general

[fiery gull]

even fp4

[proud bobcat]

My dude I don’t know where you’re getting the thing that’s it’s 3.2 exp

I genuinely don’t

[surreal zephyr]

yeah but i tried to make it more fair cuz deepshit is like 0.5 terabyte

[stray aspen]

alibaba's tiny models are smarter than deepseek v4 pro 😂

[fiery gull]

3.5 nah

[stray aspen]

3.5 is ass

3.6 was a step forward

[fiery gull]

3.6 is soooooooooooooo better that the 3.5

[proud bobcat]

Qwen’s models are great

[fiery gull]

like the 3.0 and 3.0 2507 the diference is sooooo big in the pratice

[proud bobcat]

I just wish they would stop limiting their big MoEs so much

[grim cliff]

Guys does anyone know a good model In science and creativity?

[fiery gull]

the 31b in someeeeeeeee thing is a little better

[stray aspen]

oh my god finally

[surreal zephyr]

3.5 was still better than gemma4 and deepseek v3.2

[stray aspen]

it better be good

[surreal zephyr]

three fiddy it wont even run

[split topaz]

Deepseek V 3.2 Speciale had a good rating in CritPt for that era so we can say that Deepseek can be good in science and research. But I need to check how V4 did in CritPt in AA. Sorry I didn't read your response well initially. I believe that some models can be good for science and creativity.

[stray aspen]

mimo for frontend

deepseek for math and backend

[surreal zephyr]

qwen 27b

bro

😭

[split topaz]

The benchmark of Vals.ai somehow.

[stray aspen]

it is

because they are both bad

[obsidian cargo]

Tbf it's also fifty times cheaper

[grim cliff]

Does anyone know Hofburg_2_alt or is it a new model

[stray aspen]

mimo is actually decent for frontend ngl

[sonic wigeon]

they didn't have good GPUs i guess
they had to use chinese gpus

they can't use nvidia
its not allowed

[grim cliff]

I mean its obvciously AI you can see it from the background, its not that creative but yea

[lavish current]

will the new deepseek model be free?

[grim cliff]

Its already on arena

[surreal zephyr]

5.5 frontend

[lavish current]

oh

[stray aspen]

gpt image 2

[surreal zephyr]

yes codex + gpt image 2

it used image2 to generate the assets

[rancid prawn]

guys, can someone explain me what claude code is? I need claude opus (because any other ai can't help me with my mod), I heard that sub costs 25$, but what is claude code

[surreal zephyr]

lol

1. opus sub costs 110$

2. it gives you 5 requests per day

[stray aspen]

just use a chinese model bruh

[urban herald]

meow_party

[sonic wigeon]

claude code got leaked
you can use a better cheaper model with it

[urban herald]

minimax m2.5 free api key go brr

[stray aspen]

m2.5 is bad

we have 2.7

[urban herald]

free and unlimited via my method

[stray aspen]

yeah but its bad

[urban herald]

okey 💔

[proud bobcat]

Yeah even as a DeepSeek fan this is bull

[proud bobcat]

What did you actually think is good besides Mimo

Every model for you is bad

Kimi is bad

[stray aspen]

m2.5 is old

[proud bobcat]

DeepSeek is bad

[urban herald]

still being used ✌️

[proud bobcat]

2.5 is still very capable

[stray aspen]

its bad

[proud bobcat]

Eh to each their own

I’ll have to test it out some more today to give my true thoughts

[urban herald]

still being used tho

i wonder why

[stray aspen]

cause your giving it for free

[proud bobcat]

What?

[urban herald]

yes

[proud bobcat]

You do realize 2.5 is still more popular than 2.7 on openrouter dude

Well too bad for them

[stray aspen]

whhen

[split topaz]

Well yeah there seems to be a certain amount of benchmaxxing here ig

[proud bobcat]

I think there’s just a lot of hype around models

Not everything needs to be SOTA

Again you can prefer what you want

[stray aspen]

whats this ass wawy of coding

feells like slopus 4.7

unreadeable

brainrot

[surreal zephyr]

Lol

18gb

For decent q4 quant

16gb for lobotomized with small context

Thats why they should release 20b models 😭😭

[ocean vortex]

nvidia should stop gatekeeping vram. Although that's even less likely now with all that hoarding

[brazen thunder]

Is DeepSeek v4 not updated in their own web?

[ocean vortex]

so many consumer gpus are essentially memory limited on purpose

[stray aspen]

it is

[zenith steppe]

🙂

[surreal zephyr]

its not even that

there is NO ram shortage

its ram OVERDEMAND

because openai, google, xai went:

"hey nvidia im buying all your ram for next 10 years, im paying instantly, 10x more than normal customers"

[ocean vortex]

I mean nVidia were doing this before hoarding was a thing

[devout vault]

claude is his own person

[ocean vortex]

So now they are even less likely to change that

high vram exclusive to expensive af enterprise gpus

[surreal zephyr]

tbf yes

[surreal zephyr]

7900xtx has 24gb vram

costs less than 5070ti

[desert pendant]

Guys for yall which one is better, Mimo (new model) or Deep seek (new model)

[surreal zephyr]

scammer

[zenith steppe]

It is no-bullșit claude

[surreal zephyr]

id show you something gpt 5.5 said but its too vulgar to send here (no jailbreaks btw)

[tranquil marsh]

whats the best AI model i can run off ollama>

[surreal zephyr]

qwen 27b

[stray aspen]

mimo

deepseek

[surreal zephyr]

bad

less bad but bad

[stray aspen]

deepseek si worse wdym

[zenith steppe]

This should be illegal and forbidden , some stupid people mixes risky business with gamebling at this point :v , then complain about a financial crisis and dot-com boom etc , stop buying fishes in the sea and your life will be ez :c

[surreal zephyr]

how illegal?
paying in advance is completely legal, and even encouraged

imagine going to car store and saying "im buying them all, heres 10,000,000,000$

[zenith steppe]

Yes , and this should stop being the case , how crazy it is for someone to pay 10 years in advance ?

[surreal zephyr]

thats what xai google and openai did

[surreal zephyr]

bro if you paid for your wifi subscription for 10 years in advance thats good not bad

[zenith steppe]

Well buying the whole stock is fine , buying imaginary numbers in the future it is stupid

[surreal zephyr]

its on them, if they can afford it then its good

[surreal zephyr]

yes its stupid but THEY take the risk

lol

[proud bobcat]

DeepSeek looks better

[surreal zephyr]

its stupid FOR THEM if it fails

[cedar pulsar]

How do you generate ai videos?

[inner relic]

yeh i agree

[zenith steppe]

What If the wifi company decided to disapear at some point ?

[cedar pulsar]

Saw a yt vid that you join the disc. Not seeing any channels for it rn

[surreal zephyr]

then you got scammed, pity to you

[echo sinew]

Hello! The Video Arena is only accessible through: https://arena.ai/video. More information on how to use Video Arena can be found in this article.

The Discord bot was removed from the server and is no longer available. This announcement has more information about the removal.

[cedar pulsar]

Life saver. Ty!

[echo sinew]

Most welcome!

[zenith steppe]

Or what if wifi technology becomes cheaper or less important and slow comparing to modern technologies etc

Unless you are a giga company never pay for many years in advance , those big bros are paying from other peoples money not thiers , bwahaha

[surreal zephyr]

then its good for nvidia because they overpaid?

...and they are giga companies

they earn more money than gpd of smaller countries 🤣

[visual coral]

is grok 4.3 not in the arena yet? no suspecged codenands?

[fiery gull]

grok 4.3 not released the api yet

[wicked sage]

hi guys

[urban herald]

hi

[fiery gull]

tested the v4?

[wicked sage]

❌

[stray aspen]

🥀

[primal depot]

give me gpt 5.5

[zenith steppe]

True , and they cause catastrophic nation wide or even international crisises ,
Oh like the RAM surge we are seeing these days haha

[desert pendant]

I can't lie but i think that DeepSeek and Mimo are both overhyped

[keen beacon]

predicted the future💀

[surreal zephyr]

arguable

[frosty lava]

its not top 1 in coding and its also not the smartest ai ever

[grave peak]

Deep deepseek sahur

https://tenor.com/view/deepseek-thanks-love-deepseek-thanks-deepseek-gif-9297772848012106761

[keen beacon]

The funny thing is that the guy said yesterday that she would never come out

fr

[wicked sage]

https://cdn.discordapp.com/attachments/1060035543901155360/1497238412091068536/image3.gif

[zenith steppe]

It is only good because it is 1M , also energy is cheaper in china , and they do not over price or woke restrict their models as @lunar swan said in a lengthy article on his ai magazine the other day

I feel like if china started to use ai for military and other state related tasks , the avaliability of these things will be very low and therefore they need to reduce the quality to keep it going

[surreal zephyr]

its still reasoning just low reasoning

[frosty lava]

i can't wait for it to be even better at frontend

the release happen so fast

gpt 5.6 soon

[grave peak]

That's u opinion

[stray aspen]

It is reasoning tho

[desert pendant]

both Mimo and Deep are hyped

dawg

[zenith steppe]

Claude 4.6 thinking ; was horrible ; it could easily be stuck in a thinking loop , , low or medium is sometimes better than high esp for someone who dont know how to properly prompt a reasoning model

[stray aspen]

Nah

[desert pendant]

😭 🙏

[stray aspen]

Mimo is actually good for frontend

[desert pendant]

Mimo sometimes make some weird menus

but yeah

[grave peak]

Mimo is really that good?

[desert pendant]

Not really, sometimes you have to be really specific at some unnecessary points

[pastel ember]

Any Chinese model is basically total junk, except maybe GLM.

[sterile tartan]

Confirmation Bias.

[proud bobcat]

Not really

Kimi K2.6 is awesome

It’s a very nice alternative for Opus 4.7 right now

GLM 5.1 I expected more from

[sterile tartan]

What about Deepseek @proud bobcat

[night moat]

When will gpt image 2 be fixed?

[pastel ember]

Kimi K2.6 is totally scarfing down tokens like crazy, bouncing tool calls back and forth. GLM5 is way more chill with resources and acts kinda like Opus 4.6.

[proud bobcat]

It’s solid

Reliable

[sterile tartan]

Indeed

[shrewd citrus]

it’s easy to jailbreak

very very easy

[frosty lava]

lol deepseek literally just say don't do it its bad ! and proceed to give you the full thing

don't even have to jailbreak anything it will just do it anyway

[polar horizon]

yeah bro
gpt image 2 is running laps around every other mf

Gemini get up bro
do something

[light sleet]

Bro

[sleek urchin]

Anyone know what does this mean?

Why is model rising-sun

[surreal zephyr]

codename for crappy chinese model

[sleek urchin]

Thanks

[strong igloo]

/arena

[vagrant temple]

Deepseek,Glm,Kimi or Minimax?

[surreal zephyr]

bro be playin

[grim cliff]

I like Deepseek a lot lowk

[vagrant temple]

Ok and why?I like Glm a bit more then deepseek

[obtuse heart]

GLM 5.1

[surreal zephyr]

kimi 2.6 > qwen 3.6 > minimax 2.7 ~= glm 5.1

glm bad

[noble quarry]

Why glm bad?

[gray lion]

is kimi 2.6 the best coding ai thats not on battle mode

[surreal zephyr]

overrated

[surreal zephyr]

gpt 5.5 is best

not even close

[gray lion]

i mean like on direct mode in arena

[proud bobcat]

No AA score on DeepSeek V4 yet

[proud bobcat]

GPT, hallucinate code for me

https://tenor.com/view/jarvis-erase-gif-1159901542694943356

[vagrant temple]

https://tenor.com/view/goodfellas-laugh-liotta-gif-13856770980338154376

nahh Amazon Nova ist the best for coding

[proud bobcat]

Nah

Llama 4 Maverick baby

[surreal zephyr]

skill issue

[surreal zephyr]

• most reliable model out there

[noble quarry]

Which model best for coding??

[vagrant temple]

this

[vagrant temple]

claude opus 4.7

[gleaming roost]

gpt 5.5

[frosty lava]

People just won't ever forgive openai to have removed 4o

its not hate based on real issue its pure hate just cause they removed 4o

gpt could be 10x better than opus there will be someone to say opus is better just because of that

y'all don't realise 4o was actually making people psychotic

it was really dangerous

[proud bobcat]

Right

I say as it has a 86% hallucination rate

3% improved from 5.4 btw

[sterile tartan]

Hallucination Issues

[proud bobcat]

If gpt doesn’t know something it will gaslight you to hell and back

Instead of just saying “whoops my approach was wrong! Let me try it this way”

No issues with Kimi on that!

Or Claude!

[frosty lava]

it didn't happened to me and even with gpt 5.4

neither gpt 5.3 codex

[surreal zephyr]

huh

[frosty lava]

like hallucinations isn't a real problem anymore

[surreal zephyr]

you realise that benchmark counts refuse all answers as 0% hall rate?

so opus being dumb and not even trying gets better score lmao

do you notice grok is best on there despite being one of worst models? guess why

[frosty lava]

where you get this from

[proud bobcat]

Waaa waaa waaa

Opus gets a 36% score

Waaaa waaa waa

[surreal zephyr]

api tracker

[surreal zephyr]

grok gets 13% score

"waaaa waaa waaa"

[frosty lava]

if they release a 5.5 codex it'll be even better bro

[surreal zephyr]

grok > opus

🥀

waaa waaa waa

[proud bobcat]

Yeah cause grok does thorough searching

No wonder

[surreal zephyr]

....have you even read that benchmark?

😭

[frosty lava]

A model is good at benchmark people say its benchmaxxed, but if the model ain't top 1 in the benchmark people will say see its a proof its bad

all contradictory

every single time

[proud bobcat]

“AA-Omniscience is a knowledge and hallucination benchmark that rewards accuracy, punishes bad guesses, and provides a comprehensive view of which models produce factually reliable outputs across different domains”

[surreal zephyr]

top1

in accuracy

where refusing everything actually matters

aka the BETTER benchmark

[proud bobcat]

“AA-Omniscience Hallucination Rate measures how often the model answers incorrectly when it should have refused or admitted to not knowing the answer. It is defined as the proportion of incorrect answers out of all non-correct responses”

So 86% of the time it did not know something it hallucinated an answer

Astonishing

[surreal zephyr]

printf ("i dont know");
gets perfect score on that

10iq

[proud bobcat]

Exactly because that’s what it’s about

If a model just guesses 86% of the time it doesn’t know something

That’s kind of dangerous

I’d say

[surreal zephyr]

im model refuses 100% times when unsure

then its useless

[proud bobcat]

That’s why they use search

[surreal zephyr]

BRO THEY DONT HAVE SEARCH

IN THE BENCH

😭

[proud bobcat]

Okay so tell me

Tell me this

Does it make it okay if you ask me about something important

And since I don’t know id make it up?

[surreal zephyr]

3+1= ?
a. around 4 but im not sure
b. i have no idea

[proud bobcat]

This is mathematics

Models can do math easily

A more apt comparison would be

“Hey why is this line of code outputting an error?”

A. Random gibberish that’s completely nonsensical
B. I don’t know, but I’d need more info

Claude almost always responds with B

[surreal zephyr]

correct answer is

4, BUT ONLY if you mean base10

see?

humans guess too. you thought it was just 4, while there was actually missing context

[proud bobcat]

Do you know what chatgpt told me

The reason my game was stuttering was due to vram

A game that takes less than 2 GBs of vram.

[surreal zephyr]

innate to NEURAL NETS as whole

humans too

[surreal zephyr]

missing context

[proud bobcat]

It then proceeded to gaslight me about it for 3 messages

[proud bobcat]

Arma 3 experiences stutters when used with the proton compatibility layer for Linux

I asked Claude about it and we came to the conclusion it was just the compatibility layer acting up

Upon asking gpt with the same info it fed me nonsensical garbage

[surreal zephyr]

because claude is told to waste 200k tokens on research agents before doing anything

[proud bobcat]

“Ah, I see! It’s the vram!”

“But Arma doesn’t need much vram.”

“Welll you never knowww”

[surreal zephyr]

exactly

[surreal zephyr]

how about give it full access to pc, tell it ALL CONTEXT, and tell it to test

[proud bobcat]

I would rather use sticks to make a fire than let an ai control my pc

Absolute brain dead solution

[surreal zephyr]

you realise claude has no sandbox

while codex has sandbox

thats why

[proud bobcat]

I use Claude in app

[loud herald]

<@&1349916362595635286>

[surreal zephyr]

@echo aurora wakeup

[proud bobcat]

Why the hell would I use codex for ts???

[echo aurora]

[surreal zephyr]

when ai ocr automod 🙏

autoban if see word "mrbeast"

[loud herald]

What are you trying to do?

[proud bobcat]

Ah yes grant codex access to my entire pc to diagnose why a game is stuttering

[echo aurora]

It's an image 🙁

[surreal zephyr]

ocr is cheap tho?

[echo aurora]

Automodding that word won't help sadly

[surreal zephyr]

all those scam images have word mrbeast in them

[echo aurora]

Sry what is ocr? I'm not familiar with that.

[frosty lava]

gpt won't hack you bro

[surreal zephyr]

optical content recognition

image -> text

[loud herald]

Riught but if I said Mrbeast here then it'd be deleted lol

[proud bobcat]

Basically I was making a point that chatgpt hallucinates a lot

It told me that the reason my game was stuttering was vram shortages, which is entirely false because I specified

The exact fps and frame times I was getting
Proton versions I used
My specs
Linux distro
And what I tried

[surreal zephyr]

samsung gallery for example OCRs all images you take for easier search

[echo aurora]

I'll look into mod bots that can help with that.

[loud herald]

OCR Can be difficult unless youn put specific images for it to delete

[echo aurora]

Our current system use to catch these. Unclear why so many have been getting through in the last couple of months.

[surreal zephyr]

just make it detect word mrbeast

thats it

it doesnt even have to patter match

[loud herald]

Like I said, that'd delete everything that said mrbeast not just those images

[surreal zephyr]

name 1 actual purpose of someone sending 4 mrbeast images here

only scammers do that

[proud bobcat]

Well in all honesty who talks about mrbeast in here anyway

Aside from scammers

[loud herald]

Yea GPT does hallucinate a lot, but to be fair every big model does because they are so persistant on safety

[proud bobcat]

Exactly

[loud herald]

Not my point, if I type Mr Beast it would be deleted and if they have auto ban I'd be banned along with it

[frosty lava]

Actually i don't think it hallucinate that much you just have to give it the context and access of course if you want it to do in a full restricted sandbox i guess it'll hallucinate

[loud herald]

If I sent an image of mr beasts name at any time I'd be banned or it'd get deleted

It'd be trouble

[loud herald]

Yea thats true too

[echo dome]

<@&1349916362595635286>

[surreal zephyr]

literally just ban those words

via ocr

its trivial

theres noone here who mentions crpt casins in any reason other than scam

[proud bobcat]

Yeah but

We’re talking about diagnosing something simple

Claude sonnet had no problem

GPT was throwing stuff at the wall to see what sticks

I will NOT be giving an ai access to my pc

Ever

[echo aurora]

Looking into. Thanks for the rec

[surreal zephyr]

i never had modern claude not gpt hallucinate.
but i never had gpt not listen to my instructions, whereas claude doesnt listen very often

[loud herald]

Brother 💀 if you ban those and someone sends them, its going to automatically delete or ban them

I don't understand what you don't get when I say that

[surreal zephyr]

yes thats the point

noone is supposed to send scam

[loud herald]

Dude, anyone can send a thing that says mr beast and has a twitter checkmark next to it not just a hacked account

[surreal zephyr]

anyone sends cryp toca sinos?

nope

only scammers do

[loud herald]

Holy moly, you're missing the entire point here

[surreal zephyr]

bro if you send a scam you are supposed to get muted

future is deepthinking models on instant interference (hardwired silicon)

taalas hc2 for example

[loud herald]

That will lower quality but it'd be good for everyday use

[surreal zephyr]

thinking is for the fact you can come up with idea then realise its wrong

1 month of qwen 3.6 27b outsmarts 30s of gpt 5.5 pro

and hardware scaling is ez with asic

[soft gale]

Is it just me or qwen3-235b-a22b-thinking-2507 sucks on long context Story Writing?

[loud verge]

I guess arena is really struggling.

They disabled file attachment support for Gemini flash.

[vagrant temple]

just u, because nobody uses this model

[sterile tartan]

Use 3.6 max

[soft gale]

Ohh....

[soft gale]

Oooohhhhh i see

[sterile tartan]

Pretty Good Model

[loud herald]

Disabling file attachment isnt a struggle thing, its the type of chat you're inside of

[vagrant temple]

and model

[loud verge]

You mean it's disabled for direct chat?

I didn't put any message in there yet.

[loud herald]

^^^

[loud herald]

It could be

[loud verge]

Yeah. I was scrolling through the model selector.

[vagrant temple]

here

[loud herald]

Yea it doesnt have file attachments as you can see

[vagrant temple]

ik

[loud verge]

Kimi k2. 6 is the only decent model there that supports attachments. That and sonnet 4.6

[loud herald]

Kimi 2.6 is one of the best AIs overall right now

[loud verge]

Others have poor context accuracy.

[vagrant temple]

Does anyone else think Microsoft is secretly building a super-powerful AI, maybe even on a quantum computer like majorana? I can't believe a company this massive, with all that data and money, can only come up with something as crappy as Copilot

[light sleet]

No announcement? https://youtu.be/nDjMlfNbNNY

[loud herald]

No, because they are like google focusing on optimization

[loud verge]

Gpt 5.5 is almost mythos level

And opus 4.7 seems slightly worse than 4.6

[grim cliff]

it did 2 files

wow

[sleek urchin]

Hey everyone whats the best direct > Code. Model?

[loud verge]

Kimi

And sonnet extended if it exists.

[surreal zephyr]

theres no way someone voted 4.7

[sleek urchin]

Is glm 5.1 good?

[surreal zephyr]

ragebait

[sleek urchin]

I... did

Cus its honest

and doesnt spy

[loud verge]

It's okay. Not a significant upgrade.

[surreal zephyr]

4.6>4.5>>>4.7

[surreal zephyr]

lol funny cuz claude code was exposed for spying and training on browser data

[loud verge]

They be picking favourites, lol.

[sleek urchin]

Doesnt count

[loud verge]

Y'all don't seem hyped about Deepseek at all.

[sleek urchin]

Def not

[loud verge]

Yes

[sleek urchin]

Deepseek sucks

[surreal zephyr]

its literally 3.2 exp renamed to v4, confirmed

[sleek urchin]

Qwen is better

[loud verge]

Nope

1m context

1.5T model.

Way better context accuracy and awareness.

Y'all don't know sh1t

[surreal zephyr]

yeah.... no

[surreal zephyr]

renamed on arena

kept the score

so its same model

[loud verge]

Bro trusts arena scores.

[surreal zephyr]

bro im not talking about scores im talking about ITS SAME MODEL

they didnt, thats the point

😭

[loud verge]

I think you don't understand what context accuracy means.

Lol.

[surreal zephyr]

they kept same endpoint and renamed it

they release v4 pro yes

[sleek urchin]

Ye, I dont

[surreal zephyr]

but v4 flash is v3.2 experimental

xD

[sleek urchin]

https://tenor.com/view/batman-thinking-gif-18107621495886085895

[loud verge]

When I give it a huge file, it extracts data more accurately than Gemini 3.1 pro.

[sleek urchin]

Gemini sucks

[surreal zephyr]

thats like putting gpt 5.2 and calling it 5.4 mini while doing 0 changes

[loud verge]

Most models suck at Long context.

[sleek urchin]

https://tenor.com/view/social-credit-credit-social-уважение-плюс-уважение-gif-1626328442317885176

[sleek urchin]

Claude dont.

[loud verge]

Yes. Tried it with a 550k token text file.

[sleek urchin]

"I never tried it"

NAH

Qwen 3.6 plus >>>>>>>> Gemini latest Pro

[loud verge]

Dsv4 is better. Tested it vs ai studio Gemini pro.

[loud verge]

3.5 ?

[sleek urchin]

Dunno which is latest

[loud verge]

Not yet.

🙁

[sleek urchin]

3.1*

[loud herald]

<@&1349916362595635286>

[surreal zephyr]

@echo aurora when ocr

🥀

scammers, scammers everywhere

https://giphy.com/gifs/nounish-dao-nouns-noggles-z9tbWnGzUmzs5FNWj7

[loud herald]

The more you ping him isnt gonna make it any quicker, he already said he'd look into it

[surreal zephyr]

i pinged him to delete the dang scam

bro

^^^

[loud herald]

I already pinged the mods role, theres no point in pinging one mod just ping the whole role

[echo aurora]

I've updated some of the automod triggers. Wi'll have to wait and see if it's effective. 🤞

[light sleet]

Why no announcement

For the new vid

[desert pendant]

oh yeah

what it should do now pineapple?

happ

[echo aurora]

^ scam messages

[south ether]

gpt 5.5 still not in arena right

[whole sundial]

it's not in the public api, so...

i mean, arena could get an api from openai but they would most likely wait until full api release

[desert pendant]

yay~1

[fiery gull]

Qwen 3.6 27b > plus

[desert pendant]

yall cannot be this serious

[fiery gull]

@echo aurora Why qwen's 3.6 35b and 27b not release in lmarena?

[desert pendant]

prob high api cost

[echo aurora]

Sorry to say I won't be able to go into details about why specific model may or may not be there.

[fiery gull]

Yeah the alibaba prices is crazy 🤣

[slender ledge]

Will gpt 5.5 come in arena

Is it like mythos?

[abstract plinth]

😂

[burnt sinew]

No

[slender ledge]

Some ppl say

[abstract plinth]

it will eat mythos for breakfast

[light sleet]

didnt work?

[desert pendant]

nothing

[light sleet]

<@&1349916362595635286>

[desert pendant]

no mute or nothing.

[light sleet]

yeah

ANOTHER ONE

[slender ledge]

Hacked lol

[echo aurora]

Didn't work 🙁

[light sleet]

😭

[desert pendant]

son

[light sleet]

Another here

[abstract plinth]

are u guys trying automod or sm?

[light sleet]

these r image loggers or smth?

[echo aurora]

Yeah working on changing up our regex to try and catch these.

It use to work.

[desert pendant]

how are ppl getting hacked

from suddenly

[slender ledge]

U gotta pay before like u can withdraw

So scam

Pineapple

ADD gpt 5.4 and gpt 5.5 pls

[echo aurora]

We'll be sure to put out announcements for new models added.

[light sleet]

to the guy who did the 💩 emoji

5.5 is literally mythos level

💀

[slender ledge]

Fr?

[light sleet]

and mythos is not releasing to the public

[abstract plinth]

yeah

[light sleet]

yep

[slender ledge]

Btw

[quartz light]

5.5 HAS ENTERED THE ARENA

[abstract plinth]

https://klipy.com/gifs/alien-lockedinalien--k01KQ0BS8P43P2A676341153M1A

[quartz light]

!
@slender ledge

[quartz light]

^

[light sleet]

oh yeah it did

[desert pendant]

gngImtired

[echo aurora]

Lol

[slender ledge]

Arena should make smth like paid u get opus and gpt in direct

[slender ledge]

Omggggg

Pls also direct

Lemme look

[light sleet]

HOLY SH

IT'S IN DIRECT?

WHAT

@surreal zephyr

[slender ledge]

Omgggggg

[light sleet]

IT'S IN DIRECT

[cold gorge]

On direct

[light sleet]

WHAAAAAATTTTT

[light sleet]

YEAH

EVEN HIGH ALSO

[slender ledge]

Omggggggggggggffff

Wwwwwww

[light sleet]

@echo aurora credits system sneek peek 🔥 🔥

It's time to COOK.

[echo aurora]

[slender ledge]

Pineapple don’t delete it

[echo aurora]

New usage system isn't yet released

[light sleet]

pineapple please take a look on my supabase idea

yall should add integration to supabase

[echo aurora]

Can you link?

[surreal zephyr]

why is there 5.5 but not 5.4 :/

[slender ledge]

Why not in announcements

[echo aurora]

Working in it

[slender ledge]

don’t delete gpt 5.5

Pls

[light sleet]

Designarena has this ^^

[desert pendant]

i throught ts was a joke

its not

luigiscream

[light sleet]

fr same xd

[light sleet]

And it helps very much

[desert pendant]

"wow they forgot its not april fools day"

[slender ledge]

Wym

[desert pendant]

welp... Roblox is gonna be the first victim

[surreal zephyr]

this is alredy there

tho

[echo aurora]

I think you're confusing this with the current limitation systems we have in place

[icy frost]

no bro

the credits are on the site

[slender ledge]

Wil u delete gpt 5.5?

Don’t pls

And add gpt 5.4

[echo aurora]

Where are you seeing this? Can you share a screenshot?

[light sleet]

No it isnt

no wonder

[slender ledge]

What credits

[desert pendant]

gngImtired

[icy frost]

i was checking the frontend every day and i found that u added new lines

[light sleet]

What I meant was like u can ask to integrate supabase so it handles backend and stuff.

but rn you'd have to make all supabase tables manually and give it an API key which is boooring ngl.

[slender ledge]

what happened ?

Idgi

[light sleet]

peter has blue eyes?

[abstract plinth]

[echo aurora]

I'm not aware of the new usage system being "out" as an experiment, so I'm not sure what this refers to.

[light sleet]

pineapple ofc this my only test when a new big model drops https://019dc0c1-6360-79cc-8212-9505e03b55b8.arena.site/

[icy frost]

credit-system-m1 it is

[slender ledge]

What should I ask to got 5.5

[light sleet]

@surreal zephyr 5.5 high extremely good at frontend damn

[icy frost]

really

[slender ledge]

U see enable false

So yes

[echo aurora]

Want to note 5.5 is going to be in Battle mode only

[slender ledge]

Noooooooooo

[echo aurora]

So get those prompts in quick

[echo aurora]

Yeah sorry to say. 😭

[icy frost]

i'm not a pro coder to make a script that replaces that part

[molten cargo]

is there going to be an announcement when it gets released on battle mode?

[echo aurora]

Yeah prepping now

[primal orbit]

I'm glad it will be in the battle mode at all, thank you

will it be high or xhigh or smth?

[sterile tartan]

RIP

[grave peak]

lul

[desert pendant]

yay~1

bro

[light sleet]

yep its gone

the api calls now show something went wrong

[slender ledge]

Wtf

[slender ledge]

Yes

[desert pendant]

worst arena day

[light sleet]

atleast my pineapple website was done before it was gone 😭

[icy frost]

i will use lmarena to make the credits thing system true

[light sleet]

Worst Arena Day was when all good models were removed

[desert pendant]

👀

[desert pendant]

second part of the worst day

[grave peak]

fr

[desert pendant]

we lost Chatgpt 5.5

[light sleet]

It was obvious tho

[desert pendant]

when we are gonna stop losing the good models brah

sad2

[zealous sparrow]

they will keep going into battle till we get the credit system

[desert pendant]

happ

[icy frost]

yoo it worked

[desert pendant]

gotta hop on Mimo again

[icy frost]

but no credits

[desert pendant]

-# not going in Qwen cuz QWEN SUCKS

[desert pendant]

his next action:

[primal orbit]

it would be cool to have token giveaways to premium models here on discord to incentivize user engagement.

like you participate in a poll and have a chance to win 20k tokens with some premium model.

in direct mode

[slender thistle]

LLM finally reached its limit. We didn't see any development. They took our money, LLM's are still behind from perfection and guess what.. I will wait the next 15 years to use any LLM (If possible).

Ai is not going to take people's jobs, But these people will take yours

[loud herald]

GPT 5.5 mighttttttt be coming to arena

Finally on API

[loud herald]

What are you even saying 😹

[desert pendant]

it had

however they changed it to battle

[loud herald]

They just announced its on API it wasnt

Stealth model is believable though

[desert pendant]

bro

[tawny canyon]

@echo aurora hi , i hope you are doing well , is gpt 5.5 coming to arena?

[light sleet]

bro it's already out

[light sleet]

It was out before the openai announcement

was in direct for 4 mins and then BooooOoooooOm

[loud herald]

Yes obviously, its been out since yesterday, not once did I say it wasnt out

[loud herald]

Now where would this be

Doesn't show for me so its a cache issue

[light sleet]

What I meant is u said "might be coming to arena" but it was already out in Arena before you said it

[light sleet]

It was working in direct and side by side for like 4 mins but now moved to Battle

[tawny canyon]

gpt 5.5 is not in direct chat

[light sleet]

It was

[loud herald]

Sure as a stealth model, it couldn't have been the official model

[light sleet]

It definitely was official

Or myb

https://019dc0c1-6360-79cc-8212-9505e03b55b8.arena.site/

I generated this

Try it out

[tawny canyon]

guys

[light sleet]

(Click the pineapple too btw)

[loud herald]

Theres no point in making your words bold they dont do anything 😂

[tawny canyon]

i have a site wich haves shared acc of chatgpt plus and 200$account

[loud herald]

Literally couldnt have been because it hasnt been on official API until 20 minutes ago

[slender thistle]

do you think it's a joke? What did you achieve so far?

[tawny canyon]

but its chinese

u want or not

[rustic gale]

Saw it a few minutes ago. Sent a prompt. sOmeThinG wenT wrOng. Refreshed. Gone. Don't think we'll be seeing anything new from the big 4 in direct until arena starts a paywall

(non-accidentally, that is)

[primal depot]

yra

[glacial mulch]

not in direct 💔

[vale quest]

Oml

RELEASE IT TO US NOW

RELEASE IT

[light sleet]

Dude it came out like 30 seconds in arena before the api release

[loud herald]

Many of things, this is what AI is doing for me currently, people make thousands off AI you're uneducated

[solemn nacelle]

Is pineapple the yt guy

[loud herald]

Thats called a placeholder, know what that is?

[light sleet]

No that's Peter.

[loud herald]

They wont put a paywall lol so it will never be in direct

[primal orbit]

I seem to be stuck on generating in battle mode. Is this gpt 5.5 issue?

[vale quest]

@echo aurora PLS ATLEAST HAVE GPT 5.5 IN DIRECT ITS THE CHEAPEST OPTION

[solemn nacelle]

Who is Peter

[shy mason]

dudes not getting what hes saying

[loud herald]

Its the same price as opus 4.7 what are you talking about

[vale quest]

BRUH

[molten cargo]

its more expensive than gpt 5.4 and even that isnt in arena

[light sleet]

The guy in the vids

[shadow zodiac]

anything but some useful ai

[light sleet]

Idk his specific role

[restive charm]

I live in Syria, the website isn't working for me.

[vale quest]

PUT OPUS 4.7 UP THEN

[light sleet]

But probably a staff at arena

[solemn nacelle]

Added now

[surreal zephyr]

why 5.5 not on leaderboard

[loud herald]

HOW would that make any sense

[desert pendant]

they added it

but they put it in battle

[molten cargo]

i mean direct chat

[desert pendant]

after some minutes

[loud herald]

Yea he said 5.4 isnt not 5.5

[desert pendant]

oh mb

[hollow flicker]

I miss old arena where you could talk to the model in direct mode

[cursive pewter]

Yo am I the only one seeing models that don't exist on arena xd

[loud herald]

Holy moly people must be rage baiting with those votes 😭

[plucky sparrow]

how's gpt5.5 guys

[crude anchor]

Stop bro

[desert pendant]

couldn't even test gpt

[vale quest]

Make me

[crude anchor]

Just battle.

[tawny canyon]

guys are u seeing my mssges

[loud herald]

I've been seeing stuff on twitter about it. It is much better than anything we've had yet

[crude anchor]

I don't waste my time with whiners.

[shadow zodiac]

this is straight ass if u havent made a internal with grok or gemini then u not tapped in with vgk bypass also

[tawny canyon]

i just uploaded a link were u can have access to chatgpt 5.5

[shy mason]

🍑

[crude anchor]

ban this guy

[vale quest]

Yea ban tonileni

[crude anchor]

yeah

ban tonileni

[tawny canyon]

ah yes it was deleted by moderators

[loud herald]

This is a UI, nobody said anything about backend quality lmao

[echo aurora]

We're taking care of it

[tawny canyon]

whyyyyy lil

[vale quest]

Ok

[tawny canyon]

lol

[desert pendant]

happ

guys who is better

[crude anchor]

you share links Suspects

[tawny canyon]

oh sorry

didnt know it

just wanted to help

[desert pendant]

Qwen, Deep seek (New release), Mimo or any other model you know

[vale quest]

Yo you shared your email

[desert pendant]

who is better gng

[stray aspen]

What

[onyx frost]

When we are going to use gpt 5.5 in side by side 😞

[vale quest]

Mimo

[loud herald]

Oopsiesi

[vast kiln]

When will gpt 5.5 go on the leaderboard mkay?

[echo aurora]

Gotchu

[slender thistle]

lol you made me laugh so hard that my eye balls are coming out, already guys like you who debate and defend are most uneducated, stop showing your fake acheivement, i have already seen many times like your crap. In real life project its dumb. You are a hacker i see. Then lets see if you can get backend information from ai and get api token to use Ai for free. If you can do this, i will believe your statement

[loud herald]

Thanks lol

[echo aurora]

I don't have an ETA to share, but we'll be sure to post an announcement when ready.

[restive charm]

I live in Syria, the website isn't working for me.

[vale quest]

Also @loud herald what even was you doing was that some force model extension

[vast kiln]

ok

[light sleet]

pineapple, first generation with arena gpt 5.5 high (Click the Pineapple for surprises) https://019dc0c1-6360-79cc-8212-9505e03b55b8.arena.site/

[raven stone]

Hey everyone, I want to ask, what free models can be used in the CLI? You used to be able to use qwen3.6, but not anymore

[desert pendant]

bro is truly speed

[echo aurora]

Lol the night mode

[light sleet]

Xd

[loud herald]

Nobody said they're a hacker, why do you assume so much? This is a game exploit an external memory reader and spoofcall

[shadow zodiac]

u flexing some as ui

[light sleet]

There's make "Pina" dance too

[icy frost]

btw pineapple why did u remove my prediction

[slender thistle]

so did you make this or someone else?

[loud herald]

No one is flexing, all I did was tell him I'm doing something right with it because he believes AI has no purpose lmao

[desert pendant]

happ

[loud herald]

AI

[loud herald]

Wdym

[desert pendant]

what is even happening at this point

[slender thistle]

i said that did you use ai or you are using somones else image buddy

[surreal zephyr]

and people say opus is better at ui xD

[light sleet]

Fr

[loud herald]

Oh no, its an external aimbot/esp that I am attempting to make mb I couldnt read this correctly for some reason

[vale quest]

O

[loud herald]

Brother I just told you its AI, AI make the UI and Backend for me

[echo aurora]

It's possible that your IP address is currently being blocked by our systems. These blocks are applied automatically based on internal safeguards we have in place that are based on a variety of internal criteria designed to protect the reliability and security of our services. Unfortunately, we’re not able to share additional details about the specific reasons for the block, nor are we able to manually override or remove it at this time. We appreciate your understanding and apologize for any inconvenience this may cause.

[loud herald]

Would it be allowed to use a VPN in someones case like that?

[desert pendant]

sounds crazy

[loud herald]

Its not but its better for sure, they just decided to hype tf out of it

[raven stone]

Hi @echo aurora , why were the fresh versions of Claude, Gemini, and GPT removed from the site? And will they ever come back?

[echo aurora]

I'd check with our Terms of Use. I won't be able to provide interpretations of that.

[slender thistle]

but you said i am uneducated, but then you are. Its just UI. Its just a waterless bottle. Solve the real life problem, solve upwork project then say ai is a real deal unless no matter what people say we know it

[desert pendant]

deep seek v4 and gpt 5.5 both hyped

[loud herald]

Alright

[loud herald]

I thought deepseek would be better then it was, unfortunate

[echo aurora]

Gentle reminder @loud herald @slender thistle of one of our server rules:

Treat others with Respect. Be kind, assume good intent from others, and keep disagreements respectful.

[desert pendant]

they get beat by a random model btw

[loud herald]

I literally said in the message you just replied to, the AI made the UI and Backend for me

[loud herald]

Understood

[loud herald]

Which random model is beating them

[raven stone]

@echo aurora ?

[desert pendant]

Mimo

[echo aurora]

Recently some models have been removed from Direct and Side by Side mode. This was done to help ensure reliability and availability of Arena in the long term.

[compact flame]

Dang didn't expect gpt 5.5 to be 5 dollars more expensive that Claude

30 dollar output

[desert pendant]

It is good but you need to specify what you want it to do

[light sleet]

@surreal zephyr look at this guy

[light sleet]

Is it a high chance of getting 5.5 in battle

[loud herald]

Guy decided to call for backup

[light sleet]

And backup for what exactly?

[compact flame]

But gpt 5.5 is better in my experience though

[silent tree]

same guy whos flexing his ass looking ui btw ^^

[light sleet]

looks like I already do ig..

[south ether]

how long do we think until we get a preliminary arena score for 5.5?

[loud herald]

You're pinging the GPT glazer because someone said opus is better 😂 thats just starting an argument

[light sleet]

Same

[light sleet]

I could've pinged him for various reasons and that's your pick?

[loud herald]

Absolutely zero correlation, you're spewing nonsense because you have no base to speak on

[compact flame]

I guess we can't change Claude fans opinions

[light sleet]

no wayyy

doesnt that mean ur criticizing him

and creating an argument yourself

[loud herald]

100%, but theres no shame in it

[light sleet]

Noire thinks it is

[loud herald]

If you like a model then you like a model

[light sleet]

https://019dc0c1-6360-79cc-8212-9505e03b55b8.arena.site/

[light sleet]

Yea so

[loud herald]

Idk why it matters to others what you do and dont like

[compact flame]

For me it was worse no glaze

[light sleet]

That hurts your feelings?

[loud herald]

Who said that?

[primal orbit]

https://i.snipboard.io/AIrtnY.jpg

😢 😢 😢 stuck

[loud herald]

Its actually wild how much people assume on this

[silent tree]

same guy who flexed his ass fake ui exploit btw ^^

[silent tree]

slow ahh guy

[echo aurora]

Sorry to hear that! Give these steps a try: https://help.arena.ai/articles/8691588590-troubleshooting-infinite-generation

[light sleet]

well nice, but it definitely hurts noires feelings

[loud herald]

Idkkkk mannn you arent following the rulessss!!!

[silent tree]

well atleast i dont create exploits

thats against the rules too

hm

[loud herald]

No its not LOL

[silent tree]

both the server and the game youre doing it for

[light sleet]

I wouldnt call myself a glazer

[silent tree]

ok lets exploit on games and ruin it

[compact flame]

It just depends on what tasks you use it

[silent tree]

lmao

[loud herald]

Where in the server does it say I cant talk about AI making game exploits? maybe I need to read through it again

[compact flame]

Opus is bad in it's own ways

[unique mural]

how can i gen ai videos tho

[compact flame]

Gpt is too

[surreal zephyr]

skill issue

[silent tree]

TUFFFF

[surreal zephyr]

opus 4.7 extended fails this

[light sleet]

I'll leave it to yall

War starting here

[surreal zephyr]

gpt 5.5 low passes

xD

[silent tree]

dude even 5.4 xhigh matches mythos

LMAO

"Mythos can't be released to the public"

[compact flame]

For me it was bad

[unique mural]

i thought there s a feature in this server for us to generate ai videos

[surreal zephyr]

and much better at prompt adherence and half cheaper

[loud herald]

Not needed argument*

[grave peak]

We know son

[silent tree]

what if mythos was just a fake theory created by anthropic

💀

[grave peak]

For what?

[compact flame]

What if mythos never existed

[loud herald]

Mythos being breached tells you how real it is

[compact flame]

What if mythos was just a man that pretended to be ai

[silent tree]

yet 5.5 is mythos level but beats it in some points too

[loud herald]

And how bad of a system it can actually make

[silent tree]

spud mogs

[grave peak]

Firefox just fixed bugs with him

[silent tree]

like your game exploit

LMAO

that's a nice example

your game exploit

[compact flame]

I mean who cares about hallucinations?

All models have those

[silent tree]

fr

talk about the real coding

not hallucinations

nobody gaf about hallucinations

use another AI?

Is it that hard?

[grave peak]

Ok

[silent tree]

u dont have to use same ai for everything

thats stupid

its based on your task

[grave peak]

I want to use jipiti 5.5

[silent tree]

not for everything.

[loud herald]

Hallucinations give incorrect information and could ruin a project of yours

[compact flame]

I generally use Gemini flash for tests

[silent tree]

same

[shadow zodiac]

ai has purpose when u learn to make it use its purpose

[loud herald]

Speaking of hallucinations

[silent tree]

and then you realized you got charged of 50000$

[grave peak]

https://tenor.com/view/gem-alert-gif-14590606955304525995

[silent tree]

💀

[obtuse heart]

Just use sonnet..

[compact flame]

Sonnet doesn't understand Russian well

[obtuse heart]

Oh, fair

[undone saffron]

I wonder when gpt 5.5 codex

[silent tree]

same for my language

watch how claude users will call me worshipping sam altman in the next 3 minutes

same thing jalapeno did yesterday and leaked his age but ig hes good

and the random kid

[grave peak]

Ok sam

[silent tree]

but let's not talk about it again.

yeah youre gonna say I worship sam altman 💀

ok?

Idgaf man

[compact flame]

If opus 4.8 comes out it will be more expensive probably

[silent tree]

your decision

See

Exact point.

All anthropic people r the same

they're gonna say I worship sam

Lol

[grave peak]

No bro you don't worship him

[compact flame]

Though they gotta chill with 5.6 or etc etc

I don't want to see

[silent tree]

Leha said so.

[compact flame]

Chatgpt 6.7

[obtuse heart]

Can they just release like an open source model, that's both better than opus 4.7 and gpt 5.5

[silent tree]

dude idgaf

[fiery gull]

Qwen3.6 27b no thinking>>

[compact flame]

Some people say opus 4.7 is worse than opus 4.6 in some aspects

[silent tree]

and claude doesnt have images

so sad

[compact flame]

Apparently degradation

[sterile tartan]

Mainly 50% more expensive yeah

Well they had atleast

Not anymore

[south ether]

gemini 3.1 pro > gpt 5.5

[compact flame]

They had sora well it's gone

[pseudo hemlock]

oMg @echo aurora wHy iSN't gPT-5.5 iN diReCT cHaT?

[silent tree]

does anthropic have?

[compact flame]

Because sam Altman said it's too elite to be used for free

[pseudo hemlock]

dang it man

guess ill just never use arena again!!!

someone tell me when they're giving away free money!

[sterile tartan]

Goodbye...

[echo aurora]

Hopeful the new usage system will be able to help with this!

[compact flame]

It's obvious that google has everything

It's google for a reason

[sterile tartan]

I hope it doesn't make it worst atleast

[silent tree]

The new usage pineapple system

[icy frost]

i hope it will have a dupe glitch

[echo aurora]

It won't.

[silent tree]

@echo aurora

[echo aurora]

Don't spam pls

[compact flame]

You worship Claude!1!!1!1!1

[silent tree]

Can we get this guy muted?

[sterile tartan]

In Pineapple we Trust.

[silent tree]

why do I face this everytime from claude users

"You worship sam altman"

[compact flame]

Claude is opus

[sterile tartan]

They are very Beta Males

Claude is Slopus

[silent tree]

Pop Us

[loud herald]

Its weird, claude is like the base name if you want to say any model like sonnet opus or haiku

[echo aurora]

Kk this is becoming unproductive, going to ask to move on from this convo.

[sterile tartan]

Good One

[compact flame]

Do people even use haiku?

[silent tree]

and tell these people not to call me a sam altman worshipper

[icy frost]

can i have the beta test pls'

[pseudo hemlock]

Sometimes

[loud herald]

Dude honestly no clue LOL, I have legit never seen any benchmarks or reviews on haiku

I've never used it myself

[pseudo hemlock]

I used Haiku like 5 mins ago lmao

[silent tree]

Hi Mr. VIP

[loud herald]

Dang

[icy frost]

g

[silent tree]

haiku sounds anime

[grave peak]

Can chatgpt 5.5 fix this?

[pseudo hemlock]

I just had it unzip a .zip and give me all the contexts from the files bc I needed it in text and didn't want to write a script for it

[sterile tartan]

Just like Waifu

[loud herald]

I see

[silent tree]

just like ai anoscranel tuff edition 100% 🔥 🔥 🔥 🔥

[pseudo hemlock]

Also i would've used sonnet but ...

im about to have no more claude 😭

[silent tree]

But then pineapple attacks aianos cranel

[sterile tartan]

https://tenor.com/view/orange-cat-laughing-gif-13031147940704744720

Reverse i am

[loud herald]

Yea uh unfortunate

[sterile tartan]

https://tenor.com/view/nalog-gif-25906765

[icy frost]

bro is paying for invisible ram

[sterile tartan]

https://tenor.com/view/riddler-paul-dano-new-rockstars-thumbnail-does-he-know-he-doesnt-know-gif-27552850

[pseudo hemlock]

NOOOOOO

MY SOCIAL CREDIT

[sterile tartan]

It's okay.

[pseudo hemlock]

there goes 4 generations of my family

🪦

[sterile tartan]

https://tenor.com/view/social-credit-credit-social-уважение-плюс-уважение-gif-1626328442317885176

[pseudo hemlock]

WE'RE SO BACK