gemini 3 looks good
#general
astral blaze
spare rune
Gemini for the gui and gpt for backend
hoary elbow
Gemini 3.0
zealous sparrow
sure gemini 3 pro takes 18$ for output, but it doesnt make sh like OpenAI
astral blaze
im not trying your game though
devout vault
Gemini for the gui and gpt for backend
nah i dont think so
hoary elbow
Gemini 3.0 sucks when I use it, but how did you get that good of a result?
odd geyser
Yes.
#1445330452180238369
devout vault
Gemini 3.0 sucks when I use it, but how did you get that good of a result?
paid API
spare rune
nah i dont think so
Gemini sucks at coding other than front end (imo
rugged abyss
tbf id say Gemini, its only issue really is delta time I assume. At least i find it visually more appelaing
torn mantle
gemini 3 is unplayable
hoary elbow
So it’s deep think
zealous sparrow
i dont like gemini 3 version
are you being deada-
grave plaza
guys is kat coder in?
hoary elbow
although UI wise gemini 3 was better
Are you using a computer to play it?
devout vault
asura is literally a gpt glazer
rugged abyss
gemini 3 is unplayable
Its just too hard, not unplayable
zealous sparrow
Its just too hard, not unplayable
gemini 3 version is easier than OAI
from my tests
torn mantle
Are you using a computer to play it?
yes
sharp mirage
i think gemini better cuz it looks like game not gpt is looks like a modren page isnt a game
astral blaze
Also gemini 3 is quanted right now and 5.2 is full precision because it just released
(just speculating)
zealous sparrow
Also gemini 3 is quanted right now and 5.2 is full precision because it just rel...
nerfed gemini 3 pro beats a fresh model
hoary elbow
GPT 5.2 looks more like a website other than a game
zealous sparrow
are we deada-
devout vault
GPT been releasing the worst models ever
compact sleet
I have no idea how to attack on the gpt one lol fun game though, both of them
golden ocean
openai is done for
compact sleet
Gemini 3 could attack at least with space
torn mantle
modest prism
Have we finally achieved AGI with gpt-5.2 ?
torn mantle
one thing they've been talking about in this new model is the world knowledge
devout vault
gpt-5.2
torn mantle
it has a better understanding of the world
devout vault
it has a better understanding of the world
bro gemini does too
hoary elbow
No, AGI hasn’t been achieved yet. It would’ve been big news.
hoary elbow
But it’s a good AI
spare rune
gemini 3 pro is literally in preview rn still
Well Gemini 2.5 used to be better in preview
rugged abyss
it has a better understanding of the world
Gemini 3 Has a much better understanding
I think Asura is ragebaiting
hoary elbow
But Gemini three pro is better
astral blaze
it has a better understanding of the world
lol are you serious?
zealous sparrow
google stock is down 2% but just you wait till they learn what gpt 5.2 really is
spare rune
And everyone Gemini keeps making a new stealth model it keeps getting worse
astral blaze
it's terrible at world knowledge
grave plaza
GPT 5.2 IS GOOOOOODDDDD
astral blaze
google stock is down 2% but just you wait till they learn what gpt 5.2 really is
I bought more $GOOG because it dipped
empty stump
i wonder how it does on geobench
modest prism
Should I cancel my Gemini pro?
torn mantle
hoary elbow
Are there any text 5.2 users in here? If so, how good is it with text
astral blaze
Should I cancel my Gemini pro?
yes free up the compute for us
rugged abyss
Should I cancel my Gemini pro?
hell nah
fleet lintel
what is HLE score?
hoary elbow
Do you want to use VEO3?
unborn ocean
hoary elbow
If so, then you can go on flow
zealous sparrow
no it still has the same ass UI as 5.1
empty stump
i need gpt 4.5 size model
unborn ocean
no it still has the same ass UI as 5.1
could still be
fleet lintel
that's a good question... I think its not on a new base... but not sure
cloud zinc
can u access xhigh through app?
hoary elbow
I’m gonna tell 5.2 to stop making website you guys when I try making a game
heavy smelt
Why are the chances of the video arena using veo,kling or hailuo always much higher than other models? Like Is it because these are like the top models so the randomizer prefers them over the others or something? Like I've not seen a ray-3 video generated in the arena for example.
torn mantle
oh
zealous sparrow
no the style is the same ass as before
torn mantle
need to ask it about glassmorphism
spare rune
Just changed the color blue to other colors
cloud zinc
what is HLE score?
34.5%
spare rune
The gui layout is exactly the same
torn mantle
Just changed the color blue to other colors
lol
torn mantle
no the style is the same ass as before
yea
rugged abyss
need to ask it about glassmorphism
Hey chatgpt, create me the most basic UI
cloud zinc
astral blaze
no the style is the same ass as before
It's just the openAI website slop design
neon idol
Is gpt 5.2 flop or slop?
spare rune
Apple should hire gpt
zealous sparrow
Is gpt 5.2 flop or slop?
both
spare rune
Is gpt 5.2 flop or slop?
Slop
rugged abyss
Is gpt 5.2 flop or slop?
both
hoary elbow
I believe human’s last exam Gemini three was better than it before it got quantized
sharp mirage
guys what did i mised ?
spare rune
I think gpt5.2 is thinking infinitely again
zealous sparrow
because they are already quantized to the max
astral blaze
Is gpt 5.2 flop or slop?
It's just GPT 5 2 which is to say that it is GPT 5 with a new slug
neon idol
Worse or better than Gemini 3.0?
astral blaze
With this they have proven themselves to be not capable of beating google
torn mantle
Worse or better than Gemini 3.0?
too early to say
astral blaze
Worse or better than Gemini 3.0?
depends on if you are running benchmarks
or actually using the moedl
rugged abyss
So now that we bashed GPT 5.2
Are you all excited for Gemini 3 Flash when it comes out
sharp mirage
guys try this : https://019b0eb7-e889-77d9-a52b-e7272e8d1863.arena.site/
PROTIP: You gone crsahout
torn mantle
@neon idol https://lmarena.ai/
spare rune
“Game idea: Wind-Up City (Co-op Survival Maintenance)
Premise:
Players are tiny “clockwork caretakers” liv”
torn mantle
try it
modest prism
The knowledge cutoff is august 2025. They've done a completely new pre training
zealous sparrow
guys try this : https://019b0eb7-e889-77d9-a52b-e7272e8d1863.arena.site/
PROTIP...
why is it in arabic
torn mantle
guys try this : https://019b0eb7-e889-77d9-a52b-e7272e8d1863.arena.site/
PROTIP...
it messed up awsd
sharp mirage
why is it in arabic
ask Gpt5,2
astral blaze
let me pull up the one I did with gemini 3
zealous sparrow
ask Gpt5,2
can i have the prompt to test it with googles new flash models
torn mantle
WAIT
spare rune
it messed up awsd
maybe because its arabic
and arabic is spelled backwards
haha
get it
torn mantle
what happens if we use claude frontend prompt for fixing UI slop
zealous sparrow
i alr send it
so what is the prompt
spare rune
Lowk wait for gpt 6.0 xpro
astral blaze
wow very cool GPT 5.2 is this the best you can come up with
fleet lintel
i reluctantly opened my office account to test Gpt 5.2 ...
I am sorry, i am not switching from gemini 3.0 pro
zealous sparrow
wow very cool GPT 5.2 is this the best you can come up with
im sorry this is literally 2d
torn mantle
looks good
rich panther
why does the LMArena website say that there is a Gemini 3 Pro when it is actually a Gemini 1.5?
stray aspen
why does the LMArena website say that there is a Gemini 3 Pro when it is actuall...
It's gemini 3 lol
astral blaze
Where did you do ig
gemini
astral blaze
Gemini app ? Or studio
just the gemini site
stray aspen
why does the LMArena website say that there is a Gemini 3 Pro when it is actuall...
Gemini 1.5 can't even dream about doing these things
sharp mirage
studio
spare rune
AI slop detected
zealous sparrow
Even gemini 2.5 is better
ngl if you are beaten by a model
that is a year old
skill issue
modest prism
Can gpt 5.2 generate image too?
zealous sparrow
close to a year old
bleak lake
How is 5.2?
rich panther
It's gemini 3 lol
then why does he think it's 2024?
unborn ocean
The knowledge cutoff is august 2025. They've done a completely new pre training
not necessarily, they can also just 'continue' the training (they call it continued pre-training)
and that is likely what happened
zealous sparrow
Does it come with the yellow filter?
apparently not
modest prism
It says it can output images in API page
astral blaze
then why does he think it's 2024?
because it's trained up to 2025
zealous sparrow
then why does he think it's 2024?
this model has no grounding with search
zealous sparrow
because it's trained up to 2025
also that gemini 3 pro on LM has no grounding
stray aspen
then why does he think it's 2024?
There's a knowledge cutoff
astral blaze
gpt 5.2 is benchmaxxed
I don't have to bet I know it's benchmaxxed
spare rune
sans.. is that you..
warm zodiac
gpt 5.2 is benchmaxxed
its benchmaxxed for GDPEval, SWE bench, etc
zealous sparrow
sans.. is that you..
what the f is this
spare rune
idk
zealous sparrow
its benchmaxxed for GDPEval, SWE bench, etc
what bout ARCAGI
spare rune
instead of making it by code maybe it wanted to make it by text
proud bobcat
echo aurora
then why does he think it's 2024?
say that there is a Gemini 3 Pro when it is actually a Gemini 1.5
This is likely a hallucination. When models are being trained they're unable to perceive it's own existence, as it doesn't yet exist.
vivid coral
Where is 5.2 gpt in search arena? Pleeeese 🙏🙏
proud bobcat
these benchmarks are so bull
zealous sparrow
> say that there is a Gemini 3 Pro when it is actually a Gemini 1.5
This is like...
Are you sure? I know that one of the models have no grounding, or do you have it on.
proud bobcat
no way it had a leap of over 30% on arc agi in just one submodel update
rich panther
> say that there is a Gemini 3 Pro when it is actually a Gemini 1.5
This is like...
what
astral blaze
The problem with these benchmaxxed models is that they are literally benchmaxxed and is terrible at everything else like actually using the model
zealous sparrow
0% on comedy verified
astral blaze
0% on comedy verified
gemini 3 is actually so good
modest prism
If gpt 5.2 was actually that good they would name it gpt 6 or at least gpt 5.5
zealous sparrow
a huge + of gemini 3 pro is that it knows when you joke
astral blaze
Copium.jpg
zealous sparrow
and knows it a lot
proud bobcat
oh wow... no way...
zealous sparrow
it also knowns references
proud bobcat
its peak...
astral blaze
it also knowns references
gemini 3 is peak
proud bobcat
im not a gemini fan
and also
there is no way this one model update jumped that high
astral blaze
there is no way this one model update jumped that high
Benched
stray aspen
Nice bullmarks
proud bobcat
if gpt 5.2 is peak i will switch to it
zealous sparrow
if gpt 5.2 is peak i will switch to it
boi do i have bad news
unborn ocean
ngl it really feels like they just wrote down the things that gemini 3 was outperforming them at and (tired to fix) them:
world knowledge (supposedly)
front-end (somewhat)
science tasks
competition-level math
GDPval (anthropic outperformed them here)
ARC-agi 2
vision tasks
and some more algorithmic coding benches
echo aurora
Are you sure? I know that one of the models have no grounding, or do you have it...
For fairness in comparing with other models if there is are additional tools we're going to default to not enabling. As not all models could have the same, creating an unfair testing environment.
astral blaze
Gemini 3 is funny
unborn ocean
all things that gemini was leading in compared to openai
stray aspen
Gemini 3 is funny
Gpt is such a killjoy
astral blaze
Gpt is such a killjoy
Yeah I hate it I don't know what happened between 4.5 and 5 but it's such a letdown
I think only 4.5 ever had a chance of beating gemini but we all know what happened there
unborn ocean
I think only 4.5 ever had a chance of beating gemini but we all know what happen...
they likely still have it internally and it was used to train gpt 5
astral blaze
they likely still have it internally and it was used to train gpt 5
I think it's just that 4.5 wasn't pretrained until it's lobotomized
stray aspen
It better be greater than nano banana 2
cloud zinc
It better be greater than nano banana 2
u can already test it on designarena, the model name is hazelnut and chestnut
cloud zinc
It better be greater than nano banana 2
its not better than nano banana pro
unborn ocean
well the benchmarks it had the largest gains in are literally build by openai (gdpeval and mrcr), so it is not that impressive
astral blaze
honestly I don't know how an OAI image model can beat google
especially when google obviously have every photo ever taken on their servers
unborn ocean
but they are clearly focusing on it + plus they had it for longer than the other labs
astral blaze
it's impossible to beat their dataset
cloud zinc
honestly I don't know how an OAI image model can beat google
ya they cant beat
sterile tartan
especially when google obviously have every photo ever taken on their servers
Photos: Hold My Gun
cloud zinc
gemini 3 flash image is coming next week
sharp mirage
@zealous sparrow did you get any good game with this prompt ?
cloud zinc
which will be way faster than any openai models
zealous sparrow
<@872475096743305226> did you get any good game with this prompt ?
suprisingly no, it had to be trained on your prompt ong
astral blaze
IT'S OVER FOR GEMINI 3 OPENAI HAS THE AGI
zealous sparrow
Oaky
nvm
https://019b0ece-d42e-7b99-8e85-06a5e06a3e82.arena.site
ghostfalcon just gave me this
sharp mirage
at least better than mine :/
compact flame
Damn gpt 5.2 is out?
zealous sparrow
adult mode delayed
lmao
compact flame
Is it good?
sick sky
nvm
https://019b0ece-d42e-7b99-8e85-06a5e06a3e82.arena.site
ghostfalcon just gav...
Web game?
zealous sparrow
they can just rerelease gpt-3 before they censored it and call it a day
atp just give us gpt 2 back
astral blaze
atp just give us gpt 2 back
gpt2 is open source
zealous sparrow
gpt2 is open source
right
compact flame
atp just give us gpt 2 back
Did gpt 2 even exist tho?
zealous sparrow
i forgot
zealous sparrow
Did gpt 2 even exist tho?
yes
astral blaze
Did gpt 2 even exist tho?
yes but chatgpt did not exist then so nobody cares
zealous sparrow
yes but chatgpt did not exist then so nobody cares
it didnt sure
but playground autocomplete did
which probably ran gpt 2
astral blaze
it didnt sure
I think before aidungeon nobody cared about gpt
compact flame
So how good is gpt 5.2?
zealous sparrow
So how good is gpt 5.2?
mid as hell
astral blaze
So how good is gpt 5.2?
its good, go use it, unsubscribe from gemini to free up TPUs for me, next
modest prism
Can someone test this prompt? The correct answer is 96:
1,7,18,45,?
compact flame
mid as hell
Rushed tbh
sick sky
They're rushing to the release making the same mistake Google made at the start
astral blaze
Just ping mods
I am being dramatic don't kill my buzz
compact flame
I am being dramatic don't kill my buzz
Okay
gusty helm
god damn it larry
sharp mirage
MMMMMMMMEEEEEEEEEEEOOOOOOOOOOWWWWWW 🐱
gusty helm
no snackies
sharp mirage
alr thats so damatic
modest prism
This is Gemini not gpt
echo aurora
It's a mood please understand
Understood
astral blaze
This is Gemini not gpt
why is it 96 though
astral blaze
This is Gemini not gpt
This is your gpt
compact flame
I checked chatgpt website but there no 5.2
I got scammed by the official chatgpt bruh
proud bobcat
This is your gpt
the high models never work
modest prism
why is it 96 though
Sequence: 1, 7, 18, 45
Differences: 6, 11, 27
Pattern in differences:
6 = 2² + 2
11 = 3² + 2
27 = 5² + 2
Squared numbers (2, 3, 5) are primes. Next prime is 7.
Next difference: 7² + 2 = 51
Next term: 45 + 51 = 96
Answer: 96
proud bobcat
theyre doomed
spare rune
Can someone test this prompt? The correct answer is 96:
1,7,18,45,?
Gemini flash
compact flame
Nvm you can't even send images to gpt 5.2
They be hiding chatgpt failing basic finger test
modest prism
Gemini flash
Didn't expect that. Opus 4.5 thinking and gpt 5.1 and deepseek v3.2 failed
proud bobcat
"sota"
spare rune
Didn't expect that. Opus 4.5 thinking and gpt 5.1 and deepseek v3.2 failed
Maybe they overthink it or something
astral blaze
"sota"
this literally akes no sense
maybe I am just a brainlet
I will go back to benchmarking comedy
compact flame
Where's chatgpt 5.2 on the official website damn there only 5.1
astral blaze
it suddenly got it right maybe they did pull the high precision models back out because 5.2 came out lol
proud bobcat
u didnt use high
yeah cause high doesnt work
compact flame
yeah cause high doesnt work
High said 99
astral blaze
this model is defo benchmaxxed
I hate benchmaxxed models seriously
at least gemini 3 has some genuinely vast world knowledge
cloud zinc
yeah cause high doesnt work
modest prism
"Assuming the 3rd difference stays 11"
Lol math doesn't work based on assuming.
cloud zinc
"Assuming the 3rd difference stays 11"
Lol math doesn't work based on assuming.
its prob cuz lmarena limits thinking token
weary galleon
compact flame
Def worse
glass arch
is 5.2 better than 3 pro for now
compact flame
Very rushed model
astral blaze
We need a sneedbench
modest prism
Can someone test this prompt? The correct answer is 96:
1,7,18,45,?
Also if you wonder this is a test from competitive china math Olympian
compact flame
Atp it's better to buy Gemini ultra instead of gpt pro
astral blaze
Oh wow GPT 5.2 actually got this wrong
proud bobcat
gemini is goated
astral blaze
LMAO
compact flame
Google is like cooking rn
astral blaze
I literally wasn't expecting this, I think even gpt 5 got this right
or 4
or 3 for that matter
compact flame
Or who owns Gemini I forgot
proud bobcat
Or who owns Gemini I forgot
cloud zinc
Also if you wonder this is a test from competitive china math Olympian
i am testing on gpt 5.2-xhigh
compact flame
They spend more money on gemini
astral blaze
Why it’s funny: the sign is comically specific and redundant—if it’s “Sneed’s,” why do you still need “formerly Chuck’s”? It reads like a small-town business trying too hard to keep old customers.
The joke is a subtle, dirty pun based on the rhyming scheme established by the current owner's name.
Current Owner: Sneed sells Feed and Seed.
Former Owner: Chuck (implies he sold F___ and S___).
sour spindle
GPT models always work worse on lmarena than when I use them on app/website for me.
modest prism
Still the goat 🐐
compact flame
they also have better data
Yep
proud bobcat
gemini uses high quality data
astral blaze
it got it wrong?
proud bobcat
gpt is bloated
proud bobcat
large language model
astral blaze
I forgor
astral blaze
large language model
that's it
compact flame
I'd this on yupp or smth
proud bobcat
"Extra High" im DYING
spare rune
It’s so sad it has to be extra high while Gemini flash can do it in 3 seconds
limber crag
What are the benchmarks
astral blaze
gemini remains undefeated
proud bobcat
It’s so sad it has to be extra high while Gemini flash can do it in 3 seconds
literally
this is so sad
cloud zinc
It’s so sad it has to be extra high while Gemini flash can do it in 3 seconds
it took a 1.5 minute
polar niche
Is it good
weary galleon
This guy 👉@sour spindle voted "big improvement" in my poll🤣
astral blaze
sleek crow
compact flame
Gpt x high? That even exists?
proud bobcat
OH MY GOD ITS REAL LMAOOO
modest prism
Maybe it's just a luck
proud bobcat
EXTRA HIGH NEEDED TO SOLVE A MATH PROBLEM THAT FLASH CAN DO IN 3 SECS
spare rune
Maybe it's just a luck
Let me try asking it again
proud bobcat
sam altman lock IN
astral blaze
I hear coping
proud bobcat
it was a new olympian problem from what i know
compact flame
What Extra high means anyways
astral blaze
fax
compact flame
What Extra high means anyways
Or it's just extra high no meaning
sour spindle
This guy 👉<@353540653512392705> voted "big improvement" in my poll🤣
HIPAA VIOLATION
ornate mesa
Gpt 5.2 will be better than gemini 3 if i don't care about thinkung time ?
cloud zinc
Gpt 5.2 will be better than gemini 3 if i don't care about thinkung time ?
yes
it will take longer, but it is better
compact flame
What is even pro medium how to access that model bruh
cloud zinc
What is even pro medium how to access that model bruh
with chatgpt pro
astral blaze
GPT 5.2 PRO MAX ULTRA XL
compact flame
with chatgpt pro
I know but there pro high and medium
There no like switch or smth
void elm
weary galleon
HIPAA VIOLATION
🤣
modest prism
This question does not require insane math knowledge. It just needs a little bit of creativity. LLMs suck at creativity.
rich panther
because it's trained up to 2025
how is this possible if the gemini 3 pro was released in 2025
astral blaze
how is this possible if the gemini 3 pro was released in 2025
up to jan 2025
LLMs don't have self awareness
asking it what version it is will never be right
I am still amazed at how much gemini actually knows
you look at 5.2 and it's just sad
weary galleon
X X X X X HIGH ULTRA PLUS
proud bobcat
5.2 such a buzzkill
proud bobcat
it used to be so fun
astral blaze
but I guess google could just use all of their index as datasets so it's bound to be better
astral blaze
it used to be so fun
4.5 was good
now gemini is the UNDISPUTED KING
depending on what you use it for I guess
But I firmly believe that a large language model should be good at language first and formost
grave plaza
i send a 800 lines Code and it just show me a error 🥲
zealous sparrow
June 2025 meme btw
unborn ocean
ok so guys from my experience the "larger world knowledge" from gpt5.2 is just non existent
astral blaze
ok so guys from my experience the "larger world knowledge" from gpt5.2 is just n...
I already tested this it's a lie
look at gemini 3
proud bobcat
oh its so peak
unborn ocean
like even 2.5 pro smokes it on some (but not all) world knowledge questions
astral blaze
oh its so peak
You can't tell me gemini 3 isn't still the king
astral blaze
but you guys are free to subscribe to gpt-5.2, more compute for me
unborn ocean
gpt5.2 is recent cpt + maybe less agressive quant, but still fundamentally less world knowledge than the others
astral blaze
gpt5.2 is recent cpt + maybe less agressive quant, but still fundamentally less ...
Would be funny if it's just gpt-5 but FP16
proud bobcat
REAL
proud bobcat
5.2 on its way to be the most mid release ever
astral blaze
5.2 on its way to be the most mid release ever
on it's way?
proud bobcat
ITS SO BORING
keen beacon
Gemini is boring
keen beacon
No flavor
astral blaze
actually gemini is more right because asia northeast 1 b is actually in tokyo
keen beacon
No taste
unborn ocean
nah gpt5.1 release was more mid
astral blaze
so gemini 3 have the right stuff
unborn ocean
gpt5.2 is still a pleasant surprise, expected nothing, got something
weary galleon
so gemini 3 have the right stuff
Gemini 3 Pro is not the best!
astral blaze
Gemini 3 Pro is not the best!
maybe
celest orchid
gpt 5.2 is disappointment
celest orchid
weary galleon
but gpt 5.2 is not better
Obviously. Opus is the best.
proud bobcat
opus peak
weary galleon
1,7,18,45,? Right answer is 99 or 96?
astral blaze
Obviously. Opus is the best.
opus is good if you use it for work ig
ivory quest
prompt = "A high-quality 10-second shot of a futuristic robot confidently walking through a neon-lit cyber-city street at night."
astral blaze
but I personally hate anthropic so I am biased
queen veldt
but I personally hate anthropic so I am biased
Why tho
zealous sparrow
prompt = "A high-quality 10-second shot of a futuristic robot confidently walkin...
weary galleon
but I personally hate anthropic so I am biased
I love Anthropic, but I hate xAI.
queen veldt
I think they're the most hardworking company for coding
astral blaze
I think they're the most hardworking company for coding
I mean it's good if you code
queen veldt
They're honest "yes you can use our models but there's limits"
They don't have as much money as google and openai imo
astral blaze
I think sometimes people forget that LLMs are supposed to be a chatbot
weary galleon
I think they're the most hardworking company for coding
They're the best. For psychiatry, it is the best I ever see.
weary galleon
1,7,18,45,? Right answer is 99 or 96?
queen veldt
Well most of money comes from API's so programmers are burning the API and making the company 🤑
astral blaze
Well most of money comes from API's so programmers are burning the API and makin...
It's a money hole and I am not convinced that it would be profitable for a while
weary galleon
I just grade models depending on if they are funny
I grade them on psychiatry diagnosis only.
astral blaze
I grade them on psychiatry diagnosis only.
ok
queen veldt
weary galleon
1,7,18,45,? Right answer is 99 or 96?
👆
cloud zinc
96
queen veldt
Look at this cost of claude opus on MAX mode (max context windows basically using as much code as possible)
keen beacon
astral blaze
Look at this cost of claude opus on MAX mode (max context windows basically usin...
it's not profitable
keen beacon
Yeah just like Suno
cloud zinc
keen beacon
Yup
queen veldt
Hahaha
keen beacon
Just like I said
cloud zinc
microsoft?
astral blaze
https://www.hollywoodreporter.com/business/business-news/disney-google-cease-and...
lol
cloud zinc
u mean google?
queen veldt
Imagine if Netflix signs with Microsoft
astral blaze
if I'm google I'd just kill their SEO reach
cloud zinc
microsoft?
queen veldt
Google*
astral blaze
also google has been literally scanning books and putting them on their servers FOR YEARS
blazing bison
gpt 5.2 is different
astral blaze
they never cared about legal CDs
limber crag
The dirty games have started
fair wadi
when could one expect the text-arena leaderboard to update for 5.2?
keen beacon
Dirty games been here lol
Get ready to pay 4.99 for 10 exclusive video gens lol
Micro transactions
astral blaze
honestly openAI doing this just means that every IP holder will be out for blood for them now
astral blaze
expect Japanese animuu makers knocking on their doors
compact sleet
Personalized movie will be a thing with Gen AI, it's far at the future, but I think... it's pretty much there.
compact sleet
Choose your own actor, scenario, etc etc, preferences and many other things.
astral blaze
The opposite means more gonna join
It's a licensing agreement, OpenAI is paying disney for it
zealous sparrow
https://www.hollywoodreporter.com/business/business-news/disney-google-cease-and...
we can argue this is just disney and OAI working to take google down
astral blaze
if they pay disney, that means that every major IP holder's gonna want that too
zealous sparrow
after they signed a contract
keen beacon
Y’all are crazy
cloud zinc
It's a licensing agreement, OpenAI is paying disney for it
no disney is paying openai
cloud zinc
wait for veo 4
keen beacon
Can’t wait
astral blaze
we can argue this is just disney and OAI working to take google down
It's not gonna happen, like I said google has massive leverage over everyone
keen beacon
I hope it’s good
cloud zinc
veo 3 is 6 month old
zealous sparrow
and veo 3.1 is uh
keen beacon
So is Sora
zealous sparrow
old
astral blaze
no disney is paying openai
They are exchanging services
zealous sparrow
So is Sora
sora 2 isnt 6 months old yet
keen beacon
Yes it is
cloud zinc
we are talking about release date
astral blaze
OpenAI gets IP usage and Disney gets their API calls
cloud zinc
release date is what matters
compact sleet
I remember that Google had that world model teased right?
what is it called again?
cloud zinc
google had veo 3 cooked up way before 6 month, anyone can play that game
rich panther
up to jan 2025
but he thinks that today is 2024
astral blaze
google had veo 3 cooked up way before 6 month, anyone can play that game
if nano banana pro is this good I expect the next veo to be crazy
cloud zinc
I remember that Google had that world model teased right?
genie
compact sleet
ah yeah, Genie
whole swallow
Is gpt 5.2 censored?
astral blaze
also crazy expensive
astral blaze
Is gpt 5.2 censored?
yes
keen beacon
compact sleet
Judging by the uhh trailers of it, it's at Will Smith Spaghetti moment right now iirc
iunno in 3 years in the future
might be the actual winner
zealous sparrow
https://youtu.be/2MFiI3XtdZg?si=aXP_ymuuaDH6Vy99
I mean duh they aint google to whip out 1000+ TPUs
astral blaze
More than 5.1 or less?
about the same which is chasity cage tier
torn mantle
no but honestly gpt 5.2 high has the best translation ever
its so accurate and flows better
cloud zinc
research demo is not same as product
whole swallow
about the same which is chasity cage tier
Oh i see
compact sleet
yeah thats why it's far in the future
whole swallow
Thanks bro
modest prism
Gemini 3 pro is free. Gpt 5.2 high won't be free
compact sleet
but the foundation is laid
fickle venture
Is gpt 5.2 good?
astral blaze
Is gpt 5.2 good?
It's good for sam altman's bottom line
slim gorge
#2 and #6 is crazy for a "SOTA" model 😭
fickle venture
Sora 3 is coming soon 🥀
keen beacon
Idk
zealous sparrow
apparently this is not a lie
what in the f-
whole swallow
apparently this is not a lie
what in the f-
Input output? Wtfff
fickle venture
apparently this is not a lie
what in the f-
Dang that's alot
zealous sparrow
apparently this is not a lie
what in the f-
This makes gemini 3 deepthink look CHEAP bro
whole swallow
More than opus 4.5 is crazy work
keen beacon
Cuz open ai is a premium
fickle venture
OpenAi is just boring
slim gorge
they dont care about cost anymore they're just desperate to not fall behind the competition in benchmarks
keen beacon
Crazy thing is people are willing to pay it
weary galleon
Gemini 3 Pro is completely DESTROYED!!!!🔥🔥🔥
fickle venture
Crazy thing is people are willing to pay it
For a model that can't make a great ui dawg these people afk grinding money
And just waste it
slim gorge
Gemini 3 Pro is completely DESTROYED!!!!🔥🔥🔥
100% in AIME?
celest orchid
umm
weary galleon
100% in AIME?
Yes
full tangle
Is there anyone knowledgeable about music? not just someone who listens, but someone who understands it? I want to send them a specific track and ask what a particular instrument in it is called
slim gorge
just ask ai
fickle venture
How to use this bot to create video
fickle venture
WHAT
queen veldt
I canceled subscription 1 month ago tho
weary galleon
What is AIME? I forgot
Benchmark
queen veldt
Now they're trying to bring me back
keen beacon
<#1397655624103493813>
Thankyou
celest orchid
Very nice, now let's look at the overall bench
)
weary galleon
fickle venture
Dawg these 83% people
fair wadi
Would Google's Deep Research Agent fall under a new entry for the LMArena?
Or is it the same because the underlying model is Gemini 3?
blazing bison
for general use i think that gpt 5.2 is the best model
bcs the instruction following of this model is the best of all
very aligned
compact flame
for general use i think that gpt 5.2 is the best model
What about uh coding
blazing bison
opus still the king
compact flame
Ok
blazing bison
but gpt 5.2 is not behind, it just slow
opus do in 30 seconds what gpt 5.2 takes 5 minutes
queen veldt
Making you inefficient
blazing bison
yeah
queen veldt
I remember i used to waste whole day to create 2 pages on my app
Then i found cursor and it's life-changing with claude
queen veldt
I used to copy the whole code and paste into chatgpt and than copy what it gave me etc
The thinking time + the hussle of it doing the // Rest of the code goes here comments
blazing bison
and in the end gpt 5.2 use so many tokens to solve problems that opus still cheap
cloud zinc
Like where did they find Extra
extra is more reasoning token
compact flame
extra is more reasoning token
Okay ig
tawny canyon
guys should i pay for chatgpt or just use the gpt 5.2 in lmarena?
blazing bison
pay gpt, lmarena is 0 privacy
compact flame
extra is for agentic use
What is agentic stuff I'm kinda new at this ai stuff
blazing bison
What is agentic stuff I'm kinda new at this ai stuff
when you use it in a cli tool or automate something
cloud zinc
What is agentic stuff I'm kinda new at this ai stuff
its not agentic, its just extra reasoning token thats it
blazing bison
its not agentic, its just extra reasoning token thats it
someone didnt read the model paper
compact flame
when you use it in a cli tool or automate something
Okay
blazing bison
😆
cloud zinc
someone didnt read the model paper
source?
tawny canyon
is the gpt 5.2 on lmarena same as the 1 in chatgpt?
compact flame
I wonder why LMarena didn't add this extra though or it's just expensive
compact flame
is the gpt 5.2 on lmarena same as the 1 in chatgpt?
Probably nerfed version of it
cloud zinc
extra-high is better than high, thats all it matters
blazing bison
I wonder why LMarena didn't add this extra though or it's just expensive
bcs you not gonna see the effects of it wiithout tools or summarization
compact flame
bcs you not gonna see the effects of it wiithout tools or summarization
Fair point
golden ocean
is openai cooked
compact flame
I mean OpenAi was the first one to give us AI to save some people from homework
cloud zinc
fiery gull
guys this jump is insane xD
I think this is because from (EXTRA) high
20$ per prompt, I'm so anxious 🙂
blazing bison
I think this is because from (EXTRA) high
the instant model got 14% on arc
cloud zinc
fiery gull
the instant model got 14% on arc
hm? I'll see this later
fiery gull
the instant model got 14% on arc
I'm testing the 5.2 xhigh in real use now >:). I'll see the 52.9% in the pratice
docs and text use
compact flame
Anyways imagine just buying chatgpt pro for purpose to do only homework
neon idol
I'm testing the 5.2 xhigh in real use now >:). I'll see the 52.9% in the pratice
Bro believe me
Gemini 3.0 pro is better
cloud zinc
just trust me bro
blazing bison
Gemini 3.0 pro is better
nop
compact flame
I'm testing the 5.2 xhigh in real use now >:). I'll see the 52.9% in the pratice
Damn you got chatgpt pro?
cloud zinc
Damn you got chatgpt pro?
u cant access xhigh in chatgpt pro
compact flame
u cant access xhigh in chatgpt pro
Then how do you access it
cloud zinc
api
fiery gull
Damn you got chatgpt pro?
no, I just use yupp lol, I never paid for ai in life
compact flame
Is extra high on yupp is technically xhigh or nah
compact flame
no, I just use yupp lol, I never paid for ai in life
How much credits you grinded there bro?
fiery gull
Gemini 3.0 pro is better
I'm testing now at side-side vs opus 4.5 thinking
fiery gull
How much credits you grinded there bro?
I has 120k, I never get this points ;-;
swift oyster
Xhigh loaded, anyone want me to test a prompt?
compact flame
I has 120k, I never get this points ;-;
Damn you're rich as hell bro
I only got 18k
fiery gull
Damn you're rich as hell bro
I work ever day with yupp, and I love to compare AI :D. and vote etc, it's fun for me vote and compare
compact flame
I work ever day with yupp, and I love to compare AI :D. and vote etc, it's fun f...
Oh okay
fiery gull
later I'll see later how I get this point in money
compact flame
What's the highest amount of credits you got per prompt btw
fiery gull
What's the highest amount of credits you got per prompt btw
in just one? I think is 1345, Idk why
compact flame
in just one? I think is 1345, Idk why
Oh same
I think its based on how hard the prompt is
fiery gull
I think its based on how hard the prompt is
no, I think this depend of day
but we are in lmarena server calm down 🤣
gentle pasture
Xhigh loaded, anyone want me to test a prompt?
how much u pay for that
i neve rused that
fiery gull
how much u pay for that
I think 1$ lol, not joking
compact flame
the OpenAi is hungry
Anyways uh when you finish comparing tell us how good is that extra high
Unless this extra high means it's on drugs
hushed gyro
Chat, tbh, I didn't find GPT 5.2 that good at reasoning and creative writing
native yarrow
1$ per message
native yarrow
gpt new model i bet isn't good i assume
balmy mist
gpt new model i bet isn't good i assume
what are scores? and price?
like is it better than g3 or pus 4.5?
native yarrow
i haven't tested
balmy mist
me either, after g3 came out i stopped caring fr lol
zealous sparrow
https://x.com/bcaine/status/1999212747213656072?s=20
gemini's OCR is just too good
swift oyster
how much u pay for that
its $21/M input tokens
$168/M output tokens something like this
compact flame
I mean all I can say OpenAi is good at video generating that all
swift oyster
for 5.2 pro
gentle pasture
its $21/M input tokens
$168/M output tokens something like this
oh lord
god bless gemini 3
i cheat a lot in exams
zealous sparrow
for 5.2 pro
how much is 5.2 pro xhigh
fleet lintel
is this true?
https://www.reddit.com/r/singularity/comments/1pk6n9w/openai_is_cheating_somewhat_with_their_xhigh/
compact flame
god bless gemini 3
Fair Gemini got me good grades
fleet lintel
That plus user get 5.2 medium ?
gentle pasture
Fair Gemini got me good grades
holy grail
compact flame
Though I feel like
cloud zinc
That plus user get 5.2 medium ?
yes its true
swift oyster
god bless gemini 3
OpenAI seems to have rushed the launch of this update, complete with steep pricing, in a bid to temporarily reclaim benchmark dominance. In my view, the core model is essentially GPT-5.1 with targeted post-training refinements focused on areas where Gemini 3 excelled and generated significant buzz on X.
fleet lintel
yes its true
🙁 this is very wrong .. very scammy behavior
hardy lion
OpenAI seems to have rushed the launch of this update, complete with steep prici...
which areas are the ones they targeted the most?
zealous sparrow
Im so hoping this doesn't happen
gentle pasture
OpenAI seems to have rushed the launch of this update, complete with steep prici...
they wanna win the 2025 ai race
compact flame
Im so hoping this doesn't happen
Fr or we are all cooked
cloud zinc
which areas are the ones they targeted the most?
arc agi, math, swe
gentle pasture
whats swe?
swift oyster
google model is most efficient
Google is making use of the TPU's - Good for them. OpenAI are going back to AMD
hardy lion
I did a few of my favorite mathy rating systems questions side by side to gpt-5.1-high and gpt-5.2-high and they were either the same or a couple 5.1 was better. I was suprised
swift oyster
I did a few of my favorite mathy rating systems questions side by side to gpt-5....
Yes, hearing this just makes me firm that the base is 5.1 .
zealous sparrow
Yes, hearing this just makes me firm that the base is 5.1 .
Probably so is.. The UI is the same as 5.1
swift oyster
the base is gpt 4o
No, GPT-4o was one of the best models to come from OpenAI.
zealous sparrow
sour spear
openai model are slow and clunky
This. Gemini 3 is fast. GPT-5 is slow af and often stops mid-generation. Let alone the small context window. I preferred Gemini 2.5 Pro over GPT-5 already, simply because I could iterate so quickly.
blazing bison
i don't like gemini 3 bcs it just ignores your instructions
zealous sparrow
i don't like gemini 3 bcs it just ignores your instructions
it doesnt ignore system instructions on AIStudio tho
blazing bison
it ignores for me
zealous sparrow
Not for me
cloud zinc
not for me
blazing bison
each model is good at different use cases
cloud zinc
sure.. vibes
sour spear
it ignores for me
It does what I tell it to, but if my instructions are bad, it also tells me so quite bluntly, instead of hallucinating bs just to please me.
swift oyster
each model is good at different use cases
I agree, but the pricing is outrageous.
echo aurora
@sage copper You'll want to use /video in #video-arena-1 , check out the information in #1397655624103493813 and let me know if you have any questions.
blazing bison
in my opinion openai still the best $20 deal
swift oyster
in my opinion openai still the best $20 deal
Output is much higher 🙂
swift oyster
in my opinion openai still the best $20 deal
For the general ChatGPT interface, sure why not. General Public it's perfect.
blazing bison
i think the model is better, my results are better with chatgpt idk, just a feeling
and i have all ai plans
swift oyster
i think the model is better, my results are better with chatgpt idk, just a feel...
Out of curiosity have you tried our model?
blazing bison
nop
swift oyster
nop
it's free https://www.movementlabs.ai/ would love to hear some feedback.
blazing bison
🤔
sour spear
in my opinion openai still the best $20 deal
Google will give you 2TB of cloud storage in top, that's huge. If depends on your use case, but I've cancelled my ChatGPT sub after two years, it just wasn't worth it anymore for me.
blazing bison
Google will give you 2TB of cloud storage in top, that's huge. If depends on you...
ye i don't trust google with storage
i use my own
fiery gull
Google will give you 2TB of cloud storage in top, that's huge. If depends on you...
notebook lm and etc
blazing bison
and yes i created a whole storage system in my home
😆
we have things like groq so, maybe
🤔
groq was not that some years ago?
but yeah, i don't trust random ppl in disc
cloud zinc
i dont trust openai
echo aurora
@nocturne heath you'll want to review the information in #1397655624103493813 on how to properly use the Video Arena bot
cloud zinc
2TB cloud storage wow google
sweet tinsel
The MovementlabsAI stuff again, the cycle repeats itself.
@swift oyster I am still waiting on that e-mail, by the way.
swift oyster
There has been a major shift in our company, we are updating team page, we also have a new CEO. Please bare with us while we check the backlog. Thank you.
queen veldt
Yeah movementlabs might be legit company
We talked with them few days ago
I've had my doubts too
Okay why?
They aren't selling anything you can go and try the model yourself
You don't have to buy to try it
Proof for false advertisement?
vivid coral
Where is GPT 5.2 search? Is Sammy A. scared for us to test it? 🤔
queen veldt
Okay 😂🤣🤣👍
sweet tinsel
Proof for false advertisement?
Just check out my messages and https://discord.com/channels/1340554757349179412/1435953842956013620 please.
echo aurora
Where is GPT 5.2 search? Is Sammy A. scared for us to test it? 🤔
I'll check with the team if we can add to search 
lusty tinsel
I'll check with the team if we can add to search <:blobthanks:825444835460644929...
when add attach txt and code files? 😆
compact flame
Why do people keep requesting gpt pro models
balmy mist
I agree, but the pricing is outrageous.
hm?
whole sundial
they do, it's just very expensive
echo aurora
when add attach txt and code files? 😆
Tbd! As soon as we have new updates we'll be sure to share with you all.
queen veldt
Not worth the price too
compact flame
they do, it's just very expensive
Fair point
compact flame
I mean atp people just gotta buy pro subscription
Like there no way any free services gonna provide gpt pro models
queen veldt
It's crazy how the o3 was best back in the day
I remember that it was expensive and super powerful at that time now when i look at it at this chart it's funny
compact flame
But though damn gpt pro costs like a cheap laptop
That just gonna work for only a month
compact flame
OpenAi be getting greedy these days
I mean I know they provide high quality models but still
compact flame
7.5$ is per prompt?
spark python
THIS is greed
Pro = add 1 more 0
vivid coral
I'll check with the team if we can add to search <:blobthanks:825444835460644929...
As always, you the man
compact flame
Damn very expensive
Also I check
Uhh free users don't have 5.2 in chatgpt
Even team sub
queen veldt
Not yet maybe
queen veldt
It'll probably come in few days maybe they want their pro subscribers to have it first
So they can potentially make someone who's hyped to buy pro
compact flame
So they can potentially make someone who's hyped to buy pro
Yeah or maybe plus users have it too just somewhy team subs get these models after a long time
queen veldt
Well i guess for teams they want it to work with no issues since it's for businesses
balmy mist
just tried gpt-5.2 with my pokemon test and it was horrible, imma get it another chance cause wtf lol
queen veldt
Where people are not that smart
queen veldt
https://019b0f51-896f-786c-a11b-8ae511bdfd57.arena.site
This is terrible ui 💀
balmy mist
ik lmaoo like wtf
gemini 3 and opus 4.5 blows it out the park in that test so maybe it just not good with UI?
granite tide
try inputing "good ui" lol
atomic lagoon
it's free https://www.movementlabs.ai/ would love to hear some feedback.
Its not that bad, but on the coding side of things its probably where GPT 4o is
torn mantle
terrible
atomic lagoon
What
leaden meteor
So, all those gains in ARC-AGI-2 does not mean much in arena's vibe testing for gpt-5.2-high? I would have thought it atleast competes well with gemini 3 pro but everyone saying it is only as good as 5.1 on text arena?!
empty stump
I dont understand the upgrade
atomic lagoon
So, all those gains in ARC-AGI-2 does not mean much in arena's vibe testing for ...
5.2 is a lot better at coding, thats all I know
OpenAI likes doing updates on their models to upgrade like one thing at a time and its gonna kill them
leaden meteor
Yes, I am wondering more about its non-coding ability...
atomic lagoon
Okay cool, it would make sense if its still like 5.1
Because 5.1 was an AGI update
empty stump
ok its always coding and math but what about general use
atomic lagoon
ok its always coding and math but what about general use
Because I guarantee you majority of users are coding or doing some kind of code, literally LAST update 5.1 was about general use
empty stump
5.1 was disappointing
atomic lagoon
Sucks to suck then man, I don't know
leaden meteor
Yeah, but 5.2 ARC-AGI-2 is much higher than 5.1. And that does not seem to translate much to non-coding performance based on people opinion about 5.2/robin high here. Lot of peope here test on arena with puzzles and abstraction ability. So, I am surprised that even with such a high ARC-AGI-2, robin/5.2 is not sota on text arena....
atomic lagoon
Yeah, but 5.2 ARC-AGI-2 is much higher than 5.1. And that does not seem to trans...
Idk where you got puzzles and abstraction ability from lol, but uh the second its released everyones going on webdev and making UIs with it
They're testing their coding capabilities
It happens with every model
empty stump
like claude can be natural and coherent, and doesn't always tell you what you want to hear and its good and fast without reasoning even though it may struggle with deep reasoning tasks
queen veldt
Openai making confusions not with models only but also with versions too 😂
Guess what version is coming next
5.21o
Codex max super ultra x high
atomic lagoon
Openai making confusions not with models only but also with versions too 😂
AIs have confusion with models because their data isn't updated to the current day. Gemini 3 did the same thing
queen veldt
No i mean they made the models be confusing at first with the names
Nano mini regular low thinking extended thinking pro thinking
sour spear
Codex max super ultra x high
Like Apple 😉
torn mantle
idc anymore
oak granite
Hi
pseudo summit
i dont think this is Google either
proud bobcat
Hot take
GPT 5.2 and 5.2 high are different models
Like completely different
I think 5.2 high is that garlic thing they were preparing
Also I’m not surprised with 5.2 honestly
It’s mediocre
cedar lotus
How to creat video
echo aurora
How to creat video
Check out #1397655624103493813 , basically you'll want to use /video in #video-arena-1 . Let me know if you have questions.
golden ocean
sour spear
Gemini 2.5 still going strong 💪👴
proud bobcat
GPT 5.2 is hella expensive as well for a gpt model
I don’t know what the use case will be
Cause coding and roleplaying are becoming favored for api usage
I think this model will mostly be a minor release regardless of the benchmarks
People would still rather pay for Gemini 3 and Opus
burnt sinew
is 5.2 good or nah
burnt sinew
People would still rather pay for Gemini 3 and Opus
they are all free
with data sharing
sour spear
is 5.2 good or nah
Mediocre. Neither particularly good nor really bad. It's just another incremental update.
burnt sinew
Mediocre. Neither particularly good nor really bad. It's just another incrementa...
worse than gemini 3/
sour spear
worse than gemini 3/
Definitely
fiery gull
worse than gemini 3/
I used gpt 5.2 xhigh vs gemini 3.0 pro antigravity agentic mode and gpt gave a better result but it was only 1 work
Today I'll use 100% hard work in text and docs 🥱 and I'll talk who is better for me
Still the opus 4.5
sour spear
"Create a 3D voxel animation in JavaScript, depicting a pelican riding a bicycle."
Gemini 3 did a really neat job. And most importantly, it was fast.
burnt sinew
"Create a 3D voxel animation in JavaScript, depicting a pelican riding a bicycle...
speed isnt really a big thing imo
much more time is saved from quality
when you are coding
fiery gull
"Create a 3D voxel animation in JavaScript, depicting a pelican riding a bicycle...
Imagine people from 2024 seeing this 🤣
Imagine 2026 💀
Anxious...
burnt sinew
Imagine 2026 💀
"Create GTA 7"
sour spear
Imagine 2026 💀
Imagine 2030.
fiery gull
"Create GTA 7"
Not joking, ps2 games in one prompt
echo aurora
Imagine 2030.
"Create GTA 8"
fiery gull
Gta san andreas in first prompt 💀
fiery gull
In 2 minutes lol
burnt sinew
Gta san andreas in first prompt 💀
it already can right now for sure, it just needs prompting
and assets
proud bobcat
It would only have 50 errors
Dwdw
Ultimately in my eyes gpt is not a profitable model
fiery gull
it already can right now for sure, it just needs prompting
Hmmm maybe the anti grativy gemini 3.0 pro create a gta 3 with one prompt 🤔
burnt sinew
Hmmm maybe the anti grativy gemini 3.0 pro create a gta 3 with one prompt 🤔
https://onecompiler.com/html/445jfmjgx i one shotted this with gemini 3
fiery gull
https://onecompiler.com/html/445jfmjgx i one shotted this with gemini 3
I'm mobile now, I'll see it later
fiery gull
i think it should work on mobile?
Don't work 😐
burnt sinew
Don't work 😐
ah
fiery gull
For me
burnt sinew
well its flappy bird 1:1 clone
neon idol
Sorry, I just use 3 works, and both tie
Oh
fiery gull
2026 will be fun, see a model creating a gta 3 in first prompt and you think it is bad because the gemini 4.0 ultra create a gta 5
proud bobcat
2026 will be fun, see a model creating a gta 3 in first prompt and you think it ...
Gemini 6.7 creates one shot steal a brainrot
jade egret
gpt 5.2 good?
jade egret
oh
proud bobcat
It’s a fine model but it doesn’t really like excel at anything
So I think opus and gemini still rule
torn mantle
hmm