#general
1 messages · Page 340 of 1
is it gonna be free
🤣
never bro
5.3 instant is free
no spud for me then
just like no free user can actually use opus 4.7
its not
If it has api day1
wdym i just used it on lmarena
you'll probably be able to try it on arena yeah
They gatekeeped 5.4 but not this
yeah battle mode and in the arena environement
I have a problem with arena ai can someone help in arena ai if we go to direct chat mode and select gpt 5.3 latest chat it shows that we can not share any files like picture pdf etc but in chatgpt app gpt 5.3 support files so please fix it
They make 6o, rename to 5.5, and not even gatekeep
Chat are yall talking about 6o
what in the infostealer is that
arena disabled some features because they are broke
fr
RAT
Mean?
Account thefter
ampro is 6o gonna release today
They renamed it to 5.5 but yes
So its disappointing
theres nothing to fix gang
😭
thats how they made it
5.5 and 5.5 pro today
Yes it is
oh good
Thank you. First time I see this one
So renaming it from 6o to 5.5 is dumb
5o or 5.5o wouldve made more sense
its not good enough to call it 6
Its new base
but guys don't expect it to talk like 4o, 4o was very sycophantic
Exactly
Fr
I’ll double check but it doesn’t have that functionality enabled. Will flag if this is something we can change.
5.5 sounds bad
Yeah
Please fix if you can
how are we judging model over a name now, going +0.5 for a new generation was happening but year ago
its long gone now
every new gen is 0.1
for most companies
6o would make more sense for an omnimodel tho
i don't know if its omnimodel its ampro that say it but idk
Yeah
But its new base bro
It is omni
i looked everywhere on X and thing like that no one ever talked about 6o
🥀
guy did supergork peak?
no
super gork
im dunno you guy use it may i get a review?
even 5o no one is talking about it
mythos is way better
so idk bro where you get those info
i still have no idea what mythos is
its the most powerful anthropic model
mm sure
anyway posting this again cause this is weird
i even typed spud omnimodel and gpt 5.5 omnimodel and again no one talked about it ever
Its bad
but like gork is uncen right?
Because its spud aka 6o
image 2.0 is stupid
Exactly
but it makes good images
people when a codename called "boomshakalaka" appears: thats gpt 10!
like no offense but
gpt image 2 is the spud
its a new unreleased open ai model
Spud is gpt 5.5
spud is openai next model based on a completely new architecture
its 5.5
it literally release today
spud aint 5.5 tho
Bro..
When is Sora 2 going open source
never lmao
They renamed it i guess
when is open source going Sora 2
5.5 pro is arcanine alpha
nah it cant be that bad
Sora 2 model needs to be passed to another company that can host it lowk
its a shame that its gonna die
i don't know about if its good or bad
It is cuz 5.5 is omni
in a few months
im saying its 5.5
Spud/6o is released as 5.5
your the first one i ever saw talking about a model named 6o
no one else talked about it
Sam altman said on live - its new base and its omni
Base = first digit
So its 6o
yes but he did not said 6o
they gonna name it 5.5 that's 100%
sure
Yes
And hence they name 6o "5.5"
It means its bad
or its just that's what they've been doing for a long time now, every new model its 0.1 +
Every finetune is 0.1+
This is new base
First digit
but they still name it 5.5 anyway i mean i don't think it will ever happen that gpt will go 0.5+ now
or 1+
Bro its not about increase
it was happening year ago
Its like minecraft
1.21
Imagine minecraft 2.0
This is spud
New, from zero
Omni
Yes but 5.5 still mean something its half of the 6 so maybe now you can expect a very good model every 0.5
we getting a useless fine tune before the spud
No because its confirmed its omni
UNLESS
they release 5.5 and 6o today
And 6o is gatekeeped
Guys
5.5 is api
nah thats not gonna happen
🤔
When do you think gpt 5.5 is getting released
In 2 hours its confirmed
It's better be better than Claude 4.7 opus
4 more hours
Its confirmed omni model is coming today
So either 6o is 5.5, or both are coming
It is
I'm a newbie so can I know what Omni in chatgpt means
Can watch videos and audio
Gpt 5.5 spotted.
Ohhh
but um
how do you know
5.5 won't be in arena soo
every gpt model has been bad at frontend
Uh why?
but good at backend
It was leaked in codex + its omni
Is deepseek v4 omni
It'll be in battle only
i just hope your right
I guess because it's expensive?
frontend level of claude but pricing and usage of gpt
Exactly
that would be cool
I wonder what would be the prices...
Opus like
no it's just that arena doesn't put any frontier model in direct chat anymore
cause all of them cost too much
for arena
Fair point
And pro will be probably 200$ for output
is deepseek v4 omni
yeah always pro model, i just don't see point of using a pro model anyway
Math research
DeepSeek is good but shorter responses and not as good prose
Probably
Js drop it already fr 😭😭😭
Yeah, but for doing actual task the only purpose of using pro is if you want a good one shot, but usually you want to keep going and keep working on it
bro is sam altman
and keep working on it is not doable using a pro model lol
We will just end up using deepseek v4
Deepseek sucks, no good vision
600$ god dayum
They say v4 will ve omni tho
Oh shi better not this again
no it won't happen again that's old price
of older model when it cost much more than now
O1 was big.
Spud is big too
yes but now everything work differently
the price just ain't the same
we'll just end up using deepseek v4
Mythos:
mythos will be big
even mythos won't be 600 dollar output
deepseek deepthink will be big
It uses 2x worse tokenizer than.gpt
So resl it is 300 out
backpedalling is crazy bro
Fr
no for u
No they are bro
you were adament on it being 6o
is gpt 5.5 coming to arena?
5.4 will
ohh that's still cool
right now you can almost use 5.3 codex with fast mode and it'll be same usage as 5.4
almost
WHAT is the best ai model for now
I use 5.4 xhigh fast
You don’t need vision for some models
but you don't care about usage ?
DeepSeek has been an amazing number cruncher for me
Pro100
how did you get this
its a tweet ❤️
Late news bro
yeah and it give you enough usage ?
to use with fast mode
i guess it depends on how much you use it
It gives like 50x plus usage
In practice
Lol
isn't it like 5x but actually 10x with the boost
10x of pre nerf plus
5.5 in base64
Thanks
so much teasing and no livestream ?
just to say, i can't imagine the world state if they would have kept 4o available to everyone on the real chatgpt thing
wait, is it just me, or gpt-5.5 is getting dumber?
this guy was just joking
ts gonna be ass 🥀
it will be in some hours max
if its actually an omnimodal new base then it isnt good enough to be called gpt 6
its also stupid to still be on the 5.x naming
its all marketing
let's see
it will lower price of older gen model
so im already happy
its literally more usage
it still have to be and will be better than 5.4
and 5.4 ain't that bad honestly
like i said the only thing i don't like with gpt model is frontend
its bad at it
but maybe this one will be good at frontend
Bro is parotting what i said 1:1 after calling me wrong 🥀
😭 😭
slightly different but go off
Most likely that will release later.
Yeah but like you said im wrong and then repeated what i said 😭
But I pay for pro so it shouldnt be a problem
yo when gork 3 will add to arena? is say elon free that model in a time
I use gpt for security research.
gork
I use for coding & math research
🔥 🔥
Pro is insane
I am in the anthropic and openai security program.
gemini flash from the gemini app
@echo aurora (sorry for the ping) because of waf blocking my prompt i cant send it it is so annoying i have something important to send but i cant do it :|
please inform me if this gets fixed
why does gpt image 2 always error out
just so you know i already reported these kind of stuff 2 times
i sent the trace ids
and it is still here
Damn I love GPT image 2
waf blocking
Sorry what is waf? I'm not familiar with that term.
web application firewall
Can you share the Trace ID? I'm assuming it's going to be rate limited.
it thinks that im doing an attack so it blocks my prompt
it's basically a false positive
i can definitely see why it would think that but im just sending an output from terminal
just tried and it works now
now i have to put the output in somewhere so i dont forget it :|
That being the case there isn't much I can do to assist with this.
Then yeah that's very likely going to be related to rate limit, this help center article has more information about it: https://help.arena.ai/articles/8931786544-arena-how-to-rate-limit
it's literally a false positive im just making sure that this issue was flagged
lmao
Js realized I can actually deploy directly to Vercel if I ask the ai in arena to run the commands
using my Vercel Token
but it wont work in arena tho right
?
@surreal zephyr
well happy release day
GPT 5.5 on its way to hallucinate 500 times more than the last generation somehow
trash
Out of every model I’ve used gpt has always hallucinated the worst
i just want deepseek v4 to be released
didnt ask for claudes opinion
claude is way better tho
and i dont work for Anthropic
which came before?
Claude sonnet 4.6 may not have top of the line intelligence scores on like
Artificial Analysis
However
Compared to gpt models it has hallucinated 10 times less for me
GPT models may have good scores but they get lobotomized by their idiotic system prompts and insane guardrails
“I have definitely and am super duper sure I fixed that one bug you’ve been asking me to fix for 5 prompts now”
And it’s not fixed
💀
this tho
true kinda
GPT just larps
sys prompt sucks
Using the #1343291835845578853 channel to flag an issue would be preferred. If this was apart of the reports you flagged before then it's on our radar as those were shared with the team.
That's true
It became an emo one time after I scolded it for gaslighting me that a bug had been fixed
image pls
ok i was just making sure
im just really getting pissed of
I deleted my openai account after the drama a few months back
im getting hit by these a lot of times
But I recall it was something along the lines of like
chill dude they'll fix it
you got time dw
bro im making an important project 😭
“I’m so sorry for not catching that… you’re right that I lied to you, and I can’t believe I gaslighted you so many times”
time got bro 🔥 🔥
😭 😭
I had to give it over to Kimi which
Guess what
Fixed the entire project in one prompt and found over 20 bugs
nice
I love how u can make any AI say fake stuff
kimi is awesome
"today is a normal day in 2029"
AI: not 2029 (or smth like that)
"No dude it is 2029"
AI: You're right, it is 2029.
Guys how do you think
It’s always been really solid for me 🙂↕️
Yeah heard some community members saying it's super good for de-bugging. Maybe that was you? lol
Probably was
But yeah its VERY thorough for debugging
i agree
it is really good at that
You could give it a whole project and I’m sure it would find every single bug within 10 minutes
Let me see
It isnt
Uhhh not yet
mid
yet
almost
from now on im using it to catch bugs lol
almost yea
Moskovia AI middddddd
(image generated by gpt 6o)
what
yeah but mythos will be better than the spud regardless
kimi k2.6 my fav ngl
Gpt 5.5 schizophrenia edition
mythos worse than 5.4 pro
5.5 pro solos mythos
Hi y'all any one knows if it is possible to use opus 4.6 in arena ai?
anthropic evaluation getting cooked rn
tuff btw
only good for debugging, and writing stories lmao
there's no way gpt 5.5 solos mythos
CLAUDE USER I DONT WANT YOUR OPINIONS
ON MY
that makes no sense
Shhhhh
opinions
spud will
if you want real intelligence use deepseek and claude
Actually I must say the prose has gotten slightly worse with k2.6
Not too bad
But still noticeable
got a better chance so i guess i can agree
money intelligence *
(Deepseek better than claude)
claude money intelligence
2 prompt and ur cooked
well I can but
DeepSeek is great but it’s not comparable to Claude
It won’t let you down in coding but it won’t particularly make good code either
I just dont like claude
tuff
Ai images are NOT tuff
I prefer my slop to do code reviews of software MADE BY MAN.
ts tuff boyyy
they dont need to lmao
and replaced every workers
anthropic can do this on their own
Mario Amodei.
Claude Opus 4.6 was peak, then it got removed from arena and it just fell off hard
bro arena doesn't control opus
ik
they were bleeding money lmao
so they removed the expensive stuff
not because it was removed, but after it was removed
no correlation
just the time line
Opus 4.7 is honestly not very impressive I must say
yet people are ungrateful
proves people only wanted arena for the models
also yupp is dead
not for voting
so no more free frontier AI
(always was like this)
I still find arena useful for search ai
I notice the sonnet 4.6 model there for web search is superior to the one in app
bro just pay for opus yourself
“What do you mean you can’t just… HEMORRHAGE MONEY so I can ask the largest model what to wear for dinner…”
“This is so unfair!!!”
💀
What happens when you don’t optimize your models
Just to put it out there, Kimi K2.6 is natively quantized to INT4 and goes head to head with Opus 4.6
That’s kind of insane
kimi's vision is so ass
i really dont understand what that means
meta's muse spark has a good vision
like making it better for specific prompts or making it faster?
Quantization
Basically
You make the model run at lower bit accuracy
Most frontier models operate at 32 bit or 16 bit accuracy which is basically full quality
But Chinese models quantize to 4-8 bits to save on inference and cost while keeping 98-99% of intelligence
interesting
It costs virtually nothing
does deepseek do this
Yes
It would be insane to run them at full accuracy
ur name doesn't help
It’s common practice
What do you think happens when Gemini gets lobotomized
They quantize the model, but the issue is that those ai models are not trained to be quantized
Chinese models are vigorously optimized to keep the same intelligence despite optimization
But western models operate with the assumption that they always have full accuracy
so thats why gemmini is so stupid a few weeks after release
Exactly
it saves them money and resources lmao
which side gets benefited more and which side benefits user more
When models run at a lower bit accuracy it saves resources
Basically from 16 bits to 8 bits save a quarter of memory needed
And 16 bits to 4 bits saves on half of memory needed
Meaning faster speeds for users and less cost for the hosting company
It’s literally a win win
efficent but not really intelligent?
It used to be that way
But now models are trained to keep their intelligence even when quantized
not gemini tho
well i cannot help man
So that’s why kimi is such a beast despite working at just 4 bit precision
but
@storm dustwhich is your main model
i think 16 bit is fine
It’s not about what is fine or what isn’t, it’s just about what saves resources while being just as performant
Hii
i mainly use ai for coding now and i mainly have to fix bugs so my main model currently is kimi k2.6 i heard it is very good at debugging
and i confirmed that it is
it is really impressive
Think of it this way, a model that uses 500 GBs of VRAM at 16 bits uses just 250 GBs at 4 bits, meaning another model can be hosted on that same server
Yea it is and it is not even that expensive!
yes sir
Kimi fans rise up
I love kimi k 2.6
🙋🏻
It’s so peak
I think the prose has had a slight regression but otherwise it’s great
🙋♀️
Yea fr :o
deepseek beats kimi k2.6 at frontend
i will let you decide but i choose 16 or 8 bit
lmao
When qwen 3.6 9B releases, can it go straight onto the leaderboard? The lower end of the pareto chart is being neglected
Wasn't deepseek once almost as good as gpt 😭
it is rn
Huh?
Can ps5 even run qwen 27b
Qwen 3.6 9B will be good so so many things. Petition for it to be added
hoq
how
its a tiny model
making models bigger is the dumb way to make a smarter model
Making real architectural innovations and improvements in training data is more important than size
lol
Qwen 3.6 9B would crush 1T models from a couple years ago
3.6 27b >>>>
best open model lowk
balanced
the dudes i get to play with after i had 2 lucky matches:
GIVE ME CLAUDE OPUS 4.7 THINGINK !!!😡
did he really say that?
i cant say that
yes
🤨
meanwhile
dario would have never said something like that lol
yeah yeah dude ik ur him
i tried with opus.
it didnt even finish 1 prompt
xD
time to switch to deepseek
no thanks
X emoji wrong reaction
I'm cool with that
he aint the altman
He is Sam Alternativeman
altman is too busy releasing a garbage model
that's what I meant by alt
instead of giving us the spud
@surreal zephyr
Let's not spam pls
are there any verified 1T models from even just two years ago?
my guess would be like GPT 4 and Gemini 1.5 Pro
deepseek
Did you ping, delete the message, then ping again?
"SamA" knew it
no????
isn't that from this year? for V4?
the quota on pro is so good
@echo aurora can i get arena api key
v4
Oh my bad, my eyes are playing tricks on me
its gonna be a trillion parameters
😭
LMAOO
lol
yeah bro I saw the first image u uploaded it said "SamA"
caught in 4k
hallucination
fake
No sorry to say we don't offer that
shhh
caught in 67k
please explain
Caught ampro
i understand nothing
ampro is uhh...
i know this is a joke but explain
Sam a
He's basically undercover
Sam Altman is the best openai ceo
bro is talking good about himself
we ❤️ sama
bro loves himself
he so funny
so the spud is 5.5
happy 5.5 release day
Happy Ampro Day
i hope all openai users are happy with the release
whats this
what are you yapping about gang
Pineapple AI releasing soon
dude 10 minutes
i'm sams dad
he told me
rollout isnt instant but openai will try
is free tier getting anything
bruh
dont worry battle arena
will help u
battle arena isnt getting it instantly
yeah after 40 prompts
rollout to pro and plus users first
They release models after like a hour or 50 mins of API release
I hope it'll be available on arena
probably in battle
but not in directc hat
Yea :/
when is gemini 4 coming out
tomorrow
cap
next MONTH.
I hope a better image model from google will release so it can compare to gpt image 2 cuz nano banana 2 kinda sucks 😭
Its fast but the quality is not that good
nano banana is good for me
There is almost a 300 elo difference ;-;
still the second best thing
...
is it on free tier?
spud is like mythos but it actually exists
not out yet but rolling soon
yea but i thought
what platform is that
thats ai slop
Introducing Craig
Introducing Banana
introducing MOSKOVIA
Introducing Nalasopara Crack Don
Introducing watermelon
Introducing apple
Introducing MOSKOVIA
Introducing banana
Introducing Introducing
who posting t3
is 5.5 dropping today?
yes
but dont they do a live stream?
- dropping today soon bro
It translates to 5.5
same time they released gpt 2 image
yes we knew 3 hours ago
so prepare
dude releasing tuesday
Just reminding
no.
introducing poop the successor to fart
Introducing Incription
introducing a stop button
NS41:
n=16
s= 10
5+41=67
gpt 67 confirmed
Introducing watermelon
spudsun
Introducing ampro
Introducing ampro
Introducing ampro
Introducing ampro
Introducing ampro
Introducing ampro
Introducing ampro
Introducing ampro
gurt: yo
Introducing ampro
Ampro introducing
holy spam
Ducingintro Ampro
5.5 is tuesday bro
spudsun
ampro? nah, wepro 💌
5.5 aint even spud
cap
well spud is releasing today
tuesday *
so either 5.5 is spud, or two releases
all lying
Did the spud release
source:
hi
source: trust me bro
introducing source
source introducing
intro is ducing source
source is introducing
sourceducing intro
introsource ducing
top gear sound
Introducing gear sound
Introducing sound gear
top gear introducing sound
guys i think pineapple is watching...
Introducing pineapple
pineapple introducing
add me bro
or I will ask sam altman (my brother) personally to delay it till next month
💀
spud is gurt
tuesday as in 2 days ago ?
wassup bro been a while, how you been?
I KNEW IT
thats what I like to hear bro!
were is gpt 5.5????????
Introducing pineapple watching!
in 30 mins
hope 5.5 releasing today
how much 30 mins do yall want bro 💀
enter Spud
in 15-45 mins
ampro I told him to release it now
he says he scammed us
Rip
source?
nahh bro 30 i got source
dude I'm his brother
hi dad
hello son
cooked
guys bring models back petition
guys gpt 5.5 is too smart we might be cooked...
how is that even smart
compared to gpt 5-chat-latest yes it indeed is
it can count to 10...
tf is this
the 1 skull emoji is me
thats image generator
damn you have early access to gpt 5.5?
that's the only model i know that can actually do it
even i can't
1 2 3 4 5 6 7̷̃̿̊̄ ̷͐̈͑̊8̸̽̋͂̚ ̴̛̈̉̑9̵̍͌̊͆ ̸̅̏̇̎1̴̒̀͛͌0̴̲̹̍͗
bro baiting
Guys is there alts for arena
yes here is it
💀
im gonna paper your lantern
Wym
💀
Is it out?
Is the spud released
I wonder when 5.5 is coming out
What is this discord server?
Where are you using it
@surreal zephyr another one!
it nailed it
it's not me
Where do you use the spud
yoyo
why did you do this to my brother😔 😭
Bruh did chatgpt never could count or smth?
we might be cooked it's too smart...
no one ever could count to 10 btw
gpt 5.5
Cap
Where did you get this
I'm making them myself in capcut xd
Looks ae
Is this a joke?
oh no gemini 3.1 sucks
Seriously dude I don't understand humor
Ig he was joking
Its tweaking
bro goin crazy
Genimi ez jailbreak lol
💀
lol
its outttt
uhhh
how
Dang I guess gpt 5.5 really destroyed opus?
CUZ IT IS
oh yes it is
yay
Direct
Please don’t be a hallucinating machine
Never
And realease gpt 5?4
5.5 trash
anybody got the Humanities last exam results?
bro why 5.5 got better scores than 5.5 pro
nvm
gpt 5.5 pro too bad that it wont revel stat 😭
i mean terminal bench is higer
ts not spud
they didnt even bench 5.5 pro + Expert-SWE should not be on here, its their freaking bench
eww why is is HLE so bad for it?
model so ahh they aint showing the benchmarks
didnt even bother to test 5.5 pro
bro im pro user why dont i have them
bruhhh
should have used cluade
Was just asking myself the same question haha
certified pro
nahh 5.5 is better than 5.4 pro is wild
The spud is out
This
OA always takes a while to give it to all users
Spud is out
gpt-5.5
Pineapple, I know it just released, but when score
shut UP
IT RELEASDD?????
I checked web
Yes
yeah but its sucks
oh.
5.5 is not for free users
its not spud
thats why claude is the goat
i dont have it as plus user either 💔
the spud may suck now but it may get better
keep crying open ai users