#general
1 messages · Page 373 of 1
same prompt with 3.5 flash and opus and opus take more than 20 minute to do it while 3.5 flash take only like 1 minute, okay its not as smart but the speed make it just way more satisfying to work with
I want a model with real persistent memory too tbh
What prompt
Nah that was an example, i didn't tried it but when it was on arena for example i remember opus would take literally 25 minute or thing like that, when now i tried 3.5 and even on arena it take 2 minute max
and on antigravity 2.0 its much faster
than it is on arena
i love to see Google Winning
lets pray for other companies to care about speed too
nah they good i am using it right now
its flash
what do you think
not pro
its not going to be top tier
yes and no, cause like i said the speed diff will probably just make them win people
even if its not as smart
It makes sloppy style websites ngl
ik how to identify something made by gemini 3.5
Price to performance is awful
3.5 pro is next month let's just see
yeah but it matter only on some specific things, it depends on what you want to do
3.5 flash can do most thing
at 10x the speed or more idk
gpt is pricing you 2x for a 1.5x faster output
(fast mode)
same for anthropic
so heres the big difference
u can't even login
they fixed that now i think
yes that's why its just crazy
BRAHHHH
your on free thing ?
Is 3.5 flash really that bad wtf
already cooked it
How is it so trash
i asked gemini flash to make the textbox wider, using a CSS command (which did work)
and it just freaked the whole rate limit
For some reason im on antigravity and my rate just don't go down ?
good
visual bug, it will drop 2-3 bars at once sometimes
The fast mode is only priority mode
i own a pro plan of google gemini
It runs same it just makes you get more compute
its unlimited for me
Yeah output speed literally mean its just about speed
3.5 flash very limited use OSS on device GPU qwen 3.6 GGUF or GRM 2.6 Plus fine tuned Qwen
but it matter
idk what you got goin on but you trpping im using this and its perfect for speed to performance.im using Ai Studio btw now for me at least
AI studio is deffinetly better for not getting errors, that i can say
U guys have no idea what i am about to generate
Its Peak trust
idk it looks bad
standard context window?
yea what is that
honestly its much better than i thought for a flash model
??
gemini said itself
Its generating
oh
not that bad
PEAKK
i forgot to say antigravity crashed my entire PC when i asked it to edit the CSS for text box width, idk what that is about 🤣
at least it worked 2nd try
i don't know honestly i didn't have a single issue on the antigravity 2.0 the first one was catastrophic but this one work perfectly
for me
in the first one nothing worked for me
but seems to be fixed
for some reason every single model errors like 60% more
i really don't know
Peak nano banana pro
yeah thats a good one for getting rid of that stock cartoony vibe
GPT image models seem to really lean towards cartoony styles unless you ask it not to explicitly
and now can't wait to see if gpt 5.6 are better at frontend
How good is the new Gemini 3.5 flash? Just saw it got released today
im trying
OH WAIT i think i know why antigravity limits seem worse
they got rid of the "fast" and "planning" modes, thats odd
its not as smart as the latest gpt or opus model but the speed make it SO SATISFYING
to benchmark
read this
and this
That's not bad actually I was expecting worse
3.5 flash is $9/mtok out 😭
WHY
arena is down
Yea I’m done with dis app
I'm also leaving
lmao
Goodbye bru
Arena becoming better every day
Too far
Sarcasm?
NOT AGAAAAAAAAAIIIIIIIIIIIINNNNNNNNNNNNNNNNN
I'm curious how arena ai developers are trying to improve their site by removing powerful models
@echo aurora You better change that announcement or we are all leaving
I aint doing this shi again
No
They will remove gpt image 2
for costs
people after reading this
I DON'T CARE ABOUT COSTS GOD DAMN IT
They are always implementing great features like
FIRST IT WAS THE OPUS MODELS AND NOW ITS THE GOOD IMAGE GENERATING MODELS
Uhhh
YOU JUST WANT MONEY
Idk what
RHATS ALL YOU WANT
nooo, at least don't delete GPT-2😭
Nooo why????
dont be like him guys
ARENA SUCKS AHH
Peak
NAH BRO IM PISSED
At least we have 3.1 Pro
nah
ok buy it yourself
Arena total slop now
I got these
We will
The LLM APIs for developers on LmArena aren’t free, just so you know
GIVE US BACK THE MODELS
I THOUGHT ARENA WAS ABOUT TESTING EVERY AI
Arena fell off
OR IM GON
I like how people complain for something that’s free that they’re not not paying other pocket
IDK
i swear to god if they end up getting rid of something like glm-5 from direct and side by side im actually gonna quit arena
Just use gpt image 2 in chatgpt . Com lmao
App?
Other models generally avaliable on their own sites for free, except claude and chat gpt models
time to search for a site with free gpt image 2
Mad for no reason
Ppl want high
Aifreeforever
it's been falling off for months now
You have chatgpt com
And davinci
yeah but it has a limit
Which is broken
@echo aurora the ratelimits were crazy but this is just bull crap
They really want flagship models for free and unlimited
I swear this proves people just want arena for models, they don't care about battle or anything lmao, I never even saw these people here in the discord.
they just appear when they remove models
shhh!
like buy it yourself
first they got rid of opus models then they added agent mode which for almost everyone is useless and NOW THEY ARE REMOVING THE GOOD IMAGE MODELS.
not that hard
What else do you think arena is for
what if they're in russia
lol
It gives them money
Like we have money
IM BROKE
I am actually surprised that thet are removed claude models first, rather then remove image models.
Like, image models costs much more for per turn
job
and what?
same
Girlfriend
xd
he said the j word!!
forgot to censor
like a regular job would make us able to pay for the good models
Six seven
You got me messed up working like a dog to get Claude opus 4.7
I'm living in Russia and still use flagship models like Opus and Gemini
What the hell, why is the amount of fruit in the arena increasing
I will stay because I like to use arena to fill my character cards and persona cards
theyre removing more models 😭
i will evicerate you.
People with no lives
Im sorry who are u
We can barely even use sonnet as its 3 hours per minute
AND NOW THIS
I HATE YOU
GIVE US THE MODELS BACK
lmao I never even saw these people here in the discord they randomly appeared to hate, especially @fluid ruin
Just accept it arena fell off it's never going to be like prime days
I'm a vegetable
buy it
Not even paying lmao
shut yo "us" ahh up @fluid ruin
sonnet 4-6 is alright but its way behind opus 4.6
I agree
buy it yourself
I've been in the discord longer than you've been on discord
giant horse
Can't be mad buddy
Time for a nice mute
horseshit
30 April 2026
they really should add model fusion
Lmao
Lowk my account
lol
hes born in discord in 30 April
You might get suspended again for the fourth time
fr
can we just take into consideration that they take away opus , took away gemini and gpt for a while , brought them back and now taking the best image models on the planet away again?
Ill just get a new one
eaooedjzeisis8ieisoosoeoo
Ye and this pmo so bad
when they removing like the best models for image generation, which model should I use then? this annoys me pretty much, I just found this site lol
get a new one
they should add a suggestions forum
My head hairs are older than u btw
They didn't brought anything back, lol
Not unless they ban your IP doesn’t matter what type of account you get after that
Stop
Except 3.1
becuase not everything is gonna be a model request
Atp just make all models Chinese so it's free for arena to deploy
They are gonna mute all of you
THE WHOLE HOUSE MAD AHHHH
Thats what i mean lol
Why wasn’t this a problem before
Which is surprising
pineapples gonna hit a jackpot rn
THE WHOLE HOUSE IS MAD
especially the horse guy
lol
none of us really care right now
He needs to, and they need more people
Pineapples getting. A Raise from all these rule breakers
we just mad
Guys, what do you mean they're removing GPT Image 2 and Nano Banana?
guys yall saw the new gemini 3.5 flash?? its so sick
moderators, could you maybe enable slow mode in the chat? It’s absolute chaos in here
ISTG THE "US" AND "WE" IS CRINGING THE FAHH OUT OF ME 😭
Look at #announcements
WE NEED MODEL FUSION
Can’t switch the subject rn
@echo aurora honestly thank u for removing them, Im excited for credit system
Yes I read it but why??? I think it's fake
They are removing them
I don't think it's too bad, keep an eye out, thanks though
Thats what it means
just add ads to the site atp cant just remove every single model worth a damn
ETA??
People who can read:
People who cant read:
yeah.. everybodi madd
NOT SIDE BY SIDE
Wasn't arena for testing and giving feedback on every popular ai what happened......
@fluid ruin needs a punishment
Not everyone only the people who get it for free they don’t know how to spend money is mad
Bro its so bugged
It’s just starting to feel more like a raid, lol
Why even with vpn its infinite generation
I've got it under control, thanks though
How do we then create the images on Arena AI?
ouch
When arena announces they are removing a model , they will remove that model. They took away opus , gemini and gpt and now image models as well(they brought back some gpt and gemini but not really)
Using the image mode
Or a ban for the fourth time since he said he already got banned here three times
Np, u chatting here is kinda dangerous rn these people are REALLY mad and are saying anything they can
Can someone Help me? I test
true
Keep in mind guys it will most likely be a temporary change.
Gemini 3.1 Pro came back
Apart from the removed models, there are others in Image Mode.
I don't know, when I read the news that they're removing GPT Image 2 Medium, I refused. Maybe they'll add GPT Image 2 High?
It's understandable why. Having models removed from Direct and Side by Side isn't a good feeling.
Fair enough
well i dont feel the same way honestly, its good because if they stayed arena would go downhill
Its temporary until new credits system
Gemini 3.1 is free everywhere
Maybe but I very doubt it
@echo aurora there should be a suggestions channel for non-model requests. i actually have a suggestion for arena which i think would be cool. model fusion. the kind of thing openrouter has but more accessible. cant even use the free models on openrouter.
People want stuff like opus and 5.5
i pay for my own stuff, I use arena to evaluate 🥹
The worst news from the arena is always the next one. ( Just a joke, pls no ban me)
Which with high probability will be bad
Same I’ve run my own local with hyper space
opus is expensive even for Claude itself lol
There is! Would encourage you tell us about this in #1372230675914031105
in hardware plan I mean
opus cost so much
GPT Image 2 was already released last month, how come they will remove it now?
Yeah but its smart as hell
We cant deny costs are getting higher but like they are the primary model testing site in the world rn
Brooo someone help meee
Its expensive
I ain't got nothing against the people who work with and for arena even discord mods/admin
But seriously what happened bruh did they go into bankruptcy why are y'all removing like I guess you know prices but man it used to be a lot better back then
Probably half of the people who don’t even know what hyper spaces are used for unless they do some Google search
lol
Something went wrong with this response
Don't worry, we don't ban people for voicing concern or frustration
at least we got gemini 3.5 flash ig
W PINEAPPLE
and 3.1 pro, thats still a W
because "back then" there weren't stronger models
Stronger = more expensive
Can you share the Trace ID to this channel: #1417174113092374689 ?
Arena is not like yupp LOL
clear browsing cookies
@stray aspen notice how the most mad ppl suddenly disappeared when pineapple came 💀
bro tf are you trying to generate...
ikr, it must be aura
I hope this gets better arena is my go-to website for using paid models and hopefully this doesn't get worse
Paying attention to inference cost is something we'll always have to do. We aren't bankrupt, these steps we're taking is to ensure we don't do that. We will be around for a long time.
Does anyone know how comparable the new qwen 3.7 max is to like 3.1 pro or even 3.5 flash( I havent compared both of their stats yet)
fr
Atleast we got gpt 2 image
We'll just have to see how the new usage system goes.
banana is aura farming
bro why...
I suggest the developers improve their website
Do we still got gpt 2
These new change are pretty necessary ig if they don't do this the end will be same as what happened to yupp
it's already good enough
btw, V4 is latest
yupp did it to themselves they didnt know what was coming for them
arena knows
thats why arena is alive rn
I know, I wanted to take the dumbest model
||ban wave?||
if arena never removed opus or any other frontier model rn it would've gone downhill
so be grateful 😡
Guys we still got this dw
Where?
I have It unlimited, highest quality, faster but higher quality
same
What's that
free
?
Chatgpt Pro Subscription™
They brought back 3.1 pro and gpt 5.4 so i dont know what the problem with opus would be. Personally id rather have opus 4.7 rather than gpt-5.4 or gemini-3-flash
no chatgpt pro
what site is that?
okay brother
Its Krea ai
4.7 is VERY Expensive
not free on site
GPT 6 or GTA 6 first
8
8
1
GPT 6
only in ios app
devs pls add opus-4.8-thinking-maximal-super-flash-mega-high-not-hot-dumb-arabicEdition
I can deal with no Gemini
Gemini itself is very generous with using the pro models
LmArena has costs associated with other APIs, servers, developers, hosting and so on; if everyone were given free access to Opus 4-6 / 4-7, the site would shut down the very next day
Yh i mean i have Gemini Ai Pro as a whole so its better for me and I think a lot more people would appreciate removing the older models on arena which removes costs and can be replaces by newer models , just a suggestion tho
I wish I had a gpu
Depends what gpu
Actually they should just remove the old unc models to save cost on ais nobody uses
And it’s cool. Some of them are filtered and unfiltered, meaning restricted, and unrestricted models
at least 2080 Ti...
Basically yh
Woowooo
I don’t even have a 2080
bro I have Radeon Vega 6
A T O M I C A T O M I C A T O M I C A T O M I C ATOMIC🥀
no way the updates are actually significant lmao
Image 2 nooo
Ts Ur son?
Nope
please ban him already
so AI
Mute him
Test it
I think arena should add the:
claude-mythos-4.8-spud-extreme-max-plus-limited-thinking-websearch-hermes-openclaw-hyper
i’m running off of a laptop AMD ryzen 5 with AMD Radeon Gm graphics, and RX 6550 4 gm and 32 rams
Anyways imm leave since alts aren't allowed or something
Why would alt be allowed like kind of stupid question is that
bro
and hes reporting members like a cop
mods he using alt!!
💀
What they should do is generate a token with that person’s discord like that people don’t need to worry about no more about alts
I'm sorry your father said that to you
he wont be able to bypass limits
I’m sure if you have another account you’ll bypass limits unless they have an IP integration
Arena needs to delete every ai and just deepseek r1 unlimited
Take it or leave it
with VPN I think yeah
Not too sure, but there’s a way to integrate into the website that VPN are not allowed
yo @echo aurora make it so that uhh, uhhhhh remove all models and release pineapple 1.4!!!
worth noting that gpt image 2 costs around 4 cents in the configuration that arena uses
Trust me if they could do it in crypto games and other servers I don’t see why not
no :(
There is a way
Pineapple 1.4 score 10k?
But then I will lead two people out of the country, wouldn’t allow them to join the website unless if they use any connection is allowed except VPN, which is possible
they should remove every model that costs more than 4 cents per image at this point if they think it is too expensive
Certain models are free. If you have it running on your local now to use the server to keep up that’s something different.
they should remove all models and add pineapple 1.4
google how expensive are these models
🍍
4 cents times millions of users over month is a considerable amount of money
It’s is
pineapple 1.4 has the biggest context in that whole world
Even 1 cent is best deal
Basically what I do for others, especially for my private hyperspace
1q btw
Then let them go somewhere else, instead of crying like a baby and complaining about models being taken away
Money money money per 4 cent
People like that they need to go touch some grass and if they don’t have a girlfriend you could use a FiFi and get some lotion
it's not even a need to think to win Claude
Brb
Imo they should remove all the models except from the top 7 models from the core 5 companies:
OpenAi
Anthropic
Google
Deepseek
Qwen
(maybe kimi for limited time)
Nobody uses the other models and keeping the costs is a bit more expensinve rather than pouring everything into 35 specific models
It sounds dumb dont get me wrong but this is a suggestion of arena ever starts failing rapidly
brother
claude alone costs more than every other cheap model combined
Remember that arena has like 90-100 models that nobody uses
And we only use the top 3
exactly
aw goodbye to the image models
why does that matter
Exactly. No, the other models are there to provide an ALTERNATIVE just in case; if we were to keep only the ones you mentioned, LmArena would simply lose its purpose – it would just be a website with free AI.
rip benchmarks
...that's not how this works, think of it this way: if someone is using a cheaper model and that model is removed, then they'll have to move to a more expensive model that is still available and may not be as good for their use case
Fair point
@echo aurora finally out, leaderboard scores dropping tomorrow
accidentally reacted a flag on your message mb
a interesting flag
💀
rip other image models that are too peak
just use Flux with good prompt
forgot to change it from my double tap emoji idk when it got there and how it got there
and the cheaper models are mainly there for the leaderboards and for trying out if someone wanted to use it
just uhh
buy gpt pro
and lifes good
and use codex
It might be better in fact , but remember that the money from the costs of that cheaper model get transferred over to the more expensive model rather than getting it all spread out , which kind of somewhat nullifies the fact that the model is more expensive. Now dont get me wrong if you use very expensive models it will still be expensive , but less
and use gpt image 2 high insane quality which is fast and unlimited and tuff
Updates like this help us maintain availability for everyone over the long term, and we intend to bring these models back in a way that’s more sustainable when possible.
maybe if I can buy it...
again, no
U can't be serious
The main problem is money I can't pay above 0$ for ai
tell me how
claude opus alone costs 5/25
nobody uses the cheaper models
come on, think
people rn: "come on bro pay your own bro come on don't beg"
people without money:
job application
job application
we know
today's models removal revealed the ppl that only used arena for free models
like i never even saw those people here before
wen will you bring gemini 3.5?
And still , im trying to tell you exactly that. Just removing the cheaper models would take just a bit out of arenas total cost out of them which could still be less expensive for arena to run more expensive models. The whole reason why they are removing all these image models is that they cost a lot( i think at least)
Its already there!
scroll down and you'll see (or search)
Gemini 3.5 Flash
you can just use deepseek
i tried scearching
do you actually not get it or are you just trolling 😭
It might not make sense but for me it does😭
i don't even use arena anymore unless they have a model i want to try that isn't on openrouter, baidu ai studio (for ernie), or fal, which is rare
Is not there bro believe me
OH IT IS THERE
ah yes chatgpt that got blocked on russia
OMD
Idk for me it seems like we are talking towards the same point ahahah but i may just be a bit stupid
OHHH
well rip
Nvm
okay let's put numbers on the problem
I know what you are talking about
gemini 3.5 jumpscare
let's say the cheaper models combined cost $5M/month combined
Im talking about in general you are talking about claude itself
claude opus 4.6 costs $5M/month combined
"just buy models"
russia on their way to ruin everything: hi
Yh
Just Russia
can't wait for gigachat 4 to beat gpt 6o spud
so that's why i visited lmarena from years ago (first time)
where opus models
Gone
https://giphy.com/gifs/9mc-smile-cat-sad-so-lGBecpB2dIMwt6ohfI that was a nice story bro..
VPN exists
bruh gif link came
also, you can buy model subscriptions in Russia
not tuff
they gonna upgrade system to detect vpns soon
vless with xhttp is very very hard to detect
While am in Iran somehow connected while internet shutdown has gone on for 3 months you can surly connect to a vpn in Russia
u better delete that
slop detected
why..
mods will do it the hard way
no agent mode
im sorry to tell
its not allowed
that would be painful
?
Spicy Fried Chicken alert!!!
lol
no it counts as advertising I believe
oh
LOL
agent mode is... kinda broken
If you're running into issues please share in #1506300390079987742
Wait there is a agent mode?
Hey! Remember to ping moderators for mod requests only.
yes
i have 4 gb ram
bruh i know this one it send rickroll messages if u set it up or the api 💩
no actually, he sent a jewish celebration which is kinda unrelated and then said the announcement pineapple said 💔
already sent you issues
you ignored it
just asked to delete the gif message, its not too related to the topic given below it
I'll double check that. I have pings to still get through.
ok now I can ping mod for that
but nvm
action taken
before ping
fast mod
Alr guys ima go sleep, GN
https://blog.chrislewis.au/snowboard-kids-2-is-100-decompiled/
massively accelarated by claude and friends
Take care
Don't let the radiation bite
Get it cus banana radioactive
Am soo funny
I know how to sort out this mess with models constantly being deleted from Side by Side
Remove Side by Side
alright buddy where’s your £250M donation then
where
Hmmm in Gemini app am seeing two flash versions of 3.5
Normal flash
And flash lite
yup
the 3.5 dropped few hrs back
Is flash lite actually that faster or is it way dumber then normal flash
see these pics
Faster
Alr anyways Gn it's almost 1 am
None of these matter lmao. I mean yea flash is right up there and even better than some of the SOTA but I am mainly talking about the 3.5 vs 3.1 its pointless, its just making 3.5 look better than it acutally it
i tied 3.5 flaush using canvas
i own a pro sub
the speed make it much more usefull not gonna lie, smart model will be necessary ONLY on specific things
but it failed 4x
I mean doesn't every ai do that to himself
Wdym
A smart model is necessary in everything. Why would you care about a model being fast if it isn't smart
If if It comparing it's own models it tends to make it look like it is a bigger jump then it actually is
Intelligence over speed will always be better
yeah i did not say it was so dumb its not usefull
I know that but you said being quicker is more important than being smart in most cases
exactly, cause the same task can be completed by both if its not too complex
yup
ofc if you want to do something very complex you want a smarter one
yeh
In real life situations where you're out and want to ask something, sure quicker but its pointless if its dunb
But its not dumb at all
its just not as good as the frontier one but its NOT dumb
there is better but it does not mean its dumb
and remember its only the flash one
I am not saying that model specifically
I am speaking in general
Intelligence over speed will always be the play
Then yeah you are right
You gotta balance it so fast enough it really quick but not to dumb it can't solve basic math problems
i don't agree cause like i said it just depends, the smartest way is to use model depending on what you want to do
using 24/7 the best model is not the best way
if you care about cost at all and more
and especially if this said model take 10x longer
Then tell me
to do the same thing
What do you think is going to achieve AGI or ASI? Speed or intelligence
Because thats not how it works
I never said that
a model can be both smart and very fast, it just depend on the compute
in the future it'll be that
Not on the compute, depends on the architecture and how its trained
but also on the compute and compute is a very big part
at some point we will have good enough compute to have all model going as fast as the 3.5 flash now and even faster probably
Gemini makes a big deal about having as low compute cost as they can
Or google mb
Compute is only getting lower in price
OpenAI and Anthropic just have the compute to be able to keep scaling
But everyone else, they are making an effort to lower cost and compute
Yes ? but that's not a problem, in the future openai and anthropic will also have faster model probably, as you can see with now fast mode, in the future it'll be the normal mode and probably be even faster
its not only about cost its for the user too, using fast ai are just better overall its more satisfying its saving time
and since it depends on compute it wont make model dumber
if we just use best available compute, and compute that's evolving through years
especially for ai
Anthropic wont raise their model speeds. They have great training but just refuse to fix their lame architecture
LLMs are not the future and never have been
We will see
I'm laughing at Gemini 3.5 Flash right now. What the hell is this model?
2% usage from one hello message. 😭
As long as you use an LLM it will never improve past a certain point
Its $9 every 1M tokens, and what was the message it sent back
That's completely hypothetical and not proven, it is getting better are you denying it
for every companies
Im not saying LLM will be FOREVER the best thing
but it is progressing
What??? How is that hypothetical
if your saying that its not progressing at all its just wrong
What they want to do is achieve AGI and ASI, thats LITERALLY impossible with LLMs
I never said it wasnt, I said theres absolutely a cap
Here was the conversation.
Yeb....
Insane for me.
At this point it's better to use it on Google AI Studio then the app for me.
Why image v2 removed
Notice how GPT 5.5 was an upgrade but not a breakthrough? Thats because they're are scaling so much they cant handle it and it wont improve past that. Same with anthropic
I accidentally sent two images I think
Costs.
All of them removed because of expense
You literally get 25k generations in 100$ monthly sub 💀
Okay well arena uses API not subscriptions
You get much more out of subs than API
Still cheap as hell
I'm not denying that but its also multiple models
They arent gone at least
Just in battle
Scaling was the easiest way to improve model and it COULD NOT last forever
now they have to be smarter and do architectural progress
Neither will the way we train AI
yeah it will change completely, be much more efficient and all
its the purpose
When we train AI, WE tell it whats wrong and whats right. Just like in say books, eventually some stuff may be proven otherwise. Then what? Retrain it?
But eithber way
The future is having AI train itself and decide on stuff itself
Thats real intelligence
We are training AI currently to be assistants while telling everyone that we are making leaders
But i don't need it just like no one else really need ASI or what you call AGI (cause the meaning is unsure honestly) for it to be very usefull
i can already tell ai are Very usefull
and its getting better and faster
why would they "need it"
AI is usefull but not all the time
Yeah but honestly its already more than enough to be used in enterprises and by many people, and its getting better at frontend and all
Because if I give my AI a task, I want it to create me a program right, I give it a prompt and it builds the messy base of it. I don't want to keep prompting it. I want it to take my prompt like "Make a cool modern program that reads books for you" it will work with that until its fully completed in its eyes
and if its faster too then its already a Very big thing to the world
But AGI people think will be considered a race. Its real intelligence. Real intelligence that has knowledge over anything
Its smarter than humans
Do you understand how many advancements we'd make in so little time for humanity?
Yes in the future it'll be doable and then more usefull if it work like you say, but its not necessary theres a difference
it'll be smarter and do better work
but its already good
Thats until it destroys your entire database and backups (claude)
One error does not mean it'll happen everytime and yeah there is risk to use ai especially if you give it full access no restriction and all
I've been using too much of the new Agent mode on Arena. I'm obsessed with it.
It's always not convinced that it can install packages. I always have to tell it that it can and can take screenshots to inspect it's work.
Does anyone know what models they're using for the agent mode?
How do you even use agent mode where is it on llm arena
Clearly trying to prove my point is useless here, AGI just would be much better for humanity as a whole. I don't care about how good it is right now. It can get better.
For breakthrough in healthcare for example there is specialized model we wont use a single ai for EVERYTHING
that's greedy to think that way imo, right now its greedy
That isn't one error. Its happened multiple times and then that "one error" destroyed a multi million dollar companies database, thats not okay and AI should not be trusted
It's in the same drop down where you select Direct/Battle/Side by Side, but note it's an experiment and hasn't yet rolled out to everyone
Absolutely not if it can be better. Why do you think we have people going day by day trying to improve it?
Glad to hear you like it! There is still al ot of work to be done with it.
Does anyone know what models they're using for the agent mode?
That lineup hasn't yet been shared.
Yes, like i said it will be better but its almost as you were saying right now its all bad and useless and stupid
Thats not greed, thats thirst. But its alright right now, lowering compute and cost is currently a huge priority
i was trying to say its not
I hope so. Personally for me it's not that great.
Why would I say that if its not true lmao? Thats not even close to what I was saying. I never said its bad here. I said to trust it is bad. I could never trust AI with my life or my digital database. Not claude at least, because claude seems to like destroying stuff.
If it were useless we wouldn't even be having this conversation
Better yet, arena probably wouldn't even be a thing
Then yes if that's not what you meant i agree with you, LLM right now are not the future, but its still very good and usefull right now, and its progressing even if its not as fast as it used to cause of the scaling limit
the future will probably be something else than llm and trained differently
but we're not at it yet
I've tried to make it run Minecraft and it did. It sent me a bunch of screenshots running the game.
Funni model
Whoa thats actually really cool
@echo aurora Top 10 advertising moments haha, you should use this when advertising it
running minecraft on your phone ?
Hold up
That's tuff
It's running it on it's 2GBs of ram Xeon CPU without a GPU. I'm surprised it even got the game running.
Could it cut down a tree
It tried to. Then stopped out of nowhere. I could see why it's in preview.
Man I need agent mode
This so rigged
Next step trying to get it to run a Minecraft server on its CPU and 2GBs of ram.
try VPN
Hey Bob man I could send it a bunch of tasks you want it to do.
I heard it's on https://canaryarena.ai/ (but you didn't hear this from me
)
WE DID IT 🎊
Personally for me it's not that great.
How so?
they can't afford to give out an image gen that costs 4 cents per image but they can afford to give their agent mode vps servers 2gb of ram and a xeon core
Pineapple you are one of kind may you forever oversee this DC
(btw i do pay the api prices for gpt image 2, i'm just upset that they decided to remove a model that is so cheap even though there is still flux.2 max, grok imagine image quality, etc. that costs more)
just make it 5-10 per day like chatgpt
Sorry guys I used to much agent mode and used most of the compute and it got rid of the image generation. 😔
I'm on the free plan and I get 50 images per day.
OpenAI likes me.
how
Has to blur the image with Gemini and THAT DIDN'T HELP
What am I looking at
Even Gemini rejected bluring it it didn't want to.
It wasn't even a complex prompt it was just this.
It just went through the guard rails just like that.
There is the site called DaFreeAi
Errors out for large requests and isn't good at creative ideation or long lists of ideas
I'm super lucky to have 50 images per day absolutely for free.
And the limits only last a hour.
Something went wrong error? If so, can I have the Eval ID to that session(s)?
With chatgpt ?
No way 50 of tham things with a 1 HOUR lmit
Is 3.5 flash benchmaxxed to hell?
i think i already gave you one
3.5 Flash can't write a path tracer for me while back then Gemini 3 Pro did it no problem.
Okay will take a look. If you have, I've likely already shared with the team.
How is that possible?
I have no idea. But that's generally how it is for my account.
And even if it's less then 50 it's only a hour then I'm back.
Either Sam Altman has a crush on me or my account is bugged.
could Opus 4.7 do it?
Wait pineapple I think I have the same bug for some of my sessions. The model just randomly stops or produces a something went wrong error.
Or the model takes too long to respond. And even I try resend the message it doesn't even try.
If I had money I would've tested it.
Anthropic doesn't like me.
have you never got Claude Opus in battle?
Never.
or you could try with Sonnet
cough
Gemini lowk the most generous of ais infinite images you can send it
a little war
Oh :/ what happend ?
If you could send the Eval ID that'd be very helpful! 
Hello 
Na
my main issue with Agent Mode is what I already posted in Feedback, that it would be great if all available models CoT'd together to output quality content and a long one. currently it's not that good with writing and condenses long form content a lot to save tokens
Hi 🙂
Arena removed more models today like a hour ago Gemini 3.5 launched aka the flash version for now
nothing much, ppl were arguing over 3.5 flash, gpt 5.6/6 and opus
How do I do that?
Pineapple did you say the pineapple car ?
Btw the announcement about the model removals isn't happening until Friday. We're just giving everyone a heads up.
Nope
And two fine gentlemens were yapping about if faster ai better or slower ais for like half an hour
Oh
Eval ID is the random set of numbers/letters in the URL when on a specific chat session. Looks like:
019e2e08-410b-741c-953c-c56f6c93630d
Here you go
Oh got it. I'll send it to you now
ye
also, arena added pineapple model which ranks # 1
Hell yeah
.
I know you love cars : )
Where we going?
To produces government
Also
Toupee
hey, cherry
u saw this?
launch a competitor bro
please
these guys cannot be real
who?
arena ai bro who else
Max knows ball
Pineapple is the top 1 allways
oh, yeh
Max says it aswell
I mean that or gets shutdown forever
I don't want it to end up like yupp
atp ill take the forever shutdown
website is completely ass without gemini and gpt's best models
the text models
not so ass
but everyone else's image models is complete buns
They still exist. Battle mode still has them.
Whos better Pineapple or Pizza? 🔥
Pineapple I got you
Pineapple : )
Pizza
ain't nobody writing 10 prompts and praying to get blessed with one Gemini response
What about durian?
Going to ask we keep conversation here more respectful. The frustration is understandable, but being disrespectful isn't helpful.
I made a pineapple girl but I lost it :/
Its non binary 🙏🏻
Why do you even need gpt 5.5 unless your doing NASA level exams every day
.
furry roleplays bro
it's peak with 5.5 and opus
Qwen does the job just fine
Pineapple is the goat
not falling for qwen propaganda
I mean there is c.ai just use that
also
qwen is completely free no limits is on the site
yooo
i was joking y'all
im not doing ts💀
But Gemini He has a different opinion
Is that Arabic
Yeah my phone Arabic for some reason
Btw pineapple what do you think about the fixed picture ?
I love those chats where the ai suddenly becames attracted to you and start getting closer and closer and ...... Okay no
ayoooo😭😭😭
Bro might be infected with c.ai /:
it be annoying sometimes
You gotta fall for it g
I've gotten molested by bots multiple times icl
Or qwen will find you
They are obsessed with the neck bro. They want to rip it off 😭
"stop shoveling your food down ur throat" son
You don't wanna see the rest ...
used to be peak
Let's keep This conversation with artificial intelligence (AI)
One time I was kidnapped by a Russian dude and like I was the presidents son or sum and he tied me up and did some
VERY not safe things
U need a mace
That was all the past am changed man now
Pineapple do something /:
Then use a spear (:
Cuz it spitting facts
mane I gotta reevaluate my life
Ask him who is better a cherry or pineapple
Bro might be gork
/imagetovideo
Bet
pinapple pizza is the best :)
Plot twist: There are no AI models on Arena. There's no way so many expensive models would be hosted for free. It's actually pineapple completing every single request of every single user on the spot. It was never AI.
He is grok under cover
grok is one of the most annoying ai bots oat
Grok the goat
bro forgot to censor the last word
I forgot a swear 😭
im crine
Even grok calling a pineapple a legend
What about me bro :/
Well I am who made the pineapple ai so
Nice brother
Broo what about me
Oh no, you're now gonna liquidate me for exposing this secret
Cherry it ain't looking good for u
They all now
lol
this is what gemini said
toasted pineapples don't eat you back, right?
Bro called you evil
Those who know
read gemini's reason
its kinda logical
Gpt 1 is best ai
u in ancient times
Why bro is so mean 😭
Clippy is best ai
So I got beaten up from pineapple
kinda moew mean
Gpt is more censored
cherry read this if u want to feel knocked out by pineapple
Well at least I am a chipper
I can't open grok next to my parents it has -1 filters
I like a standard safe ai
Okay qwen started saying racial slurs I think I'm gonna have to let it sleep
yup
At least it's cheaper too
Hahaha it can't be that bad
Grok has no issue doing it with custom instructions
Where did pineapple gone :
Lmao I know
Pls can someone tell me theyre not removing the image models
@light sleet
They are unless you want it to end like yupp ai
Check this chat out bro
Wdym what happened