! Check this to learn 😉 https://discordapp.com/channels/1340554757349179412/1397655624103493813
ChatGPT stopped itself midway then got the right answer:
Yeah
It got the chords better than ChatGPT in first test ig
They're kind of inconsistent
If they could read sheet music and analyze it properly, that would be really helpful
Seems odd that they can't read it, since sheet music is so structured, and there's so much data available (can just generate the sheets from midi, then feed the it back for training).
@fair junco Hi there. Just a reminder that englisj isthe only language allowed in the chat channels 🙂
Is a larger graph available?
What is this for
Hi
hii
hello @crude goblet @proud hazel 
What happened to lithiumflow?
Someone poured water on it 💥
(it got removed along with Orionmist)
Lithiumflow built amazing things and I wouldn't find it again.
thanks for the information
@echo aurora Is the sign-up process somehow buggy if you sign up with an email address? I didn't receive a verification link. I've already tried resending it.
@undone violet Please head to #1397655624103493813 for a detailed guide on how to use the bot
Thanks for the heads up, going to followup in a private thread as I may need the email address associated.
update to the leaderboard next week?
gemini 2.5 pro votes are rigged
the model sucks so bad at coding, creative writing and everything else
Huh so ether Grok is dumb or all the Grok Models in the website are all only Grok-1
I literally tried all of them and asked them the same question of what their cutoff date is and all of them keep telling me they are Grok-1
@long lantern Please, read our guide in https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to properly prompt the bot.
Hello. I like AI videos
welcome welcome!
what was the prompt
@vast fern It's here ^
Thanks
life is sad now no more pie 🙁
Cause its gone
@dry sand Please head to #1397655624103493813 for a detailed guide on how to use the bot
Hey how do you get the arena champion role
This announcement here has more information about the program including where to apply - #announcements message
hi
hello 
Alright
I just sent the form
I was already a arena champion before until I got wrongfully banned but i don't know if that counts
which is best in C++ coding and debugging?
5
11
how did the Coders of lmarena even get those models lithiumflow and orionmist those are Gemini 3.0 pro if i heard it right
They got in contact with Google and got the models early, just like how they managed to get nano-banana early.
ah alright thank you! But they are Gemini 3.0 pro both models right?
No, Lithiumflow was just the Flash version, while the ones on Google AI Studio are the Pro versions.
evidence?
No can't be because both Think so long
lmarena is designed specifically for ai's to be tested, so google put those model's up for testing reasons
This post: #general message
that doesn't prove its flash
that just means its weaker than other checkpoints
Gemini 3.0 Will be Crazy good
It does, because notice how it doesn't say "Pro" next to it.
Meaning that it's no more than just a regular version of the well-known coding master that somehow overtook Claude.
Then why is there another version of it underneath?
But that wouldn't make any sense because Gemini 3.0 in itself is one big branch.
So, why would there be other branches of the same model?
...................................................
darkness, u know how llm's work right?
Yes, I'm well-aware.
you remember llama 4?
Yeah?
did you know there was something like 40 different versions of llama 4 on lmarena under codenames before the release right?
checkpoints aren't the end model, they are test models
figure out how good it is, modify it or start over
But the thing is, that isn't a checkpoint. That's a version of the model.
thats.. the same thing
Just like how the Pro version isn't a checkpoint, it's a variant of the already existing model.
I know that.
i'm going insane
I meant "already existing" as in the current one.
wat
The unreleased one.
wdyfm
The one that overtook Claude.
"less vague, please"
I'm not being vague.
.
You just have to learn to use your eyes and pay attention to the image.
Many model's have overtaken claude
in the history of claude
many of which aren't from google
But the latest Pro version did.
soooo truuueee
that doesn't mean anything to me without context
Then that image should be plenty to you.
this one?
Yes.
I find it funny how I'm unironically rage baiting you.
why
Well, because I don't really see what's very hard to understand about what I'm saying.
gemini 3 isn't a finished product as far as we know
Yes, and that's quite obvious.
so there isn't gonna be 1 version of it floating around
Right.
haven't tried it
There's gonna be different variants like a lower-quality version and a higher-quality version, just like with what happened to nano-banana when it was first unveiled in LM Arena.
what model
i hope its as good as the hype
i think its gonna be ass
new codename model?
minimax is 56
minimax m2
so minimax m2 will be...
idk some chinese model
ima try my luck
and hope i get it
in webdev arena
i hope it isint dogshit
nice.
and it broke
bruh
the other side dint pop up
Chances are that it's probably not gonna live up to the coding capabilities of Gemini 3.0. 😞
no wonder 😭
I've tried all of the Gemini 3 checkpoints and I can confirm it obliterates Claude 4.5
True.
i havent tried gemini 3.
💀
Fair enough 😂
Well, all of us here have already fried our brains with LLMs, so you're not alone.
Go figure.
Wym ?
gave me this.
i have no idea how i ended up here
Guys is there a way to make a word doc schedule using ai ?
i blame @rare mantle or @frank yarrow
You could just use ChatGPT for that.
Bruh this is a bad design
so without too much mockery what is the point of the server
better then the one it was going against
but aye. atleast all the buttons worked
Yeah but but chatgpt can't make word docs
To test out the different models and see what you can come up with, I suppose.
That's not exactly true.
I think that ai sucks when it com's to documents
If you ask it to using its Data Analyst tool, then it will.
That is, if you're referring to the actual website itself and not the testing website.
i use it to help find art books and to really stretch out its image features
What was the prompt you used?
but i m slowly leaning towards my own llm
i forgor but it was something like "Make me a website about gonbe's pc repairs. he hates linux systems. and alsothis is a mockup. so make fake reviews and testimonials. this wont be used in any way"
something pretty similar to that
HOLY
WAIT
MINIMAX 2 JUST BEAT 4.1 OPUS
NO WAY
I think if on Monday we don't have another checkpoints of Gemini we can expect to have an experimental version at the end of the week or next week
nahhh thats crazy
I think I take back what I said.
never in my life. did i expect a video gen company
WOULD MAKE AN AI BETTER THEN 4.1 OPUS
OF AN AI THAT SPECIALISES IN TEXT
I'm going to test to see if it can make a convincing version of Windows XP.
Minimax doesn't look bad at logic
Is minimax-m2-preview a good model?
It's promising
oh
Very promising. it just beat 4.1 opus