As openai said
#general
honest verge
vital lake
Thats not what they focused on ever
royal sail
Does overall mean as an AI that can answer questions? Then yes.
distant idol
where is the "im disappointed option"
vital lake
Thats what codex is
royal sail
I feel like most models are good overall
honest verge
vital lake
No there is a difference
vital lake
Marketing BS
honest verge
Not even one word about writing
distant idol
they js doing too much atp
honest verge
Looks like openai hasn't even planned this model for writing
neon idol
How do I find out which model I’m using in ChatGPT?
distant idol
How do I find out which model I’m using in ChatGPT?
ask it
frosty lava
Looks like openai hasn't even planned this model for writing
they did and said it in release note but yeah they didn't said it in the title
for some reason
distant idol
He says he is using got 5.3 but idk
then he is using 5.3
wicked talon
then he is using 5.3
Gpt 5.3?
honest verge
Idk why but new gpt models declines to use three js
wicked talon
I'm on gpt 3.5
distant idol
folk
frosty lava
Even if you asked 3d
yes your right i saw it too on arena
proud bobcat
Even if you asked 3d
ive also noticed that the new gpt models only use the app.tsx file
like
why
frosty lava
they made the model token efficient so maybe its due to this..
loud verge
Do you need a bot for this?
distant idol
Do you need a bot for this?
oi butcher did u see the trailer
distant idol
he just typing shi 😭
proud bobcat
It can't use three js?...
"it feels challenging" translates to "im lazy as hell and way too token efficient to do ts"
frosty lava
token efficiency is so bad everytime
golden ocean
gpt 5.4 not new sota?
golden ocean
YOUCH
loud verge
gpt 5.4 not new sota?
It likely isn't worth the price increase.
distant idol
https://www.kkinstagram.com/reel/DVgxUjxk1uZ
nah the scene between homewrecker and solidermale was insane
golden ocean
is bad writing?
inner relic
nope
frosty lava
can i know why gpt 5.4 on the codex app doesn't have 1m context ?
loud verge
nah the scene between homewrecker and solidermale was insane
Daddy issues 💀
proud bobcat
yeah i hate this model
loud verge
Though
proud bobcat
this is awful
loud verge
Gpt Plus will still be cheaper than Claude.
distant idol
opus better
yes
vital lake
Its not that bad tho, its not like Opus level
proud bobcat
opus smashes gpt
frosty lava
It also sucks at code
its bad at front end
proud bobcat
Its not that bad tho, its not like Opus level
yeah but you get lower quality code for the same price
vital lake
Wdym Opus is $25 output?
honest verge
Gpt 5.4 is so bad for three js like you have to use a 3d engine so it actually will do something 3d but it just can't use three js for some reason
proud bobcat
wait what?
honest verge
I asked it 6 times
frosty lava
i tried gpt 5.3 codex and opus 4.6 and i can tell opus 4.6 is only better at front end
loud verge
Wdym Opus is $25 output?
These glazers High af.
honest verge
It still couldn't use it
frosty lava
i tried gpt 5.3 codex and opus 4.6 and i can tell opus 4.6 is only better at fro...
for good looking
vital lake
These glazers High af.
Wdym bruh...
wraith wren
its bad at front end
Tbh I tested it out rn and it's kinda bad at back end too.
Claude managed to get it first try while gpt 5.4 failed. But.. Meh
proud bobcat
Gpt Plus will still be cheaper than Claude.
delusion
vital lake
frosty lava
Tbh I tested it out rn and it's kinda bad at back end too.
Claude managed to ge...
i don't know i test with codex app
proud bobcat
claude still won at a 6% diff
honest verge
Gpt 5.4 humbled
frosty lava
that does not make any truth in term of capabilities for back end
honest verge
It couldn't beat Claude
frosty lava
especially considering the 6% diff
distant idol
opus da goat
proud bobcat
Tfym delusion? Claude rate limits are trash.
i can use it everyday with quite generous message limits, more than gpt may i add
loud verge
that does not make any truth in term of capabilities for back end
My point was that codex has nearly caught up to Claude with v5.3
proud bobcat
and i dont have paid tier
honest verge
i can use it everyday with quite generous message limits, more than gpt may i ad...
Also Claude is not lazy
It will do whatever you ask
proud bobcat
My point was that codex has nearly caught up to Claude with v5.3
ehhh....
frosty lava
My point was that codex has nearly caught up to Claude with v5.3
Yes, its really just front end that's bad
proud bobcat
5.3 is great for backend work but for frontend its miserable
frosty lava
always been like that for gpt
loud verge
5.2 codex was worse than sonnet though.
frosty lava
5.3 is great for backend work but for frontend its miserable
yes so don't say its bad at coding
it have bad taste
proud bobcat
oh no i mean 5.4 is bad
frosty lava
but its not bad at coding
frosty lava
of course cause its a codex version
proud bobcat
codex is finetuned for speed while sacrificing knowledge
it is not fine tuned specifically for coding
loud verge
frosty lava
idk about this honestly in codex its doing very good job
distant idol
gpt 5.4 is fast ngl
frosty lava
5.4 is made with "token efficiency" so yes it'll be lazy
loud verge
gpt 5.4 is fast ngl
I heard other people say this too.
proud bobcat
its fast but lazy
frosty lava
yes token efficiency = lazyness at max
loud verge
yes token efficiency = lazyness at max
Opus does this too but with scientific research.
proud bobcat
https://swe-rebench.com/
48 problems out of the whole benchmark is not a basis
distant idol
I heard other people say this too.
its fast butcher
proud bobcat
lemme see something actually
distant idol
5.4 feels like a small update
5.3.5
proud bobcat
why does it constantly want to use the app.tsx
stop shoving everything into one file
honest verge
stop shoving everything into one file
It's like ancient lvl
distant idol
stop shoving everything into one file
he is nervous dont mind him
honest verge
Old ais also put everything into one file
frosty lava
again its always due to this stupid token efficiency
proud bobcat
real
honest verge
So 5.4 is basically old ai
proud bobcat
oh wow finally
frosty lava
they should make a separate model not token efficient
proud bobcat
its not gonna only use one file
honest verge
oh wow finally
FINALLY
proud bobcat
claude opus is still faster though
honest verge
How did he use 3d?
proud bobcat
its basically caught up after getting an error
distant idol
oh wow finally
1m20s
proud bobcat
How did he use 3d?
magic
lmao opus is doing this in half the time
thats insane
hmmmmmm
distant idol
lmao opus is doing this in half the time
ts js a downgrade atp
proud bobcat
now gpt is just creating new bugs
LOL ITS REMAKING THE WHOLE FILE
meanwhile claude just finished the tank suspension engine
awww there was an error
distant idol
awww there was an error
loud verge
Pookie wants a hug
proud bobcat
its honest
frosty lava
There's a difference between real opinion based on facts and hating for free + faking things
anyway i can't change people.
distant idol
and its broken
act surprised
honest verge
And all over again it puts everything into one file
loud verge
Are y'all good at vibecoding?
I was trying to make a web app for a task but I'm not sure if it's possible.
honest verge
WHAT THE HELL IS THIS IT'S GPT 5.4
distant idol
And all over again it puts everything into one file
four mins 😭
loud verge
I kept failing with the ai.
proud bobcat
thanks gpt!
honest verge
Maybe because gpt didn't install fourth package
Because I expected more
Gpt is so garbage
proud bobcat
gpt did install 4 packages for me
it just ignored the entire physics engine it installed
proud bobcat
thanks gpt!
here
honest verge
It ruined everything
proud bobcat
gemini 3.1 pro has the same exact problem too
everyone saw what claude was doing and tried to copy it
loud verge
Hmm...
honest verge
Claude is so good for 3d and frontend
frosty lava
honestly token efficiency shouldn't be on thinking model, not that much
honest verge
honestly token efficiency shouldn't be on thinking model, not that much
Token efficiency nerfed gpt 5.4 to the ground
distant idol
gemini 3.1 isnt good tbh
proud bobcat
dont worry
honest verge
I won't use gpt anymore
distant idol
I'm already scared for gpt 5.5
dont even get ur hopes up
proud bobcat
gpt 6.7 will finally fix all the issues
loud verge
honestly token efficiency shouldn't be on thinking model, not that much
Maybe high variant would have been better.
frosty lava
If it will have even more token efficiency
its not impossible that they work on it too.
distant idol
gpt 6.7 will finally fix all the issues
📆
frosty lava
and fix this
loud verge
gpt 6.7 will finally fix all the issues
6.96969
loud verge
at this point even deepseek v4 will mog gpt
I want this to happen but highly doubt it.
proud bobcat
deepseek 3.2 was already a substantial upgrade to coding on the same V3 dataset
distant idol
@echo aurora is this guy dead or sum
proud bobcat
why would you ping the mod
honest verge
<@283397944160550928> is this guy dead or sum
@gork is this true?
loud verge
<@283397944160550928> is this guy dead or sum
I last heard him say he's travelling.
distant idol
why would you ping the mod
ive been mentionin him for 3 days
loud verge
why would you ping the mod
That's an admin.
distant idol
@gork is this true?
yes twin ✌
loud verge
Good thing he's not an azzhole.
distant idol
eta pizdets blyat
whoa
proud bobcat
GPT 5.4 IS SO BAD
grass is green
distant idol
GPT 5.4 IS SO BAD
bruh atp they should js remove it
inner relic
Can deepseek v4 release tommorow
distant idol
and rethink ts
proud bobcat
deepseek save us
i will forever be a deepseek fanboy
deepseek has deadass never failed me once
honest verge
bruh atp they should js remove it
It was supposed to revive openai and stop unsubscribing
But it made it worse
distant idol
It was supposed to revive openai and stop unsubscribing
nah the unsubscribing is another story
proud bobcat
it needs to cook in the oven for another month because god knows what they were smoking
distant idol
crazyy
loud verge
deepseek has deadass never failed me once
No way. What did you even use it for?
proud bobcat
No way. What did you even use it for?
math usually
loud verge
Lmfao
proud bobcat
its a KING at math
distant idol
WHAT IS THISSS
wdym bro its the new ohio tank model
proud bobcat
oh my god
oh my god
it doesnt even drive correctly
let me record this goofy ahh
loud verge
Lol. I love these guys.
loud verge
Father .... Is this you...
Mother when
frosty lava
Stop acting like its a new thing, i said it 10 time bro and everyone know gpt suck at front end
proud bobcat
frosty lava
and now with token efficiency its lazy too
yes
but please do not over react like that
loud verge
Meinkraft drift.
distant idol
nephew 😭
proud bobcat
nah what do you mean
proud bobcat
this is agi bro
proud bobcat
what was the prompt
"create a detailed 3d simulation of a tanks suspension with a map, driving controls, accurate physics, and polish. do not cheap out on code, think to the maximum and no shortcuts."
albeit not the best prompt but come on
come on dude
claude opus worked with less and produced 5 times better results
loud verge
"create a detailed 3d simulation of a tanks suspension with a map, driving contr...
Can you try this on 5.2 high and see if it's better or worse?
honest verge
It will crush openai
hardy merlin
hhaha wtf
honest verge
claude opus worked with less and produced 5 times better results
Opus is so good
Because it actually does best it can
frosty lava
i love ai that are good at front end too
but all i care about is the actual capabilities
not only if it can make something look good
but if it can work with real project
glazing is as bad as free hate, when its too much its too much
loud verge
Lol.
loud verge
glazing is as bad as free hate, when its too much its too much
5.5 codex will beat anthropic. Trust.
frosty lava
5.3 codex is really solid already, but if they start adding token efficiency and making it lazy..
honest verge
5.3 codex is really solid already, but if they start adding token efficiency and...
I think they should remove 5.4
And actually make it better
At least without token efficiency
frosty lava
what they should remove in my opinion is the token efficiency
honest verge
Or less
frosty lava
bad result are mostly due to this
loud verge
At least without token efficiency
Adding high variant would do it ig.
Medium is mid. Even in 5.2
frosty lava
i don't mind if the ai take 2x more time to give me a real good result rather than 2x faster but i need to give it 6 more prompt to a good result
loud verge
Okay. What should I say?
honest verge
What's better gpt 3.5 turbo or Claude 2.1?
loud verge
What's better gpt 3.5 turbo or Claude 2.1?
Never heard of Claude 2.1
honest verge
Never heard of Claude 2.1
It actually existed
loud verge
Might have. I was free user. I used bing chat.
distant idol
Might have. I was free user. I used bing chat.
the good ol days
frosty lava
in my opinion its not even a question anymore if ai are better than human at coding, it do the same work as human for much less time and cost
now in general intelligence its not at human level that's not the same thing
distant idol
in my opinion its not even a question anymore if ai are better than human at cod...
yess
loud verge
- Hmm... How did I even miss it.
loud verge
the good ol days
Everything felt same until O series models arrived.
honest verge
It already had 200k context
While gpt 5.3 codex had 256k
Crazy
It also had first tools
loud verge
It also had first tools
I thought perplexity did it first.
vital lake
frosty lava
context doesn't equal intelligence, 256k is not bad and we do not really need more than that, the fact that they increased it to 1m is very great but not necessary
loud verge
There's this thing called context accuracy.
Gemini has fat context size but low context accuracy.
Makes up or forgets stuff.
honest verge
Gemini has fat context size but low context accuracy.
Llama 4 scout has 10m context
honest verge
But after 100k context
Model accuracy starts decreasing
And after 1m
It's almost nothing
distant idol
It's almost nothing
yes
honest verge
Can't imagine what 10m feels like
loud verge
Okay my Claude had a seizure.
frosty lava
Okay my Claude had a seizure.
that's due to arena limit of 10 minute
many problem with claude opus 4.6 thinking
on arena
loud verge
that's due to arena limit of 10 minute
I used one in past 48 hours.
This was it.
Pretty sure the limit is like 3 or 4 every 30 minutes.
frosty lava
we can't use claude opus 4.6 properly on arena
frosty lava
Pretty sure the limit is like 3 or 4 every 30 minutes.
im talking about another limit
the thinking limit or whatever they call it
loud verge
im talking about another limit
Tokens?
loud verge
the thinking limit or whatever they call it
Weirdly enough, the web search variant doesn't have this limit I believe.
loud verge
Or it's limit is clearly longer.
frosty lava
on thinking
distant idol
its confirmed by pineapple the admin
what does he even do
frosty lava
what does he even do
what do you mean
frosty lava
yes and he work for arena on the website too it seems and take feedback from people
yes he take feedback and have the info
loud verge
Not sure if he has evil plans in mod chat, lol.
distant idol
yes and he work for arena on the website too it seems and take feedback from peo...
he appears one time a day
and then dips away
frosty lava
he appears one time a day
i think right now he is traveling but i see him very active last few weeks
distant idol
i think right now he is traveling but i see him very active last few weeks
i need to kirkfiy him
loud verge
Opus might fail again so next time I'll proceed with 5.2 high alone.
slim gorge
for gpt 5.4 is the 1m token context window only available in the api?
gentle vector
I need it on descord can i?
distant idol
I need it on descord can i?
it got removed
loud verge
for gpt 5.4 is the 1m token context window only available in the api?
I believe so.
gentle vector
it got removed
Thank you
candid storm
Guys you can still make 7% on polymarket by betting against gpt 5.4
I already put al my cash in it
distant idol
claude been there for ages
candid storm
Its free money
slim gorge
hmm idk it seems gpt just keeps getting worse and worse with new versions lmao
slim gorge
ah i see
distant idol
yea
distant idol
there is 2 rs in apple trust me
honest verge
feels like we're back to 2020
Cursed emoji
frosty lava
5.4 doesnt have non-reasoning i think
it have this non reasoning they said on the release notes
slim gorge
then why release 5.3 at all if they gonna release 5.4 a few days after
distant idol
Cursed emoji
is it crying or mad asf
loud verge
Cursed emoji
slim gorge
honest verge
distant idol
Tears don't fall 2 ways from one eye.
he is a cyclop dude
loud verge
Okay. Opus had another seizure.
honest verge
what I'm doing with my life...
distant idol
Okay. Opus had another seizure.
ur opus gonna have a brain cancer
distant idol
what I'm doing with my life...
what even is this expression
honest verge
loud verge
I guess you are new to this.
quasi atlas
@gentle vector Note that Video Arena has been removed from the server. More information can be found in this #announcements
loud verge
The chicken talks.
loud verge
Can you share what your opus output was like for comparison?
proud bobcat
Can you share what your opus output was like for comparison?
Hold
To be fair this was sonnet 4.6
But the output was generally similar
gleaming roost
🔥
undone geyser
gpt 5.4 not generating text
gleaming roost
on codex app ?
needs to be configured
undone geyser
ok changing to opus or 3.1 i guess
frosty lava
needs to be configured
how to do that
gleaming roost
edit the config.toml file
echo aurora
<@283397944160550928> is this guy dead or sum
Not yet, just traveling 
I'll be back next week though
frosty lava
edit the config.toml file
i put model_context_window=1M in features ?
gleaming roost
i put model_context_window=1M in features ?
echo aurora
Cursed emoji
Thanks, I hate it
frosty lava
thanks
vital lake
5.4 isnt working?
Or is it just me?
echo aurora
5.4 isnt working?
Or is it just me?
It's likely seeing a lot of use right now. Which tends to effect the rate limit resulting in an error for some sorry to say.
vital lake
Oh okay
stray aspen
fierce kelp
What did you expect?
fierce kelp
And that is what we got lol
hardy merlin
😆
vast fern
how is gpt 5.4
vast fern
what is this
hollow ivy
So this could be the end of the AI bubble, right?
vast fern
So this could be the end of the AI bubble, right?
no wayy
abstract tundra
what is this
gpt 5.4 empty response
hollow ivy
Or will Anthropic keep it from bursting?
hollow ivy
isn't OpenAI like the (BTC-)bitcoin of the AI-world?
wicked talon
5.4
hollow ivy
i mean, it was the first company, who created a working chat-LLM
hollow ivy
Will Ilya publish something this year?
vast fern
whose that
hollow ivy
So, currently, only two labs have a realistic chance to develop AGI first?
(Deepmind and Anthropic)
Or has xAI realistic chances to be first?
wicked talon
So, currently, only two labs have a realistic chance to develop AGI first?
(Deep...
Agi probably won't come out till 2050s
That's if it's possible
hollow ivy
The downfall of OpenAI could cause AIs to become more expensive?
wicked talon
The downfall of OpenAI could cause AIs to become more expensive?
I mean ai is already quite expensive now
Gemini is really the only one that's cheap (for consumers) and not Chinese
wicked talon
and Grok-4.20 has its uses
I've not actually tried that
hollow ivy
when will Deepseek V4 come out?
short sequoia
coarse forum
is gpt-5.4 working with anyone
it gives me this
wicked talon
is gpt-5.4 working with anyone
it gives me this
I'll check arena.ai
tulip kindle
is gpt-5.4 working with anyone
it gives me this
it's gives to all
coarse forum
why
tulip kindle
They'll probably spend another few hours setting it up. That's always the case with a new, large model.
wicked talon
is gpt-5.4 working with anyone
it gives me this
Yes
vital lake
is gpt-5.4 working with anyone
it gives me this
Same
grand raft
is gpt-5.4 working with anyone
it gives me this
same
vital lake
coarse forum
did anyone use it?
slate sluice
Is there a usage limit, and is it free? Do I need to upgrade to the paid version after using it for a while?
wicked talon
Is there a usage limit, and is it free? Do I need to upgrade to the paid version...
There's no paid on arena.ai
I don't think
And yes there is usage limits
Usage is different each model
stray mortar
does gpt 5.4 have any big difference or is it just the usual trashy gpt update
slate sluice
And yes there is usage limits
What are the limits for Claude 4.6 and where can I see my limits?
wicked talon
does gpt 5.4 have any big difference or is it just the usual trashy gpt update
Trashy gpt update
wicked talon
What are the limits for Claude 4.6 and where can I see my limits?
You can't
It just tells you once you used it all
coarse forum
did it answer you?
wicked talon
did it answer you?
5.4?
coarse forum
yea
stray mortar
it only works for me in codex, hasnt rolled out anywhere else for me
wicked talon
it only works for me in codex, hasnt rolled out anywhere else for me
Ik somewhere
5.4 has rolled out too
coarse forum
No 5.4 on arena.ai didn't
where did u use it
slate sluice
You can't
Thanks, bro. So, is there an API available? Do I have to use it from the web interface? Because it would be nice to integrate it with CLI or other things.
wicked talon
where did u use it
Poe but there is extreme limits
wicked talon
Thanks, bro. So, is there an API available? Do I have to use it from the web int...
For arena.ai?
Arena.ai doesn't have apis
Unfortunately
If they did they would probably get sued till openai/Google owned them
slate sluice
ty mate
wicked talon
Yw
slate sluice
@alpine pasture You should make him a moderator @wicked talon
wicked talon
<@563563711000870913> You should make him a moderator <@830934542725545995>
Lol I think becoming a moderator is an official process
Not just handed out
slate sluice
You deserve this You help everyone
wicked talon
when gpt-5.4 works in arena.ai
can sm tell
Give it a few hours for developers to set it up.
Should be the next few hours
Always editing lol
vast fern
@echo aurora why are the responses empty
wicked talon
<@283397944160550928> why are the responses empty
I believe the AI is still getting set up.
vast fern
when gpt-5.4 works in arena.ai
can sm tell
me as well
coarse forum
ok
wicked talon
Give it a few hours and it will probably be good.
thick pawn
Is the website extremely slow for anyone else right now?
coarse forum
Is the website extremely slow for anyone else right now?
yes
wicked talon
Is the website extremely slow for anyone else right now?
Normal if it has a high load or maintenance is going on, which I believe it is.
Personally I'm guessing maintenance.
golden ocean
thick pawn
Normal if it has a high load or maintenance is going on, which I believe it is.
I use the website a lot and it's never been this slow. It's barley usable at the moment
And just got this message
wicked talon
I use the website a lot and it's never been this slow. It's barley usable at the...
Well, maintenance is 99% sure happening because GPT-5.4 isn't working, and probably a lot of people are using it.
wicked talon
And just got this message
I used to get this a lot.
The solution was clearing the cache.
stray mortar
gpt 5.4 seems to have significant improvement in ROBLOC
wicked talon
gpt 5.4 seems to have significant improvement in ROBLOC
GPT-5.4 is a big flop.
stray mortar
idk, works well for me
thick pawn
Alright, I'll wait on using the site for a good while then
stray mortar
GPT-5.4 is a big flop.
well specifically codex to be exact
vivid oasis
I use the website a lot and it's never been this slow. It's barley usable at the...
fr
wicked talon
well specifically codex to be exact
Ah.
wicked talon
fr
I'm taking an educated guess and saying it's maintenance combined with high load/demand.
stray mortar
Codex is nice in my opinion.
yeah the normal model seems to be the usual garbage GPT update
wicked talon
yeah the normal model seems to be the usual garbage GPT update
Yes, that's why it's being called a flop.
stray mortar
Yes, that's why it's being called a flop.
i agree then because the average consumer likely doesnt even know abt codex
frosty lava
After testing the model in my use cases, which is coding, modernizing old project, fine tuning rewards gpt 4 is doing even better than gpt 5.3 codex.
wicked talon
Like "set a calendar reminder."
wicked talon
It can.
frosty lava
It can.
no, not once a model went worse than a previous versions
in no companies
cause the model is training it legitimately won't regress, it can stuck at some point but regression ? no
and if it did regress they wouldn't release it
stop spreading fake information
the cost efficiency features is making it lazy not less capable
i agree that they should remove it, but the model itself is more capable
than 5.2 and atleast as good as 5.3 codex on coding task
wicked talon
Hmm, it could I just see it as a waste.
frosty lava
Hmm, it could I just see it as a waste.
And why ?
wicked talon
And why ?
It's a near pointless upgrade
I would rather them spend a little more time on a model that can actually be better to the average consumer.
frosty lava
Why is it a pointless upgrade ? if we compare it to 5.2 on all general purpose task its better
its also better at creative writting and talk less robotic
this model ain't a codex version so its not specifically for it
dire kiln
is gpt 5.4 not working in the arena?
frosty lava
In term of coding capabilities its still atleast as good as 5.3 codex, but yes its not an improvement
But this one is made for general purpose not coding so we can't blame it i think
the real problem is that they implemented a token efficiency features and that make the model lazy and the result worse
lucid geyser
But this one is made for general purpose not coding so we can't blame it i think
No more codex apparently
sleek arrow
is gpt 5.4 not working in the arena?
no
frosty lava
No more codex apparently
what do you mean ?
lucid geyser
what do you mean ?
They said they merged codex with chat
frosty lava
They said they merged codex with chat
Where did they say it ?
whole sundial
<@&1349916362595635286>
wicked talon
This is a £20 request...
lucid geyser
Where did they say it ?
Some tweets
long temple
hello guys any help or more information please
deft spruce
i found some interesting gemini2.5 maxinum input token is 1048576
deft spruce
hello guys any help or more information please
..just wait 35min or use another model...
long temple
what the best model for image Up to this this
deft spruce
i found some interesting gemini2.5 maxinum input token is 1048576
and also i found that gemini 3 maxinum input token is 1081274
deft spruce
what the best model for image Up to this this
gemini-3.1-flash-image-preview
or
gpt-image-1.5-high-fidelity
plush river
Do I have an API for Claude models?
deft spruce
"error": "Bad next turn request in Battles in Direct mode. DIRECT was requested, but expected a BATTLE turn |
...i hate battle turn in direct mods
echo aurora
<@283397944160550928> why are the responses empty
Sorry what do you mean by this?
echo aurora
<@563563711000870913> You should make him a moderator <@830934542725545995>
We may have a community moderator program in the future. If it’s something we do we’d be sure to put out an announcement.
echo aurora
"error": "Bad next turn request in Battles in Direct mode. DIRECT was requested,...
Sorry is this an error message you’re seeing?
deft spruce
Sorry is this an error message you’re seeing?
yup
deft spruce
no....cause i deleted that chat already...
trim maple
How to make video
deft spruce
no....cause i deleted that chat already...
oh wait i found it
"error": "Bad next turn request in Battles in Direct mode. DIRECT was requested, but expected a BATTLE turn | sessionId: 019cc0b1-f208-7c30-9249-4c0778fe5bdd | userMessageId: 019cc0e5-8011-7c07-acd2-759b099df236"
burnt sinew
is arena really slow for anyone else
deft spruce
and also
deft spruce
deft spruce
ctrl shift r maybe
ok i did and that's not of cause of that something went wrong
"error": "Bad next turn request in Battles in Direct mode. DIRECT was requested, but expected a BATTLE turn | sessionId: 019cc0b1-f208-7c30-9249-4c0778fe5bdd | userMessageId: 019cc0e7-b504-7bf8-a331-abfeda7d9725"
this is the reason of something went wrong
well after
deft spruce
Is 5.4 good?
..gpt?
undone saffron
gusty iron
hello
quiet skiff
Are the models taking longer to respond?
deft spruce
i changed the browser(chrome->EDGE) and SAME
"error": "Bad next turn request in Battles in Direct mode. DIRECT was requested, but expected a BATTLE turn | sessionId: 019cc0df-eb6e-760e-b742-83e1904c2785 | userMessageId: 019cc0fd-4729-716e-8ccd-6bb783626f63"
}
thick pawn
Are the models taking longer to respond?
The whole website seems to be running slow currently
humble ether
Yup
brittle hollow
Seems very laggy
agile iris
@echo aurora Happy Timezone!
wicked sage
THERES GPT 5.4 NOW?????
fickle venture
THERES GPT 5.4 NOW?????
Yeah
fickle venture
i changed the browser(chrome->EDGE) and SAME
"error": "Bad next turn request in ...
whole sundial
<@&1349916362595635286>
echo aurora
The whole website seems to be running slow currently
Thanks for letting us know. I’ll flag to the team
undone saffron
guys i found free image generation tool
Free, but..
icy remnant
what to do with something went wrong ?
ocean vortex
what to do with something went wrong ?
wait 30sec and try again. If no success with a few attempts wait 1h+ (wait out arena traffic) and try later
This mostly applies to Google models btw
I don't think arena is getting hit with rate limits nearly as bad for other providers
regal girder
Does anyone know what to do about this? I tried changing browsers and still get the same error 😥
hollow mulch
Does anyone know what to do about this? I tried changing browsers and still get ...
Im dunno im got this issue every time im have start new chat
ocean vortex
Does anyone know what to do about this? I tried changing browsers and still get ...
literally above your message the entire thing is the answer LOL
regal girder
Ah, okay! I thought it was about another variant of this error 😅
obsidian prism
Hi guys. Does anyone know why my chat with claude opus 4.6 isnt working anymore when trying to ask it anything ? Any help would be really helpful, thanks
its been like this for a few days now and I dont know what to do in order to fix it
glacial swan
where chatgpt 5.4-codex???
fickle venture
where chatgpt 5.4-codex???
It hasn't come out yet
glacial swan
why
golden ocean
true
edgy spear
Hi, fellas. I'm Spandan Roy, #kakarozoid. I am an expert in prompt engineering and learning a bit of frontend engineering (not imp. info) and very enthusiastic for AIs.
toxic verge
Get early access while it’s open
raven otter
Take a moment out of your day to listen to peak https://youtu.be/HjAmB6vDGmI?si=PDbOrV0kLwlSml9H
fickle venture
https://www.martini.film/invites
Free?
wintry swift
Donde puedo generar video?
toxic verge
Free?
No, but they give you 50 credits to start off with try it out. Its alright I’m playing with it right now so far kind of confusing.
honest verge
It explains everything
wicked talon
https://www.martini.film/invites
What do I put for organization bruh
placid flame
@wintry swift Note that this is an English only server. Also, Video Arena has been removed from the server. More information can be found in this announcement #announcements message>
golden ocean
toxic verge
It explains everything
Open ai getting a lot of backlash
gleaming roost
I don't know about you guys, but I'm loving GPT-5.4 with 1M context
golden ocean
did openai lose the race
toxic verge
No
golden ocean
is gpt 5.4 new sota at anything
noble nova
ugh this Something went wrong while generating the response. Please try again. error is so annoying for me aswell
i have checked the feedback and bugs forums
did all the things and it still doesnt fix anything
toxic verge
is gpt 5.4 new sota at anything
It’s strengthening where the popular demand is probably based on the statistics of what they see of how it’s being used
misty pond
is gpt 5.4 new sota at anything
what do you mean anything 👀
golden ocean
real
noble nova
if i start a new chat it will just start doing the same thing over and over again quite fast after i start chatting
toxic verge
Rate limits
You need to slow down and working intervals breaking down the chunks , not all in the same message
obsidian prism
if i start a new chat it will just start doing the same thing over and over agai...
for now, the only fix I did is to change the ai model to gemini 3.1
noble nova
hmm
noble nova
You need to slow down and working intervals breaking down the chunks , not all i...
alright
obsidian prism
but still opus 4.6 is the one I need
toxic verge
There is a monthly subscription and then there is the API
Especially if you find yourself using it often.
obsidian prism
isnt claude opus more than 20$ ?
rocky mauve
There is a monthly subscription and then there is the API
No subscription is worth it
toxic verge
No subscription is worth it
That’s a very common point of view from a lot of people.
rocky mauve
That’s a very common point of view from a lot of people.
It’s a fact from my end, I use a service with all the newer models for completely free, sadly can’t NameDrop as I’ll get banned, but it’s out there, don’t waste ur money on subscriptions
misty pond
anyways chatgpt has 2x codex limits until april right?
noble nova
It’s a fact from my end, I use a service with all the newer models for completel...
interesting
misty pond
lol
misty pond
We need 4x limits
lets see
honest verge
Imagine pro subscription 4x
toxic verge
isnt claude opus more than 20$ ?
misty pond
people buying anotheer monitor just to see their 3 codex agents in fast mode
toxic verge
misty pond
ah annual sub pricing.
toxic verge
honest verge
Is Claude max subscription worth it?
toxic verge
ah annual sub pricing.
It’s $20 a month without the annual
17$ if you do end up getting annual you save three dollars a month
misty pond
yup i know. still a good deal
honest verge
Or pro 17$ is enough
toxic verge
I just think so especially if you use it
misty pond
rather than the api for cli coding
toxic verge
And you rely on it and you don’t want any of the headaches that come with errors
wicked talon
Gemini sux
Pro doesn't
toxic verge
No, you’re right. I’m just being picky and spoiled. 😂
honest verge
Gemini 3.1 pro just overthinking
Maybe it's good
But 500 seconds for one task
It's too much
toxic verge
Google over extended
They integrated Gemini into everything from Google search to YouTube to Gmail and everything in between
wicked talon
toxic verge
probably don’t have enough compute they have to cut corners memories is expensive
wicked talon
They integrated Gemini into everything from Google search to YouTube to Gmail an...
True
But that's what AI will be
toxic verge
No doubt
wicked talon
In everything we go in ai will track us
honest verge
Is llama dead?
wicked talon
And we will be able to use AI at the same time.
toxic verge
Is llama dead?
No
honest verge
I'm still waiting for llama 5
toxic verge
I'm still waiting for llama 5
It’s code name is avocado
wicked talon
Wait 18.99 Vs 7.99 lmao
toxic verge
toxic verge
Wait 18.99 Vs 7.99 lmao
Yeah, most people are gonna sign up just because of the ecosystem
wicked talon
https://www.analyticsinsight.net/news/metas-avocado-ai-the-new-model-set-to-riva...
Meta ai is doggy
wicked talon
Yeah, most people are gonna sign up just because of the ecosystem
I got it free so I just use it now lol
Only use Gemini that's it
toxic verge
There are other services if you do some research where people buy shared accounts
wicked talon
There are other services if you do some research where people buy shared account...
Yeah
toxic verge
And you have like five people sharing one account or something like that
I used to use that service often
honest verge
There are other services if you do some research where people buy shared account...
I think Google bans them
wicked talon
I almost did that for chatgpt
honest verge
Just like openai
toxic verge
Yes, some services offer you have to download a special app
It’s a specific kind of browser
wicked talon
Multiple accounts from other side of the world is kinda giveaway
toxic verge
I forget the name of it
wicked talon
It’s a specific kind of browser
Tor?
toxic verge
Antidetect browser
honest verge
But I still can't understand what supergrok heavy is for
Like 300$ for it
But it doesn't worth it
You can buy Google ai ultra for the same price
And get a lot more benefit
toxic verge
It’s cheaper to do api. Only normies paying 300$
Because on a monthly payment plan, it doesn’t matter if you use it one time use it 20 times
At the end of the month, you’re still paying the same with API you were just paying for usage
honest verge
Did they ever update heavy mode?
toxic verge
He has
honest verge
It's only good for search
toxic verge
They’re actually pioneer in a lot of areas that are not really appreciated by the AI community
But appreciated by users
honest verge
But for the frontend it's....
toxic verge
Because it goes against the grain
People look at it as a gimmick
Think of it like YouTube and TikTok
toxic verge
Yeah, but that’s a lot more than some of these other models could say lol
Even Gemini is connected to Google search and it’s still
honest verge
Google search doesn't allow getting information so quickly
After all it's a websites
toxic verge
To be honest with you, I think too many people in the community are so stuck on benchmarks, and all these other industry made terms and propaganda
honest verge
Does benchmarks even matter?
toxic verge
It does to a certain group of people
toxic verge
Mainly for researchers and if you’re selling products and stuff
And possibly even for enthusiast
But the benchmarks and how people use it in real life in the wild is two different worlds
Reddit
honest verge
Because censorship is slowly killing ai
toxic verge
honest verge
That's why openai is so hated
toxic verge
No, that’s just an example if you look at the leaderboard and then you look at something like Reddit where people are more vocal about things that aren’t really benchmarked
They painted a whole different picture of the reality on the ground versus what’s on paper
Although looking at these benchmarks, if you look at the capabilities that are benchmarked I suppose yeah Gemini is number one
honest verge
Then what's the point of benchmarks?
toxic verge
There’s a lot
It’s a way of measurement
It’s the closest thing we have to progress measurement
Good more than anything it’s more than likely for my speculation of popularity contest
wicked talon
Because it's connected to X
Thats a bad thing sometimes
toxic verge
Thats a bad thing sometimes
Well, who decides that? And what’s the metric for judging?
golden ocean
transformer
wicked talon
Well, who decides that? And what’s the metric for judging?
People can post misinformation
wicked talon
TODAY IS MY BIRTHDAY
Happy birthday!!!
high owl
hi
random flax
there is no way to fix ts in my project?
wicked talon
https://youtu.be/3jrGutFAIgo?si=LqtCIEXmSBkCaXwO
Its real
toxic verge
I got a challenge for somebody
To prove my point that AI can’t beat original
Try to generate an image of a chicken or dog in the same kind of aesthetic
It’s always gonna revert back to generic depictions
wicked talon
Bet
toxic verge
wicked talon
EXCUSE ME I DIDNT STOP THIS PROMPT
wicked talon
Mogged
toxic verge
Haha good try.
wicked talon
toxic verge
I guess the only way to do this is the train some Lora’s
wicked talon
Haha good try.
Was it meant to remove the cow?
wicked talon
Let me try asking for aesthetic.
wicked talon
Lemme try
acoustic garden
Why does this error happen so often?
toxic verge
If we knew we would be millionaires
It’s a very common issue that a lot of people seem to have. It’s really hard to say it could be a number of things.
wicked talon
Lmao
copper whale
maybe get the limit of the conversations
toxic verge
Nice
wicked talon
Ima Try one more thing
toxic verge
You see the fat lines? It’s almost like it’s robotic looking.
But if you look at these two videos, I kind of try to minimize the fat lines
It kinda looks like cleaner animation?
Vs something with fat outlines
You see how the movement is like really robotic AI looking vs the top ones?
It’s like 99% of videos I’d u pay attention
toxic verge
I don’t like neither of them. I don’t like the bottom videos because of the outlines
See how that’s doesn’t really have any of those fat outlines
copper whale
Ah i see
copper whale
See how that’s doesn’t really have any of those fat outlines
It's like it was carved out to make it thinner.
🧐
toxic verge
So the image on the right is original the image on the left is how AI likes to do it
Although it’s smoother and has cleaner lines
When you animate it it looks all Janky
copper whale
Is that why AI videos are often easier to detect?
Perhaps AI often has templates for creating videos
🧐
toxic verge
You see the difference?
astral vortex
Is there max generations for nano banana pro
copper whale
You see the difference?
I think it's the camera angle, perhaps?
toxic verge
Perhaps AI often has templates for creating videos
Look at the background
copper whale
Human-made videos and AI videos differ mainly in camera angles, and human-made videos generally look more natural.
copper whale
Look at the background
and the animate
copper whale
Ah, I understand now. Fat lines make them slightly protruding, like they disrupt the harmony in the frame
🧐
toxic verge
Like this is original hand run I think this is from beauty and the beast
And I’ll show you the AI trying to replicate the same thing
But when you use a reference image
Why do you remove that post?
desert abyss
Why do you remove that post?
Note that Video Arena has been removed from the server. More information can be found in this [announcement ](#announcements message)
toxic verge
I know 😭
solar fulcrum
can somebody please explain to me when i try using arena.ai no code is being generated when i'm literally using it for a project that requires code
someone with an iq above 10 please
because this is ridiculous. this doesn't happen when i actually do use the claude or gemini platform.
i'll get:
I am applying the changes now.
First, fixing src/file.ts:
Next, fixing src/file2.ts:
Next, fixing src/file3.ts:
Next, fixing src/file4.ts:
Next, fixing src/file5.ts:
Next, fixing src/file6.ts:
with no code.
and some other padded text around that which just explains some bullsquat about what it's doing in a very simplified dumbed down way with no details on how to actually fix anything
literally useless to me
free or not this is an issue and i'm very curious on how this is even an issue for a model that's quite literally designed and optimised (based on benchmarks & advertising/branding) for coding.
golden ocean
solar fulcrum
it's so absurd
it'll also mention stuff about "tool calls" like create_file
but it doesn't seem to be of any use since it doesn't actually do anything
if this is a pre-existing issue which i'm not going to check because i don't have time to look through previous chats in a server full of 200k+ people. let me know please and thank you.
this is wasting compute. valuable or not.
wicked talon
Ai image generation will get way better once AGI comes out in 2050
primal orbit
Did anyone manage to complete captcha on the website? Straight impossible.
wicked talon
Did anyone manage to complete captcha on the website? Straight impossible.
It's easy
toxic verge
Did anyone manage to complete captcha on the website? Straight impossible.
You need to login and log out and clear your browsing data
wicked talon
..
solar fulcrum
the site was fine to use after like 1-2 days
but then it just completely went downhill just because of this issue
captchas are insufferable but so is every other captcha service in the world
that's not a bad quality
toxic verge
I said this is all due respect. It’s cheaper to just pay for the subscription of your favorite model than having to swim against the tide.
solar fulcrum
i'd rather not attach my card to anything
i'd also rather not have to rely on some external third party provider to provide me with a prepaid card to use
solar fulcrum
what does anything i said have to do with what you said
toxic verge
That’s why I use when I pay for them
solar fulcrum
you're not providing or aiding in the resolve of a problem
you're just trying to find a way to sidestep it
toxic verge
Fair enough
solar fulcrum
and not even sidestep. more of a jump. to a completely separate service which provides that model just to then use that model.
after paying for it.
if i don't have to pay for something i won't.
i'm not a "cheapskate" i'm just a person who doesn't like to spend money on stuff if i really just do not have to.
toxic verge
No one said you were no one was accusing you of that
solar fulcrum
i found this site through a github repo
i understand no one was accusign me of anything
i never said someone was
i was simply stating that's not the type of person i am
toxic verge
Fair enough
astral vortex
Is there max generations for nano banana pro