#general
1 messages ยท Page 367 of 1
Very neat! Are you going to make it available? I imagine a lot of servers would like something like this, including us.
Although I would need to see it vetted further before adding.
rn i made it testable in small server, but im absolutely unable to afford hosting
and so far i only put it as bonus function in the other bot i had for other server, but its easy to make a new bot dedicated for it
why dont you sell it to someone
someone who can afford hosting
idk i just made it to help people
id happily make it free but i cannot afford to host it
doesnt use ai model
it uses small ocr
Okay gotcha, I'll think about this a bit more, maybe we can host. I'm in the middle of something atm so won't have a change to sink my teeth in until later. I'll followup in DMs
It was Safari on IOS. But it went away after some minutes and worked again.
I just refreshed some times, and it worked
Friends, can anyone help me? I need a promtp that can turn 2D renders of Lego figures into something that looks like a real photo. I've been trying for two days now and I can't come up with anything acceptable.
I only managed to do it well once, completely by accident, but I was unable to repeat it.
why in agent mode answer loading is so long
@echo aurora is there any reasoning as to the order of the models?
i think it's based on arena leaderboard scores?
no way gemini 3 flash is above gemini 3.1 pro
this is what it is when you scroll down
Yeah @whole sundial is correct, it's based off of the leaderboard. But it's worth noting sometime there is a bit of a delay. A newly added model takes a bit of time to appear.
really?
actually the leaderboard looks different than that, they must be sorted manually
leaderboard is very different
Glad to hear it. Keep me updated if things start to look off again.
Yeah let me look into. Bit unclear why it's showing that way.
gemini 3.1 pro is 6th, score of 1489. gemini 3 flash is 15th, score of 1473
thanks ๐
if the sorting was accurate it would be g3.1pro - grok4.20 - gpt5.2chat - g3flash - glm5.1 - sonnet4.5 - qwen3.5-397b-a17b
it just seems like there is no rhyme or reason to the ordering
i don't think there was ever a leaderboard that had gemini 3.1 pro below qwen 3.5
Arena lm rank leaderboards for open source is dumb
Seriously? GLM 5.1 better than mimo v2.5 pro?
I should retire as a vibe coder, it's that I don't trust ai anymore
And these leaderboards and ranking systems are dumb, who is testing these?
imo kimi k2.6 >> all other open source/weights llms
with deepseek close behind it
you think kimi > deepseek >> glm and mimo?
I am so pissed off about opus 4.7 is that it's worse and can't even close to mimo v2.5 pro
bro what
there is one question i have that kimi and glm gets right but mimo and deepseek doesn't
Does anyone know why in agent mode the responses only load after restarting the site?
Mimo v 2.5 pro made a software from scratch at one shot where the development of that software stopped 4 years ago which was incompatible, and there were no source code
Opus 4.7, 4.6 and sonnet 4.6 failed. It didn't reject me, but It failed
yeah i don't code with llms, i have an even more niche world knowledge question that only gemini 3.1 pro can get right (3.0 flash can get close if you ask it enough times), opus 4.7, gpt-5.5, and everything else fails
Also I've tried GLM 5.1 agent, ran the agent for 12 hours, it was doing loop while the UI for that software looks good, but the main core was corrupted. Failed
Gpt 5.5 is not good enough with it's agent
Gemini 3.1 pro is good at hypothesis and logic
But not good for coding
How often are you seeing this happen? We're pushing out a lot of bug fixes with Agent Mode today. May want to wait a day and let me know if it's sitll happening.
But these ai models and their company is a liar. Specifically I don't believe in Anithropic anymore
what did anthropic do this time
But for some reason Chinese LLM is growing, act as AGI for some reason
Every time I send a message, the response takes forever to load, and I have to restart the page. Sometimes, when I restart the page, there's no response at all, as if I never sent the message
Anithropic is a disaster, they are just lying with mythos
the response takes forever to load
Roughly how long are you waiting?
there model got massively dumb
mythos
yeah probably to save cost to train their new model
I waited more than 5 minutes for sure
If I swear to
Any ai models, it's always Claude, I am so tired bro
terrible compared to gpt-5.5-xhigh
again, I asked him a question, and the loading has been going on for more than 6 minutes
Gotcha. Yeah let's give it another day. If you're still seeing this tomorrow can you give me some recent Eval IDs for these chats? I don't need all, just a few would be nice.
Gpt 5.5 standard version is 10 times better than opus 4.7,
tru tho
i havent used 5.5 yet
u don't even get to use opus 4.7 for free like
is it good
they're just so greedy and don't give u any limits
Meanwhile mimo v2.5 pro beats opus 4.7, I was laughing so hard after making the software from scratch
@echo aurora Sorry for talking so much, but some really weird things are happening to me. I managed to log into my account, but now when I enter the prompt plus a reference image, I get a captcha. I check it and This error occurs
probs a phishing site
just hard refresh or use a vpn
Eval IDs? What are those? How do I send them?
hello
Imagine that an open source model is way better than a close source model, who is gonna pay for this crap? Maybe blind one? Just for entertainment?
xd
XD
wth ?
Nah man, don't show Claude, I broke up with Claude, said the last goodbye once again
@surreal zephyr look at these people
very
good
anthropic is just trying to be so kawaii
what is that ?
Ask your Claude to make a daemon tool
ur just scammed
Gpt 5.5 is God
Gpt 5.5 can do one shot, Claude requires millions
It's the random set of numbers/letters found in the URL when on a specific chat session.
You shouldn't think, but do experiment, that's real testing
Yeah this is likely still related to the outage we had earlier that's causing the captcha system to act up a bit.
anything new
Looked into this a bit more, the ordering was changed from leaderboard order. We wanted to encourage more diverse use of the models in Direct/Side by Side so we're going to be changing up the order from time to time. cc @pseudo hemlock
btw, pineapple, can you tell me what the main idea of โโagent mode is and what its features are? Thanks
wha
Which of these ten AIs is the best?
5
15
3
GLM
ig
isnt it random or based on recent models or most used models or sum?
is it hand picked? also thanks for looking into that ๐
read pineapple's message above
well is not thinking]
its probably some combination of price, speed, and usage
and some other smart people things i dont know about
well pineapple said its less common models there
"diverse use"
so like models they want to be used more i guess
its " Instead of separate hidden limits across different features and models" that what they said in the announcements
anyone know a better place to host a bot than huggingface ? ๐ญ
huggingface has unstable connection apparently
cloud host i mean
and ideally cheap - it uses little to no compute the way its made
or even free
im sure gemini 3 flash isnt uncommon
but its amazing for the price
my computer
grok 4.3 is thinking but isnt named grok 4.3 thinking
bro its not about size
maybe they cant run it 24/7
i mean yeah but if its with nothing that mean no thinking
no. grok-4.3 has thinking enabled, but it isn't called grok-4.3-thinking
i just used it
huggingface free tier would be enough if discord wasnt disconnecting
30 seconds ago
that what i mena
"Thought for 4 seconds"
so its thinking
ik
but i mean it say if it thinked or not
did any one tryed google gems
?
i need u to be mad at this claude fanboy @deep adder
unleash ur wrath on him
claude goat
at coding yeah
@echo aurora any updates on opus models?
Yeah it's done manually. And no problem! I'm very glad you did as it was new information to me too!
No sorry to say I don't have an update on this front.
anything new ?
idrk have any idea any more i am studying all the time
any plannings after the daily limits part
forget him
elite claude user
Na nothing new to share sorry to say.
It's also TBD
it was useless
but codex is better ๐
I love codex mobile
mobile for codex
also one small question
yk codex came out for mobile
something ive wished for
my entire life...
no
Melon disapproved!!
for the agent mode , i got the access any plans on combining the files e.g. html css and js are seperate for now and they are no linking as they do in the normal coding view also what is the usecase of agent mode why do we really need it
they're a billion dollar company, they can afford a cheap vps
We are planning to add upload ability, I couldn't give an ETA when this is the case, but the request is very much on our radar. I would encourage you to check out this Help Center article which goes into more details about it: https://help.arena.ai/articles/1811908126-arena-experiments-agent-mode
just a few dollars a month can get a vps that should be more than enough to handle the amount of images that get shared here, especially since video arena is gone
thx that was usesfull
i meant the person
like the person in this discord, maybe cant run it locally 24/7
gpt 2 image
true
google I/O will ready soon guys
what is google i/o
google's annual developer conference
Should be exciting!
gemini 3.2 flash is coming (they might've just renamed it to 3.5 flash before launch), pro should be coming soon as well but it may be after i/o
Google antigravity already abandoned lol
Sadly you're right
Yeah pretty sure it was renamed to 3.5.
Which is better, Deepseek v4 or Kimi 2.5?
what did i DO to twenty-one-pilots bro
V4
hello from Argentine
sorry to interrupt but do any of u know if new daily limits are larger than the previous hourly? Idk limits?
lol what is thta
hmm why Deepseek (V4 lineups) models tho? How powerful is the Deepseek V4 lineup?
@echo aurora Arena i request please if u want to add credit system its ok but once the limit reaches we should be able to continue conversations in same section after 24 hours please otherwise please don't add this system
it applied to coding section too
The best Chinese AI lineups right now?
15
34
1
Deepseek (V4 lineups)
Hey @zealous berry for the new credit system when you run out of "credit", when it's replenished (the next day), you'll be able to continue using the same chat sessions, assuming it hasn't reached a context limit.
Hello & welcome
that's what i am saying remove context limit and add credit system what's the point if context limit is reached this feature then will be just annoying we can't able to work on already built project though
hello I am BoutQci
People must have only used GLM and Deepseek then ๐ kimi is easily the best Chinese AI currently
hey
You can find Video Arena on our site here: https://arena.ai/video
And then put the image in the prompt
Pineapple
Can you @everyone and say โI lost the gameโ
Do yk about the game
Hi @echo aurora , if you remember me, we were talking yesterday about how agent mode wasn't working properly. Responses would only load after the site was refreshed. Didn't you ask me to show you the chat ID yesterday?
Hi help me
What you need help with
They're all good, I've tried them all.
Pineapple will come back in 6 hours
Sad
๐
Got Agent mode ^^
in agent mode, we will never know which model is actually working for us?
maN I miss the ai bot
@light sleet you betrayed me!!! how could you masscare pineappale
I didnt ๐
so where is it now? huh? answer this if you can
liar!!1111
It?
Hi
guys how do i fix this ? its been like this for a day , i have refreshed it multiple times yet still . Its not image generation but simple text generation , what do i do ?
Image generation is not working on any model
I will make games is finished
yay I got agent mode again
change your name first racist
proud
<@&1349916362595635286> racism
this server
I will not redeem the code for u dw
thats my gang
Dont!!!!!
dont expose me
dont do this gng dont send my pic
no
i have but i dont
its useless
sure its fast, but its as stupid as 2b models somehow
which makes it completely useless
I redeemed the code
sorry
I got free Google Play subscription
thanks
what does it say
free 1 year subscription
jackpot
i wan't to ask something is this ARENAcan make a docx file?
just ask codex
๐ฅ
anyways
anyone has image examples that COULD be false flagged as "mrbeast giveaway scam" while they arent?
codex mobile mogs claude code
i wonder if
pineapple will actually dm me
or he forgot
๐ค
he forgot
prolly
actually funny - ai taking jobs
the automod replaces 3/4 of mod team necessary here lol
MAYBE i can autoflag those too
In terms of features? or what?, for my opinion Qwen and Kimi are kindly similar
the fixed =help?
=help
ModMail is a feature-rich Discord bot designed to enable your server members to contact staff easily.
Please direct message me if you wish to contact staff. You can also invite me to your server with the link below, or join our support server if you need further help.
To setup the bot, run =setup.
lol
๐
good i guess
how it feels waiting for scammers to put the examples into reference database ๐
Hii
hi
hello
hi
wondering why there isn't any AI generation contest anymore ๐ the form still open for anyone to vote lol
and some code arena website made by them are dead ๐
@echo aurora sorry for pinging, but what exactly happened to generation ai contest?
U are using nano banana pro 2k?
What is it with arena suddenly needed captcha for every single message
It's pretty much unusable
ur not alone
from my experience it gets like that when u use it for a long time without any break
How can i fix discord Ai generatir chat
gpt the best storytelling llm
i've noticed that if u close the site for a few hours and come back it works fine
which one
Should I use GLM-5.1 or GLM-5-Turbo for coding?
any one here make ecom ?
hmm i like 5.1 , it has a MUCH MUCH less censorship
can generate fraky texts too
until 5.2 ruined everything
because of thieves, who thieves come and put the servers on exhaustion, they also put captcha
really? its your fav llm for stories ? i been testing a few, grok is interesting as well
not for storytelling btut for generate flirty lines for me
what stories u make
like do u turn some manga or book into a story
Thanks, I'll keep that in mind. But this is the first time in the day I've tried using it
Use gemini nano-banana-pro 2k is better , trust me
or generate a new from scratch from gpt
ohhh , did u use it yesterday ? i have noticed that when i use it on my laptop and instead of closing the tab or the browser completely i just close my mac's lid , it happens . So when ur not using it or done using , close the arena tabs
gemini nano banana for flirty lines ? even sexting ? im not well aware but isnt that a image gen model
Yeah, I understand but I haven't used today and it's the first time it's been like this actually
Ohhh, interesting. I've paid attention to this, I mostly use it on mobile though but I will make sure to try always closing all tabs from now on and see if it fixes it. Thank you!
welcome , u can close the whole browser and that works too
Nano banana pro 2k is the best at quality ! In my opinion .
why would you want to generate NSFW content? for Only Fans or what?
like for projects, i use stories to help me build
Creat video
no
whats the prompt
damn allah's dad
naman aso ๐
a reason I miss opus is because I can't copypaste "(NO OVERUSING CHENS (THEYRE ALLOWED BUT DONT) NO MARTINEZES NO OKONKWOS NO MILLBROOK NO OVERUSING THE 15TH DAY NO OVERUSING 47 NO OVERUSING BULLET POINTS NO THOMPSONS BE VERY DETAILED AND RELAISTIC!)"
anymore
so it do really have a tendency to use name like "Chen" ๐ญ
GPT-5.5 or CM5?
4
6
2
Claude Mythos 5
Why this happening??
ye
oh
kinda great
oh
yup
am facing error
this one
hmm
repornt in bugs section
saw that
Why is this happening?
bcz i was literally so much close to perfection
idk its happening with us too
maybe the prob is same
agent did u bad
prob this
btw, i tried Z.ai on their website instead of glm models on arena
and its kinda way too good for its pricing
but it didnt gave limit error
could be something else, i hit it 2-3 weeks ago
the one u got hit with
so did it worked after wards?
bcz i dont wanna lose my workflow
nah, i had to use another chat and browser
meaning i lost my work flow?
prob
which browser do u use?
chrome
hmm
good
so this means i have to work with other ai's Bcz i will probably ge same error
every device that has same account on arena showed error
dont know, try new chat first then go back and hit retry
same
error
on other chat
what were u makig
making my prompt more better
then log out then log in, clear site data too
oh let me check that
i suggest go with deepseek v4
meh gemini is not that good
on their own website
mine is, i have promted its system
good
the one that we do from its prefernce section
they are probably good for specific diff niches
so it works good
yeh
and not in mine i think
prob
i need creative writing and logic
i suggest, use deepseek from their official website using expert mode
i used but i dont know qwen performed better
in lm arena
hmm, any prob if u share ur raw prompt, let me have my gemini do it for u, if u like it, use it
i even checked the claude sonet
If anyone knows the solution to this problem, please help me.
can it ?
yup
and i run qwen on my laptop locally
try clearing cache and log out and log in , @thorny current told me , i also wanna check
good but 3.6 and 3.5?
3.6, the one in the attachment
is it latest one?
yup
14-20 days old
i give the prompt of
released on 21 april 2026
any u want
oh
the old one is better
kinda true
let me give old one tell him that create prompt for taekwando fight between a girl and a boy outside a dojo
DM?
if u want
No no no no no ๐ญ๐ญ๐ญ๐ญ๐ญ
just a prob, not sure
Means ???
Prob means ?
Look if it's real that lmarena going to add daily credits system its too too too much bad , like after limit we can try after 20 mins or 40 mins again but....after then we have to wait for 24 hour
Hi, i was working on arena video ai but itโs it give only 2 videos 6 sec per day any solution?
Not any solution and if you're work is image to video use meta ai it's give unlimited video genaration
Where is this ? Any website?
probably
We need to stats page I want to see how many tokens Iโve used @echo aurora
any limits?
means?
like any no. of generations per day
yes like u ca generate 10 apps
thats good
u want to see that web replit
if u wanna show
if u make an account from my link u will get 10$ as credit
and me also
make an accunt
yeah
yo
@echo aurora Please delete this scam
oh ok sry i put by mistake
Coding really, Kimi is all around better in that area
It's no problem, just note we do have a rule against advertising here.
Agreed! With the new usage system there will be more visibility into that use.
Who pinged me wth lol
any thoughts on the bot idea so far? i think adding prompts detection is possible too (like those informing about video arena being disabled) ๐ค
This looks like a known bug, I am bumping the issue with the team. Sorry to say there isn't a workaround for the user on this one.
Sorry to say I haven't yet had time to put more thought into it. Will reach out soon.
Alr. if you want to try it out you can dm me :)
Agent mode is ahh
I didn't know there is a new thing @echo aurora
will max model disappear or no @echo aurora
It hasn't yet rolled out to everyone
agent mode is cool but i still dont understand how it is different to normal chat when u select build apps other than it chooses random models
How so?
@echo aurora i want access to the agent mode
No model "provider" choosing
I can't say this is being planned
Every update on the thinking thing where it say asking Bradley brings up mobile keyboard
Feedback brings up mobile keyboard
It's rolling out to more and more users, so would encourage you to keep checking
Takes too long to respond also
Its on canaryarena.ai
But if it not there
You're unlucky
there is no direct link?
There is (https://arena.ai/agent) but if you don't haev the feature it won't work.
YES I SEE IT THANK YOUUU BRO
Too buggy
Something went wrong errors, or some other issue?
Try it out and let us know what you think!
Every ui changes brings up keyboard and is laggy
We're super interested in hearing feadback on this one.
can you answer honestly please
i need answer
so after agent its the full stack code arena being rolled out right
and also im arena champion approved or no ๐
mid
Every ui changes
Sorry what do you mean by this?
I did answer honestly.
but what do you mean this isn't being planned
โ
I thought I did approve you 
When it show feedback ui
When it updates the thinking
whats the point of agent mode
When it responds
yea i dont get it eithher
It's multi-modal. Would encourage you to check out the Help Center as it ha a lot more information on it: https://help.arena.ai/articles/1811908126-arena-experiments-agent-mode
omgggggg
this so tuff man
Meaning I'm not aware of plans to remove Max anytime soon
@echo aurora wow , best agent mode (sorry for russian GUI)
@echo aurora I've already tried Agent Mode and I'm up for a call.
so it will be removed?
max model is important btw
Keep trying it out! And let me know if you want to hop on a call to chat about, or if you prefer let us know with this feedback form: https://docs.google.com/forms/d/e/1FAIpQLSeOI51LKcpvj0wKfekhZsHSIJmMOVPX_Bq6nqb0nS14Kg6m9w/viewform?usp=dialog
it should be stay in that way
okak
Wonderful! I'll reach out privately to schedule a time that works well for you.
I'm not saying that at all lol
ah thank god
thanks
im used to max now
i hope it stays
Is there an example prompt you're able to share that'd help me replicate this by chance?
Yeah it's pretty cool
BRO
now my name color does look like banana's color xd
when will models like opus be back?
(yellow)
IT BRINGS UP KEYBOARD WHEN THE FEEDBACK UI SHOWS UP
Hey @echo aurora Just wanna ask what timezone are you in because I donโt wanna randomly ping you at 3am ๐ญ
And when think updates
Never
:(
Yep..
I don't have an ETA on this sorry to say
I'm sure GMT-7
I don't mind! Ping whenevr
Pacific Daylight Time right?
i think its good in coding tasks
yes san francisco
and wow banana has arena champion
@echo aurora How did this happen?
They applied and were approved
Ty
Keep trying it out and let me know if you'd like to hop on a call
ampros bot is so good
e
Yeah I'm excited to try this out
Btw what type of device are you using? Sorry to say, on my end I'm not replicating this.
Its amazing I've tested it @surreal zephyr approved
Samsung galaxy tab s10+ 256 gb storage space 12 gb plus ram 8x zoom 120fps
heads up - currently its unreliable because discord api connection to huggingface is awful (so it detects, scans, confirms, and then spends 30 seconds hammering the api until it lets it send the response ๐ญ )
if anyone knows a good place to host it for free (im poor) id love to know
bros poor and has chatgpt pro ๐ฅ
And this happens every time the feedback module pops up for you? Does the keyboard block it? Sorry for all the questions, trying to get a better understanding of this.
When it pops up my on screen keyboard automatically comes up
Paying 100 per month for ai subscription i use for school and studying and work vs paying 50$ per month to host a bot noone uses ๐ญ
sad ๐
excuse me
I use it
๐ก
for uhh
10 seconds per day
๐ค
Cery good
fair
No ETA yet
I mean like shell exectuion envs and stuff
It is on experiments tho
It is
only feather got it xd
Well deserved tbh
True
completed my feedback ๐ธ
I've got it and to be honest it's not that overwhelming
it is probably US only for now
not region-based, it's just random
yeah my main account has it but my alt doesn't (i'm from US)
This is correct.
Hmm๐ซช
Nice!!! Try it out! Let me know if you want to hop on a call to chat about, or if you'd prefer to share feedback this form would be ideal: https://docs.google.com/forms/d/e/1FAIpQLSeOI51LKcpvj0wKfekhZsHSIJmMOVPX_Bq6nqb0nS14Kg6m9w/viewform?usp=dialog
tbh i don't even know what to do with it

Tried it and I don't see any difference to direct chat or duel
Hi everyone, can someone tell me if it's no longer possible to generate videos? For example, in image->video or T>V
Not even the boot licking level decreased
you can still do it on the website, they never removed it on there i don't think
but it's not on the discord anymore
search vincegang in tenor gifs
my gif comes up
It's for more complex and multi-modal prompting. cc @slim tartan. Here are a few I've personally used.
For fun:
Create me a one pager one shot dnd session mini campaign. Include some old-school artwork. Provide a quick TLDR for new players on how this will work, have some pre-built classes for them.
For more serious stuff:
Role: Act as an expert rural real estate consultant, off-grid living specialist, and construction project manager with deep, current knowledge of zoning laws, building codes, and land development in Northern California and Oregon.Context: I am creating a comprehensive plan to purchase 1 to 5 acres of land in Northern California or Oregon. My goal is to live off-grid, or as close to off-grid as legally and practically possible. For the dwelling, I want to either build a very small, simple stick-built home OR purchase and place a prebuilt home (modular, prefab, or tiny home).
Task: Provide a detailed, realistic, and highly actionable guide addressing the following areas. Do not sugarcoat the difficulties; I need to know the harsh realities and legal hurdles.
Sourcing: Where exactly should I look to find suitable off-grid friendly land? Where should I look for reputable prebuilt/modular home companies that deliver to these regions?
Core Considerations: What are the most critical due-diligence items I must check before buying raw land in these areas? (e.g., water rights, perc tests, solar exposure, access).
Prebuilt Home Limitations: This is crucial. If I buy land and want to place a prebuilt home, what are the specific legal and zoning limitations? Clearly explain the difference in how counties treat Modular Homes, Manufactured Homes, and Tiny Homes on Wheels (THOWs), and what I can and cannot legally do.
The "Gotchas": What are the hidden considerations or common mistakes first-time rural land buyers make in NorCal and Oregon that I should be aware of?
Cost Expectations: Provide a realistic, itemized breakdown of expected costs. Include land, site preparation, the dwelling, off-grid utility setup (solar, well, septic/composting), permits, and contingency funds. Give rough ranges.
Oregon vs. Northern California: Provide a direct comparison of the Pros and Cons of each state specifically for this lifestyle. Compare them on: building code strictness, off-grid legality, property taxes, fire insurance realities, and climate/water availability.
Formatting Rules:
Use clear headings for each of the 6 sections.
Use bullet points for readability.
Bold key terms and legal concepts I need to research further.
End with a "First 3 Actionable Steps" summary.
You can find Video Arena on our site here: https://arena.ai/video
yeah i do understand it now, the problem is i don't really do anything that would require me to use agent mode
but i'll still come up with something and at least try it
When I talk about the political meta level of specific literature and use Machiavellis Count as cross reference I expect a well abeled conversation partner to do this himself
Otherwise I could just talk to a random idiot
Sounds good. It has pretty broad use case so would encourage you to experiment a lot with it.
hey @echo aurora why does agent mode use grokipedia as a source? aren't you concerned about it using ai-generated data as facts?
Pretty sure agent mode doesnt support multimodal input, i.e. files, images, etc?
Unless that changed recently?
I'd try out the same prompt, but in different Agent Mode chats to see the difference
i made it generate a highly detailed history of arena
File upload isn'tyet added, but is being working on.
At first glance it seems good and nice. I just tested one thing just now. But an immediate I found I miss, is the stop act button๐๐
Or the stop message or whatever
Lol yeah very understandable. There are a lot of features we're in the process of adding.
Does agent mode have any feature over normal code mode?
I heard it uses few agents not one, but im not sure if thats true
Cant it make images?
And I didnโt notice anything, but at least in normal direct, you can see how the agent think before it sends or respond. I liked that a lot, so you get an understanding on how it thinks and what itโs about to do and so on
Yes thats nice too
@echo aurora are you seeing increased errors with 3 flash on battle mode, the codearena variant.
@echo aurora I'm waiting for Side-to-Side and Direct for Video
Which is no different to talking with a bot except he may deliver meaningful answers after you feed him some instructions
I like arena
might have improved
they are suffering with upkeeping textarena, this will not happen
Can you share some Trace IDs you're getting?
@echo aurora
+Good is Arena is good
Congratulations! Welcome
Will do if it happens again
Good good
YES
am no champion ๐
why can I only see select model
Hey pineapple
Thanks
Are you planning a new contest this month?
You haven't applied!
Good question, we haven't done one of those in awhile. Maybe we'll do something for Agent Mode when that releases, although I'm not too sure how we'd go about sharing/voting on that.
What do you mean by this?
Quick question, do u miss lmarena?
Contest for the best ai vibe coded app/ai application?
Like when u first came
Yeah could be cool to see some kind of most creative use case for it. The contest submission could be the prompt they've used, then those judging what is best run it themselves maybe.
Na. I personally like the new branding a lot better.
I do ๐
Did they manage to fix the excessive Captchas?
I meant like when u first came
Like new admin
pineapple do u looksmaxx?
Also nope, we've added more really creative and talented people to the team since then. We've gotten better about process, new features, etc. Overall, things are just better imo.
No need 
Tuff
Captcha system is always evolving, it's a moving target.
Pineapple can you review these 2 models?
https://discord.com/channels/1340554757349179412/1504826368855117936
https://discord.com/channels/1340554757349179412/1504160167405818039
I miss alpha.lmarena.ai, where did it go?
Where we could put the pass type code
Will it come back
Yeah I'll flag to the team. Thanks for the reminder, sometimes I get too tunnel vissioned with #general
To access
Captcha system evolved so much this week
Thanks
It's also worth noting the recent outage we had (Wednesday) has made the captcha system act up a bit. cc @bright shard
yo btw if u send like "The Video Arena is currently accessible through: https://arena.ai/video. More information on how to use Video Arena can be found in this article" to a user trying to generate videos in server, (for example u send this as a normal user and not a mod), Could you be warned?
Like when no mod is there
And some user tries generating videos here
if u send that to let them know
Will u get warned
Or is it allowed?
i have wym?
Oh... how'd I miss that....
Boom!
@echo aurora Can u answer this
ayy now we both champions
No, why would you be warned for something like that?
Pineapple can you tell all the secret roles that users can apply?
yay
Ampro check dms
I'll apply for champion soon ๐
haha i was warned for simliar thing before
It's perfectly fine to help out other users. What our mods may do though is delete the messages once they see it. As we don't want channels being filled with prompt attempts/explanations. But yeah you shouldn't catch a warn for that.
Please do!
Okay.
Hmm okay, I'll chat with the team and let them know.
I will now.
San Francisco ๐ฅ
Or idk
why most large companies at san francisco
Champion still matters
champion is cool
Yeah
How to apply for champion?
Champion is tuff
form im #announcements
Banana u wasnt champion
Thanks
Np
๐ฅ
champion ๐
Lets go
codex glaze never ends ๐ฅ
At lwast claude can do things
meta is dead
Bro what'd i do man
I hate claude
Don't compare me with claude
(im apple)
๐ญ
All around!
lol
@surreal zephyr @silent tree @light sleet noobs give me halo vip
So the 3 masterminds (ampro, watermelon, vince) are getting arena champion but watermelon hasn't applied yet soo
This trio is cool
bad tokenizer
What's ur trio name
2.5x worse than gemini, 2.4x worse than gpt
GPT Commanders?
cuz its openai
cuz its gpt
no
not ridiculous
plus and pro are tuff
codex is life.
@sleek urchin
Smart
Applied!
plus it is still surprising how bad models are actually at retrieving knowledge without a CoT
but i also agree on it being a quantised / pruned / what-ever-to-make-it-cheaperโข kind of thing
yeah
api is the same price, so there it does not matter
@echo aurora I applied
i think it is more similar to the api version and 5.5 pro the most so
@echo aurora Please make it possible to delete unnecessary chats with agents.
yeah, codex is just subsidised to get the claude code users
Why tho ?
but they will likely also lobotomise if they can
what happened to opus in battle mode
do you think GPT 5.5 is the best at coding rn
i think they are exploring ways to cut cost on inference rn
and i think some stuff was done for 4.7 opus
no like meaningfully do u think it is?
but still the same on api and chat
do u think its meaningfully best at coding rn
just say an answer
yes or no
I wont be mad
Disproven.
not as fast as 17d but ok
๐๐ผ
ok
๐๐ผ
because its tuff (ur at halo server right)
who ru saying that to
imo they did the new tokenizer for 4.7 opus because they are using some kind of sparser prefill phase. which works well with having more tokens to choose from (bad way to put it, but i think it is pretty obvious)
openai has like 5x more users than ALL OTHER COMBINED
which is wild
2.5x worse tokenizer
so basically same sentence uses on average 2.5x more tokens than 5.5
i mean they admitted that lol
nah i think 4.7 just has a slightly different attention mechanism that is technically worse
and they did not train enough (they'd rather spend their time on claude 5)
@sleek urchin
the mythos paper says it will be base for next opus model xD
for long tasks i prefer 4.7, otherwise 4.6 is the same. the difference is really small though, idk
(long tasks less in the sense of context length and more about autonomous stuff..)
oh
i think i have not seen him here in like half a year
probably in openrouter dc ๐คทโโ๏ธ
yeah i liked our convos
Where did you get this info
he might be working in tech / ml, but not as a lab researcher
there are only very few people in these roles
Bruh I still donโt have agent
bro is distilling me today
vsc
so really basic
some kind of fork for that i think
saw it on a friends pc
for like old look and functionality
nah
i probably don't code enough anyways
so can't really evaluate
but visual studio is one of the few i have tried and don't like (at all)
Um, What model is Max from? For images?
im so cooking with the dashboard ui
Hii
Max is a router
it routes to random image models
based on your prompt.
There's no specific model for MAX
Hi
Not random it evaluates your prompt and then choses the best iamge models for it tho i dont get the point, atm nano banana 2/pro and gpt image 2 are the best ones with no other competitors so its kind of useless, pretty much the same with text, i dont think we need these kind of router models but might just be my personal opinion
Hru?
Yeah that's what I said
"Based on your prompt"
I'm great, wbu
why do u talk like pineapple
Bro might be pineapple
Yea i started writing before you said that and then didnt want to rewrite my message ๐
Oh its alr xd
Im fine^^
Yea, he is kind
thats similar to pineapples response on peter bro
u da real pineapple
poor pinapple
pineapple is pinged 13k times damm
wish I was banana 
thats some crazy flex
nice emoji ๐ญ
I'm very looking forward to seein my 2026 Discord wrapped stats lol
I love banana milkshakes
lol it'll be crazy
same
+1
I finally applied for champion xd
Fr tho
Lets goooo
Ooooo same
I review them once every couple of weeks
+1 again
so I think next weeks its feathers turn
Did too dont think im gonna get in tho ๐ญ
Advertising isnt allowed here.
you definitely will
If your application was good
@light sleet Should become a mod if he wants to
uhhh
Eh idk
We need a petition for banana to become mod
its hard to be a mod
๐ญ
my top (other server) ๐ฅ
well the thing is
If banana becomes a mod
he'll prob be quiet
like the other mods
Oh
but I think its cuz the other mods are always busy0
Or hell be constantly online, never sleeping and getting pinged thousands of times like banana
imagine if banana became mod and active mod
Yea
every mod should be a fruit or vegetable
We may do community mod program in the future. For now though, I like having that separation
forwarded to database ๐ฅ
@light sleet is now excited
I definitely am.
lol
fruits better than vegetables (agreed by Producemarks, fruit performs better than vegetables)
Fruit-2.8-pro-thinking
Better than vegetable-3.4-max
Fruit 5.5 High way better than Vegetable 4.7 Thinking Max @surreal zephyr
๐ฅ