general

[ocean vortex]

So this could go both ways

If they wanted or felt threatened by competition in any way, 128k output cap and tuning for that would have a nice performance boost over 64k I believe

It's a good model in how it manages to get the performance with less tokens than any competing model currently, but there's untapped margin nevertheless to go beyond

[frosty lava]

so gpt 5.3 codex is almost half the price compared to 5.4 ? and honestly the difference for coding is not so big, so you can legitimately take this ~2x usage

[opaque crescent]

you guys could test opus 4.6 on ondemand but its only for a few messages (til it hits 1m context im p sure)

[fickle forge]

And it's not difficult.

[ocean vortex]

5.3c is near equivalent to 5.4 in coding

[frosty lava]

yeah and the price diff is big

[ocean vortex]

task dependant marginal differences there, they are simply different

[frosty lava]

you can use 5.3 codex with fast mode and you'll just use a bit more than if you used 5.4 without fast mode

if you want to, or you can also use it in normal mode and have more usage

[red chasm]

GPT 5.4 was brought back, but only to the old dialogues.

that is, in those dialogues in which 5.4 was used, you can use it

I don't know about Gemini

in a normal, more or less new dialogue there is no 5.4

[dire gulch]

seems like a ui bug, the endpoints are closed

actually no ur right, but only some of the old chats for me

[sullen creek]

whats a endpoint

[dire gulch]

by endpoint i meant basically the way to access that specific model, i was guessing they closed it but they didnt

[modest topaz]

[vernal raft]

because its not intended to

[burnt kettle]

bro bring back claude opus

[sterile tartan]

Probably not until next gen models are Cheaper

[burnt kettle]

vruh

[sterile tartan]

The problem is Compute is not Free like Software

It takes Computers Infrastructure, Electricity and Water

Maintanence and such stuff as well

[burnt kettle]

mmh i see

[glass pendant]

Hi

[onyx python]

What's the daily limit for image generation?

[ivory latch]

is any way where we can use claude opus in free ?

[velvet furnace]

I use GLM5.1,it looks good at coding

[silk dew]

what

muse spark rank 4 overall?

is it actually good

[vague quarry]

didnt test it out but probably could be replacement of the top 3

[silk dew]

still cant move on

they removed my gemini 3.1

for me, its atleast close to opus

my favorite model tbh

[calm lagoon]

I really like Muse Spark, I'm so glad they added it

[silk dew]

is it good

?

[quiet bolt]

#Generate a video

[echo arch]

Where do you find this?

[calm lagoon]

Yeahhh, smart🧠

[light siren]

@echo aurora

[uneven peak]

😭

[hollow mulch]

Im try log in again even log out and log in back but still get that issue when try oppen a chat 🙁

[light siren]

find whats odd in the image

[echo arch]

Claude Sonnet is 19th in the global ranking

it's a mess.

[light siren]

erm so?

[uneven peak]

The one that you ask him it's odd 😭

[light siren]

incorrect

try again

[uneven peak]

First one 😭

[light siren]

wat

[uneven peak]

Extra hugs is sus

[echo arch]

Can you explain to me why gemini-3.1-flash-lite-preview is worse than gemini-3-flash, a seemingly lesser version?

[light siren]

ur soo delusional- it says LMArena instead for Arena

[light siren]

cause google is bad at making ai

you would think they would be good with the amount of data they possess

[uneven peak]

I'm confused

[light siren]

look closely at the top left of the screenshot

[uneven peak]

Ohhh it's Fake or Ai image generated

[light siren]

no

it's just my Arena Fixes extension

[echo arch]

But is gemini-3.1-flash-lite-preview actually worse than gemini-3-flash? Or are the data wrong and is 3.1 actually better than 3.0?

[light siren]

that fixes arena bugs, I made it bring back lmarena

[uneven peak]

So you have opus 4.6 in direct in chat?

[light siren]

the hell, no

no amount of changes from me could bring back models

I said bug fixes dude

[uneven peak]

Why did you want to say cute lines for your Lil sister? SUS

[light siren]

wdym say cute lines for my lil sister, she's tryna be a voice actor and wanted cute lines

[uneven peak]

You right but I got one for her I can give

[light siren]

get out.

[uneven peak]

It's really good line for her

[light siren]

Not interested.

[uneven peak]

Noo not wired i saw that in tik tok

[light siren]

fine what is it

[uneven peak]

• ~UwU~ ~my name is Senpai don't leave me please~ 🥺 ~Uwu~

[light siren]

she's a minor.

[rigid pasture]

...

[uneven peak]

Who

What?

I didn't

[rigid pasture]

I just woke up and this is the first thing I see here

[uneven peak]

That was my The Proposal

[uneven peak]

I'm minor too lol

[vague quarry]

wait...you can now click role ping

[lofty frigate]

Is it me or nano banana pro 2 messes up your actual face when generating photos

[short sluice]

ok muse spark is actually good

im using arena just for it now

[verbal marten]

Muse spark is actually good bro thanks god

[vague quarry]

yeah but im using on meta ai, i think it has no message limit

[outer flicker]

'Session not found. Redirecting to home...' what does that mean guy?

[short sluice]

its nerfed on meta ai but on arena it seems fine

it wasnt anything tou-breaking or copyrighted either

[vague quarry]

what was it

[vague quarry]

what did you say?

[short sluice]

i closed both tabs already

[verbal marten]

But only 128k token

[spring oar]

system prompt is like instruction ? or is different ?

[verbal marten]

Not good as opus 4.6 but still good i think

[gritty trench]

Opus 4.6 thinking is back ! But seems to be the only one...

And now it disappeared

[deft spruce]

gemini3.1 is back

[vernal raft]

It is not it's a bug

None of those are bsck

It's unclear if it's a UI bug, or something else

[deft spruce]

ok i think that was a bug a UI bug

[rigid pasture]

it happened to me too

its a bug

[gritty trench]

And yet, the answer generated (19 s., about physics and philosophy) was at its level, seems to me at least. Which makes me wonder...

[willow swift]

is thinking disabled on claude-sonnet-4-6

?

[obsidian cargo]

sometimes I feel thinking is disabled on this server lmao

[golden ocean]

REAL

[fast fable]

do lmarena is going to not free now?

im cant use opus4.6 or pro3 anymore

[sterile tartan]

You can use Pro 3.1 on Ai Studio

GPT plans are selled for cheaper on some websites as well

[short sluice]

yes

(unfortunately)

[spring oar]

what you mean ?

[atomic pagoda]

Umm guys where is Claude opus on the site like I haven’t been on here or the site for a while and that’s the only one that I use there so like how do I find it

[obsidian cargo]

only available randomly in battle mode for now

they're gonna roll out a free daily credits system soon and bring access back

[atomic pagoda]

Bruh that sucks like I knew this was gonna happen at some point but are there like other sites that have Claude opus

[vernal raft]

still undecided

they did not confirm those models will be back with credits in side/direct

[atomic pagoda]

Any other sites that have opus for free because that’s the only one I need

[fast dew]

Uhh why else would they be adding it

[vernal raft]

no

[vernal raft]

becase as arena gets more and more popular the cost keep increasing

[atomic pagoda]

Well I’m guessing eventually they’re gonna add that you can buy more credits and then it’s gonna be like most other sites I’ve seen like you can use it for like a few messages and then it’s just pay to use

[fast dew]

No they adding it cuz they wanna bring the top tier models back in a more sustainable way

This is the more sustainable way

[atomic pagoda]

And eventually you’re gonna have to pay to actually use it, damn the world sucks, everything has to be payed a ridiculous amount of money for

[vernal raft]

can you find me the sentence in this discord or w/e that explicitly says those models will be back with the new system?

because i havent read

[fast dew]

its literally obvious son

[vernal raft]

thats not how it works ahahhahahhaa

nothing is obvious

until confirmed

[atomic pagoda]

Well if they were back they would be extremely limited right

[fast dew]

They're not gonna bring a credit system cuz arena is getting "more and more popular"

That makes no sense lol cuz why would they still have the battle system/direct chat up

They wanna bring back the top tier models

That cost more

[fast dew]

Everyone getting free credits so i dont think so

[atomic pagoda]

Well the free credits are gonna run out and then it’s pay for more

[fast dew]

I think its per day

[atomic pagoda]

You really think this is gonna actually remain free forever

[fast dew]

Yh

[languid crescent]

I still prefer the present, I don't like the system their tryna implement... feels restricting

[languid crescent]

hope so

[vernal raft]

i mean

more than this

[fast dew]

Yh hes gonna say that but they're most likely gonna bring it back

[atomic pagoda]

You have too much hope in this because it usually goes this way they implement more restrictive ways to use it for free and then it’s you gotta pay for more

[fast dew]

I don't think they gonna make people pay

Arena a free website

[vernal raft]

we ppay with our prompts and vote

[languid crescent]

free but restricting

[vernal raft]

but those are not enough

[languid crescent]

day by day they gonna add more limits

[vernal raft]

anymore

[atomic pagoda]

Well there is probably gonna be a limit for free like you get x amount of credits for free and then if you want more you have to buy it

[languid crescent]

i like whats current, i don't like this

[fast dew]

So u like not having any premier models

[atomic pagoda]

I like how it was before like when you could just use it and get a limit after a lot of messages like that was good enough for me

[languid crescent]

I like having premier models, in fact i like using all models. It's just that the concept of "free" but limiting it is conflicting...

[atomic pagoda]

There are other websites that have AI models and credits and you have to pay for more after a bit

Like that’s what’s gonna happen to this too

[languid crescent]

def

100$

100%

[fast dew]

Thats different tho this is an arena

They not gonna make people pay for voting models

[atomic pagoda]

It’s actually sad because this was great like you can use it a lot for free

[languid crescent]

what about Direct Chat?

[fast dew]

We're still giving them data

For the models

[languid crescent]

for free

I am okay with that but they are removing that privielge

[fast dew]

Yh

No they not lol ur getting free credits

[languid crescent]

But what about the announcement that they made? They adding more credits

[atomic pagoda]

Bro just accept that this is eventually gonna happen because it already is like they already took the good models and now they’re bringing credits and next is pay for more credits

[obsidian cargo]

I'd pay for more credits

[languid crescent]

exactly

[fast dew]

Lowkey i would too

[obsidian cargo]

depending on the price though I guess

[languid crescent]

why not just keep the old models free, while premier models have credits instead

[obsidian cargo]

they probably will keep old models free

[atomic pagoda]

That’s the point and honestly I was hoping it would go more like you can pay to use this model but it’s not an extreme amount of money and you don’t have to pay monthly but like once like I know there was a site that did that like you gotta pay like 10$ to use the model indefinitely I think

[languid crescent]

Then its not free anymore...

I read somewhere in their terms that they will provide their services for FREE

[atomic pagoda]

Well that’s the point, nothing is

[languid crescent]

it just feels conflicting lol

[fast dew]

Technically it is still free

With the credits ur getting for free

[languid crescent]

as of now yes, in the future? paywall.

[atomic pagoda]

I know but they are gonna make it like everything else that is “free” like it’s technically free to use for a little bit but then it’s pay to use for real

[fast dew]

Ur just assuming tho

[atomic pagoda]

What do you think is gonna happen cause it’s obviously heading towards that

[fast dew]

Js the credits system and thats it

[atomic pagoda]

It would be great if they brought the good models back and gave credits that regenerate after some time and that’s it but I doubt it because they already removed the good models from direct chat

[languid crescent]

shi happens i guess 🤷‍♂️

[whole sundial]

you forgot the part where it says that they reserve the right to make users pay for the service or parts of it

[languid crescent]

hmmm, i didn't see this

[atomic pagoda]

It sucks, but does anyone here know a site that has claude opus for free to use like anyone

[languid crescent]

FEES AND PURCHASE TERMS. Company currently offers the Service free of charge. However, we retain the right to charge for the Service, or any features or components thereof.

[fast dew]

Theres none

[languid crescent]

damn

[deft spruce]

FEES AND PURCHASE TERMS. Company currently offers the Service free of charge. However, we retain the right to charge for the Service, or any features or components thereof.

DAMM im late

[bright shard]

@echo aurora I need some help with this: when I send a request and the reCAPTCHA verification says it's infinite, it never finishes and the message ends up saying "error due to not completing the verification".

[languid crescent]

https://klipy.com/gifs/theres-nothing-we-can-do-1

[red sigil]

Are some video model available on discord here

[whole sundial]

video generation is no longer available in the arena discord server, it is only available on the arena.ai website here: https://arena.ai/video

[red sigil]

If I want to use specific AIs and not battle mode

Please tell me what should I do

[atomic pagoda]

Go to direct chat or of the model isn’t there try and find another site on google like I’m doing now

[whole sundial]

you can't, it is battle mode only

[night moat]

Side by side and direct

[deft spruce]

hows openclaw?

[sterile sun]

@echo aurora gimme ur juice

[wicked sage]

what the hell

LOL

[ocean venture]

https://tenor.com/view/omniman-gif-26206182

[bright kayak]

#3 is impressive

[sweet comet]

Damn GLM

[whole sundial]

congrats to glm 5.1 for being the best coding model you can actually use on arena in direct and side by side modes!

[gray lion]

wow

glm is goated

[echo aurora]

Hey @bright shard sorry to say is a known bug. Team is aware of this and are working on a fix. Unfortunately, there isn't much on the user side to get past this.

[silent tree]

@echo aurora release gpt image 2 in direct chat or larry is coming 😡

[solemn osprey]

is glm actually that good

[deft spruce]

mmm....

glm

[formal delta]

Same question lol

[deft spruce]

i didn't use that

but i have to use it

[exotic tartan]

No way glm is that good..?

[solemn osprey]

since you are voting on what the web app looks like for the most part it might be skewed towards how well it can do react/web design and not general swe

[gray lion]

it is pretty high

[rose pike]

Where’s GPT image 2?

[sullen pollen]

removed

[rose pike]

Oh no. Why

[solemn osprey]

i wish there was like a gallery feature so that you see what each of the models' generations look like

[silent tree]

expensive

[gray isle]

can't even do a proper PDFLATEX. there's no way, that's the score of that thing. 🤦‍♀️

[surreal zephyr]

gemini cooking, glm trash. expectable

[rose pike]

Sounds fair

[surreal zephyr]

not out yet, demo ended

[gray lion]

7th place is not trash

[sullen pollen]

too high costs people abusing it

[surreal zephyr]

bad score, speed

iq

[pale obsidian]

5.1 is absolutely insane

[gray isle]

it is completely trash, didn't serve my request at all. always errors all the time compared to Kiwi or Sonnet.

[deft spruce]

so.....GLM 5.1 is GREAT? (but not a than a OPUS?)

[gray isle]

it's like it can't even follow a hard prompt

[gray isle]

It's below GPT 5.4 High in my opinion, unless it doesn't get any errors at all. i would rate it maybe in between pro preview and sonnet

[muted flower]

why is arena's sonnet 4.6 so different from claude's sonnet 4.6?

[silent tree]

My GPT IMAGE 2 In Arena Direct Chat Release Prediction

- Monday, April 13

- Wednesday, April 15

- Saturday, April 11 (Tomorrow) (Highly Unlikely but Yeah)

[signal pelican]

what the hell? How is GLM 5.1 is better than Gemini?

[solemn osprey]

people voting based on vibes

[signal pelican]

yeah no

they has to be abusing the voting system

[gray isle]

or some random prompt they do, that isn't really useful. like creating a roleplay story for no goddamn reason

[light sleet]

It's Sunday 100%

Gpt image 2 drops Sunday

[silent tree]

It might be tomorrow

[rugged abyss]

To be fair, Open Source Models are getting better, the chinese labs have really intelligent workers to improve rapidly, you guys should check out GLMs Twitter.

[signal pelican]

"batter"

[rugged abyss]

my bad

[floral canyon]

most chinese models just use destilation

and they copy claudes' outputs

[rugged abyss]

Destillation isnt that huge anymore. I think they use private code training data from china as these are usually not available for outsiders.

[signal pelican]

I have used GLM5.1 and I can ensure you that it's not better than Gemini, not even close. It gave me more problem by not following the instruction and more often had trouble processing its own agency in batches.

it's just absurd right now

[floral canyon]

kimi still uses it

it claims it's claude

lol

[rugged abyss]

All chatbots sometimes mentions each others names, for example I think it was ChatGPT calling itself Gemini in french or something, there are lots of these cases

[floral canyon]

Kimi for sure did use destilation

Deepseek team admitted to it too

it's not a secret at all

[rugged abyss]

Yeah no doubt distillation is part of it, but definitely pure distillation

[floral canyon]

Honestly idk how well they could do without it

they're playing on hardcore compared to the usa

less data and weaker hardware

[rugged abyss]

If you think these models are just distilled copies, you're ignoring that they're training on massive, proprietary datasets Western models don't have access to, while being restricted by hardware sanctions to innovate more efficient architectures than the probably bloated models that we have rn

[queen veldt]

Yo the muse spark is actually good

[obsidian cargo]

I wish. but major releases never come on fridays or the weekend.

[floral canyon]

uhh in kimi's case the code performace is alright

but it def would be way worse if they didn't do it

[silent tree]

Nvm, Gpt Image 2 is confirmed at Tuesday!!!!!!

[floral canyon]

china has a lot of data for sure but i wonder how much of it is just them spying on citizens lol

[light sleet]

😭 😭

[rugged abyss]

Lets be honest tho, Open source is improving way faster for what they invest though, of course a bit of stealing and distilling is part of it, but i wouldnt be surprised if we have claude mythos performance at the start of 2027 or end of 2026, but opensource

[floral canyon]

and the models definitelt did not receive all the data considering the censorship

[floral canyon]

isn't mythos like 5x more expensive than opus

[gray isle]

Me not needing to know who is this Assistant B. It's literally GLM lol

[obsidian cargo]

source? last time someone said that it turned out to be false lmao 😭

[floral canyon]

it's not only about the best model but who can run it too

these models burn money to answer simple emails

and mythos is mostly hype for now

[silent tree]

I meant in arena, because gpt image 1.5 in arena also dropped at a Tuesday

[floral canyon]

trust me bro benchmarks and lots of weird ideology put into the spec

they hired a psychiatrist or smth

these mofos really play dumb and act like they can't be sure if it's sentient

[light sleet]

It could release in the weekend too who knows

[rugged abyss]

I doubt its just hype, of course benchmarks dont reflect real World performance entirely, however the jumps there are insane and from what results have been showing its quite good.

[floral canyon]

gpt-2 was 'too powerful to be released publicly' as well

and it was 2019

anthropic is probably going public

they just leaked the source code of claude code using their frontier mythos model

lol

[rugged abyss]

Yeah but now we know more about AI than ever before you know? Its like if you eat a new kind of food its gonna be great, but if you eat more variety of that food you can tell whats best and what isnt

[floral canyon]

idk man it just looks like hype

there is no evidence for us to test it anyway

this technology needs hype to survive otherwise it will die

[rugged abyss]

But lets be honest, well just get a lobotomized version

[floral canyon]

its just a cash burning machinea at this point and no one found a way to reliably earn cash off of it

[rugged abyss]

naah, ai is too big to just die, its here to stay, no doubt about it

[floral canyon]

i didn't say disappear

[indigo knoll]

When will Deepseek V4 be released?

[floral canyon]

i just think the hype didn't meet any of the expectetions

[rugged abyss]

Well you kinda said its gonna die

[floral canyon]

excpectations*

[light sleet]

Opus 4.6 released Friday in arena

[floral canyon]

i meant the hype

literally every 6 months we hear it's gonna replace people and it didnt for the past 3 years

investors have their patience too you know

and the truth is that AI works great until it doesn't

its just a tool with lots of hype and fake promise of AGI

but it does have a lot of potential

and some people already make great things with it

so die, no but the hype wears off

[gleaming roost]

I find it strange that some models are rising in the rankings while we no longer have access to others, so I question the validation of the data and the positions in the ranking

[rugged abyss]

Yeah but keep in mind, one of the biggest issues is probably long context windows, and these are growing insanely fast. I think once this is resolved we can actually see some jobs be minimized. Im not saying that they will literally take all our jobs but will be working alongside us

[rugged abyss]

Well we dont have a central definition of AGI, so its hard to know

[floral canyon]

nah man its just cringe hearing them talk about agi

its just token prediction and nothing else

[obsidian cargo]

yeah but thats because anthropic likely released opus 4.6 the day before, on a thursday. If OpenAI released gpt-image-2 on a thursday, arena would likely have it on a friday, but openAI wouldn't release gpt-image-2 on a friday or the weekend.

[floral canyon]

its not about the context size only

[rugged abyss]

I think its mostly openAI

[floral canyon]

top models have 1m tokens and they still 'forget' things in the middle

[rugged abyss]

Never said its only that, but its still a great issue

[silent tree]

Openai hasn't released gpt image 2 yet? I thought some people were having access to it in the chatgpt app already?

[obsidian cargo]

its not officially released, yes

[floral canyon]

No. the mythos spec has that narrative. They release this paper every year and act dumb like its surprising them

[obsidian cargo]

no announcement or anything

[silent tree]

So the API isn't too

[floral canyon]

not really its just not reliable 100% and never will be

and you don't get consistent results

[obsidian cargo]

yeah there's no gpt-image-2 api yet

[rugged abyss]

scammer ahh

[silent tree]

Well since the Demo is over, we might expect gpt image 2 to be officially released in the next few days.

[obsidian cargo]

gpt-image-1.5 dropped on december 16th, which was a tuesday

but there's no guarantee gpt-image-2 will also drop on a tuesday

they just dont do major releases on friday/weekend so that people can be around to fix it if things go wrong

[floral canyon]

honestly so far what we got from AI is scammers using it to scam older people or governments create propaganda lol

or artificial OF models

[rugged abyss]

Naming only the bad parts while ignoring the benefits🥀

[floral canyon]

and social media somehow got even worse

[ocean venture]

just go to the official discord server of openAi bro

[floral canyon]

i meant the image generation and video generation

[light sleet]

what

[floral canyon]

but im curious what benefits we got from sloppy videos

[rugged abyss]

Image generation definitely has its purpose, video generation, I myself dont like it

[obsidian cargo]

I can't, I got permabanned from there back in the dall-e 2 days because I tried to generate a picture of a portal made out of meat, and they wouldn't let me appeal

[whole sundial]

video generation is no longer available in the arena discord server, it is only available on the arena.ai website here: https://arena.ai/video

[ocean venture]

https://tenor.com/view/ishowspeed-laugh-clap-gif-18203230752197306459

[floral canyon]

what are the benefits of video gen tho. must people use it to create fake narratives or straight propaganda

[rugged abyss]

Id say like using Nano banana 2 to create professional looking product images is a great usecase for example

[whole sundial]

image generation is on the site too, just not gpt-image-2, at least not yet

[floral canyon]

or maybe OF type content

and the fact it costs a ton to just use these models

i don't think its noble at all lol

[ocean venture]

maybe one of the mod things that was a gore lel

[rugged abyss]

I dont use Video generation myself, but I do think it has its use cases to create good editing faster, mostly talking about VFX or similar. Also like previz cases so customers can see what a engineer is proposing faster, without any extra costs. Or for example, education purposes. If AI video gets good enough to create working simulations or showcases of complex physics examples for students to visualize them better, I think this has its purpose too

Even though im not using it myself

[floral canyon]

I'm not sure what physics simulations you mean but sure

i feel like it has niche use cases and 99% people use it for slop

sora was discontinued

the most useless product ever created that burned 15m $ daily

[rugged abyss]

I meant more like explanation videos to visualize examples, in this case phyisics cases. But I do get your point how its being used for ai slop videos like those weird fruit videos

[floral canyon]

yeah

but AI is def gonna be useful in war

already is

USA must be very happy about it !!!

[gray isle]

isn't AI already in existence before it became Consumer/Commercial

[whole sundial]

this is funny though, every time this happens the bot does this

[echo aurora]

Lol sorry about that

At least when they're actioned their ticket is also removed

[solemn osprey]

😭

[echo aurora]

A Trace ID didn't appear?

[vagrant temple]

how can i fix this error in antigravity?

[solemn osprey]

i refreshed the page but it got stuck at reading the directory structure

[quartz pike]

I LOVEOPENMODELSILOVEOPENMODELS so muchhh

[echo aurora]

I think what is happening is you were getting the infinite generation bug initially https://help.arena.ai/articles/8691588590-troubleshooting-infinite-generation. If this continues, can you create a post in #1343291835845578853 and provide more details there about this?

[solemn osprey]

could it be also because it genuinely has to produce a lot of code?

[thick blade]

Is arena back

[civic plaza]

Muse spark seems very good. But damn the generation takes time

[remote saddle]

hi

[vagrant temple]

yes but this model is very sht by simple tasks

[solemn osprey]

probably happening again

[frosty lava]

for me the most frustating thing about arena right now, its in coding when i give it a task and it simply will either hit a memory limit and have to simplify it, or something went wrong when its only the first prompt of the day i give it !, and that leads me to never being able to actually build something !

i have to try 6 time the same prompt and on different account for one of them to work

[solemn osprey]

it would be nice to know if the context limit was reached

[vagrant temple]

at this point change your browser

[vagrant temple]

.

[frosty lava]

im simply on google it should work

chrome

sometime i find a way to fix it but it potentially also reduce the quality of the work, its to explicitely say to do the task in an efficient way, that won't hit the memory limit, (don't do too much postprocessing or thing like that)

but then the result is just not as good as it would be usually with the model

so there is no great solution

[solemn osprey]

whatever this model is in code arena can generate a working gameboy emulator from scratch in one shot which is pretty insane

[light sleet]

Yea that's a gpt model

[frosty lava]

which model was it ?

[solemn osprey]

lol of course it happened again

[whole sundial]

it has to be spud, it should come out next week

[light sleet]

It's Claude Pineapple Ultra bro

[solemn osprey]

3.1 Pro

[light sleet]

bro it's claude pineapple ultra

[feral kernel]

how do you have that

[frosty lava]

no but.. i wanted the codename

[whole sundial]

anthropic has never tested a single model on arena before release, there is no way that it's claude

[solemn osprey]

battle mode, just came up

[light sleet]

claude pineapple ultra bro

[solemn osprey]

https://019d783c-a2ca-7136-9dc9-9599b4318468.arena.site/

[vagrant temple]

ho?

[light sleet]

Don't you remember? It dropped last week by @echo aurora collabing with anthropic

2 days until pineapple juice goes back to his home in Arena Parallel Universe :(

And the Arena pineapple comes back

[vagrant temple]

where is opus 4.6 and gemini 3.1 Pro?

[light sleet]

On a vacation bro

they wanna chill too

😭

[solemn osprey]

hidden

[light sleet]

vacation *

[solemn osprey]

you can only get them in battle mode

[light sleet]

they're in hawaii

[whole sundial]

hmm i got this model

[vagrant temple]

unfunny bro

[light sleet]

damn nice

[echo aurora]

There were some models removed from Direct and Side by Side recently, more info can be found here: #announcements message

[light sleet]

not meant to be funny tho

[civic plaza]

Yeah I noticed

[light sleet]

Cuz they're actually in a vacation type and will come back with new stuff

[echo aurora]

Would you mind sharing these Trace IDs in #1417174113092374689 , I'll take a look in a bit, but worried this will get lost in #general

[vagrant temple]

thx for a serios answer

[echo aurora]

What should happen after?

[light sleet]

pineapple care to explain this?

[light sleet]

the normal pineapple comes back

[dim oracle]

literally joined this discord to search if anyone else was looking for scorch as well hi 😭 https://www.reddit.com/r/lmarena/comments/1shrvq8/scorch_triaging_mystery_model_perhaps_mythos/

[echo aurora]

No problem. It's important to note these models are still in Battle, and that we do have the intention to bring them back to Direct/Side by Side when we can in a sustainable way.

[echo aurora]

hi

[light sleet]

I predict gpt image 2 dropping somewhere around Sunday or Tuesday OR Thursday.

😈

[echo aurora]

Yeah the context limitation is a frusterating limitation to deal with at the moment. We're hopeful the new usage system we're exploring will be a helpful step to make this a better user experience.

[light sleet]

yo pineapple wanna play guess the ai logo

(guess the AI name by its logo)

[vagrant temple]

nano banana pro will be stil better

[light sleet]

gpt image 2 is better than nano banana wym

[vagrant temple]

hahahaa

[vagrant temple]

gpt glazer?

[light sleet]

since when did I say I'm a "gpt glazer"?

it's my opinion

[vagrant temple]

ok

[obsidian cargo]

gpt is kinda ass overall but gpt image 2 is better than nano banana, wonder how long it'll take for nano banana to surpass gpt image 2?

[surreal zephyr]

which model dude

[solemn osprey]

Gemini 3.1 Pro

[surreal zephyr]

nah it literally is better, but its not out

gpt i2>nb2> np1p > gpt i1

[solemn osprey]

Gemini models are good but you have to steer them to not be lazy

or else they will take the easy way out

(talking about code)

[vagrant temple]

true

[orchid olive]

sonnet 4.6 or glm 5.1

which is better

[vagrant temple]

for what?

[orchid olive]

for coding

[vagrant temple]

glm 5.1

but opus 4.6 is better then glm5.1

[golden ocean]

anthropic needs to put claude mythos into arena as anonymous model

[vagrant temple]

nah its to powerfull

[slate hare]

@echo aurora I want to ask will you return Gemini 3.1

[rustic geyser]

Whats the best model currently that can be used in Direct.

[solemn osprey]

muse-spark?

[vagrant temple]

for what?

[rustic geyser]

Ig for text. So not coding or images like that

[orchid olive]

I have tired the glm 5.0

The AI spent a whole half hour trying to figure out whether the bug was caused by user behavior, rather than admitting that maybe the code it gave me was wrong

[rustic geyser]

I tested it. Its so slow I think

[vagrant temple]

its only good for fronted coding

[golden ocean]

https://tenor.com/view/ева-елфи-gif-7266118905556669339

[rustic geyser]

Is opus the best for backend

[vagrant temple]

yes

[rustic geyser]

Either way opus isnt in direct anymore 🥲

[orchid olive]

opus is the best for everything

[solemn osprey]

wish there would be a gallery for "same prompt but different model" to see how each model responds to a prompt

[vagrant temple]

u can use antigravity with multi google accs

[vagrant temple]

.

[rustic geyser]

True

Ig I might do that

[solemn osprey]

dont they rate limit you in an insane way

[golden ocean]

Claude Opus 4.6

[solemn osprey]

like 1 prompt and thats it

[rustic geyser]

But the limits are weekly i think

[vagrant temple]

yes but u can use multi google accs

[solemn osprey]

how long until you hit the limit?

[rustic geyser]

Tbh last time I tried it it had really good limits

But u could only use it weekly

But if u have many many accs it might be usable

[echo aurora]

We don't have an ETA sorry to sya.

[vagrant temple]

yes

[rustic geyser]

Anyways for Text. Is Grok 4.2 Reasoning the best?

[slate hare]

Thank u, is not a problem

[orchid olive]

i think is gemini

[rustic geyser]

Gemini 2.5 Pro?

Or 3 Flash

[orchid olive]

3

[vagrant temple]

flash

[rustic geyser]

So 3 Flash is best for Text in Direct?

[vagrant temple]

3.1

[solemn osprey]

3.1 Flash Lite is like talking to a lobotomized toddler

[vagrant temple]

nah

[vagrant temple]

u can use ai studio for gemini 3.1 pro and flash for free

[rustic geyser]

Whats the limit?

[vagrant temple]

idk but its high

https://aistudio.google.com

[rustic geyser]

Hm alr

Ima try it

[honest verge]

Glm 5.1 is already live everywhere

Great model but too slow

[vagrant temple]

yeah

[civic plaza]

Hmm. It just doesn’t answet at all

[orchid olive]

China's AI is theoretically suffering from a serious lack of computing power

trying the zai

[primal orbit]

it's slower than glm

[vagrant temple]

no

[civic plaza]

I mean it is. But it answers, with no answer at all. It just sends an empty answer

[surreal zephyr]

bro muse is literally free with no ratelimits on official meta app

[surreal zephyr]

its literally fastest sota llm it reaches 350 tps

[vagrant temple]

You pay for it with your data

[surreal zephyr]

...yeah?
wait do you not know how lmarena works?

[vagrant temple]

ig

[surreal zephyr]

this whole thing is a data collection operation lol

thats why its free 😭

[vagrant temple]

but meta is meta haha

[primal orbit]

yeah muse doesn't work at arena atm

worked earlier in the day

[honest verge]

Still I expected more from meta

I know it's literally free SOTA but

I thought they will make something bigger

[rose jackal]

guys

because arena is ahh rn

what other site i can use

[vagrant temple]

design arena

[rose jackal]

why would i need frontend

i need chat

[vagrant temple]

How am I supposed to know that, bro?

[rose jackal]

uh

because frontend is useless, no?

no one needs frontend in our time

[vagrant temple]

rlx bro

[rose jackal]

im chill

just saying

hold on

[vagrant temple]

https://tenor.com/view/you-sure-about-that-gif-6823241795540399689

[rose jackal]

lemme create peak frontend

with this ai

67 tung sahurs

[vagrant temple]

yarrak haha

[rose jackal]

funny meme

[vagrant temple]

peak

[meager sun]

Bro trying to create roblox game

[vagrant temple]

nice name

[rose jackal]

why would anyone need ai for coding in roblox

easiest language

[meager sun]

Not English is.

[rose jackal]

wym

[meager sun]

English is easier language

[vagrant temple]

I tried coding a Roblox game using only AI, and it just doesn't work at all, haha

[surreal zephyr]

its called spark

its extremally small

and its not that bad for the size

[orchid olive]

I fed roughly 13k tokens of script into z.ai for glm5.1, and workflow prompts of process

But the GLM 5.1 just completely ignored the whole flow

[neat acorn]

I'm a little upset that the Opus models are gone from arena.ai...

[thorny schooner]

I hate this this is so actually damn annoying ( because for some reason I got this error and now I can't go forward yet great just great)

☠️

I but look look at this stupid stuff it looks like it's gone but for some damn reason it is asking me to retry but it just gives me the same error and it just gives me the air either way even if I don't do retry

🥲

[humble tusk]

🥵

[humble nexus]

someone know why opus models go away?

[humble tusk]

they are VERY expensive

[copper swan]

Hows my ai ui?

It's fully offline

[humble nexus]

Someday maybe we have Opus thinking Local Models When Technology reduces VRAM consumption trough New Implementations like turboquant or engram but just maybe hahahhaa

[humble nexus]

nice

[copper swan]

Do you have any suggestions to add features I add wake feature with memory it can open youtube my download desktop increase Decrease volume

I added face lock too

[humble tusk]

it seems to me that even under the most positive scenarios, it will be impossible to fit trillions of parameters into 8 GB of VRAM

[calm lagoon]

@echo aurora Muse Spark doesn’t work

[humble tusk]

same

[echo aurora]

Are you getting the blank response?

[echo aurora]

Team is aware of this problem. Working on a fix asap

[vagrant temple]

same

[vagrant temple]

u can use this for free in the meta ai website

[humble tusk]

who will send more of my data - the arena or the meta site🧐

[vagrant temple]

both ig...

[sonic kestrel]

bro

they removed opus 32k

gpt 5.4 high

gemini pro

what now

[vagrant temple]

ai studio

[humble tusk]

sonnet 4-6 soon muhahahaha

[long abyss]

Hello Iam New here

[abstract hinge]

@echo aurora ban

[subtle rose]

https://giphy.com/gifs/H99r2HtnYs492

[obsidian cargo]

I mean theyre a real person before they got hacked so they should just be kicked in case they get their account back and want to rejoin

[calm lagoon]

I’m from Russia, I need vpn

[calm lagoon]

Yeah

[echo aurora]

We ban in these cases, we have an appeal process.

[obsidian cargo]

fair enough, though I got hit with that hack twice and both times getting back into every server I was kicked from was annoying enough -_-

[vagrant temple]

hey

[rose jackal]

biggest glm fan

ahh

glm is the slowest model

of all time

[frosty lava]

using gpt for backend and then qwen 3.6 (free 1000 request on cli) for frontend is really great

[outer zinc]

why are the top 5 models not usable in direct chat anymore?

[vagrant temple]

but the best

[rose jackal]

not even

even sonnet 4.6 is better

[vagrant temple]

hahahaa never

[gray isle]

how is it the best? when it errors a lot... you must be joking..

[rose jackal]

you never tested glm in hard works

[calm lagoon]

Yeah, me

[frosty lava]

its slow but that's not the model fault its due to the hardware used to run it

[vagrant temple]

openclaw isnt hard work?

[calm lagoon]

But I’m fan Muse Spark too

[rose jackal]

i mean sum hard backend things

[golden ocean]

https://tenor.com/view/ева-елфи-gif-7266118905556669339

[vagrant temple]

yeah thats true for backend is opus the best

[rose jackal]

gpt 5.4 is one of my favorites tho

for backend

[frosty lava]

i think for backend its actually gpt

[vagrant temple]

hahaha funny

[frosty lava]

for frontend its opus

the best

[rose jackal]

BRI

BRO

Opus for frontend

is like

[frosty lava]

(if you don't look at the price)

[rose jackal]

nuclear bomb vs child

opus is too strong for frontend

you dont need to use opus for weak tasks

at all

[frosty lava]

if i can only use it for hard task then i don't see any point in using it at all

i will use it 2 time a month ?

[rose jackal]

probably yeah

im using gpt 5.4 cuz its 2nd top ai and much cheaper than opus

[frosty lava]

your logic be like : ITS SO GOOD i wont use it for it !

[vagrant temple]

no muse or gemini

[cunning heron]

gemini 3.1 pro is the best, but no more on lmarena 😄

[rose jackal]

because every new model will do frontend good enough

[frosty lava]

what if you want the best front end

[rose jackal]

theres no point in wasting opus

[frosty lava]

take the best model

[rose jackal]

qwen

3.6

[frosty lava]

qwen aint better than opus

[vagrant temple]

gpt glazer AHHHHH

[rose jackal]

for frontend its almost the same

[frosty lava]

are you sure ?

[cunning heron]

u sure thats best for design?

html css js

[rose jackal]

gemini is prob better cuz svg and stuff

[cunning heron]

bcs im making fivem scripts with ai

[rose jackal]

but you better use special ai design sites

[cunning heron]

ye thats reason why i need gemini 3.1 pro back

[rose jackal]

its free

wdym

literally

[cunning heron]

but now i need to work and i cant find some good model for that

[rose jackal]

on google ai studio

[cunning heron]

where

[rose jackal]

its completely free

no limits

[cunning heron]

ye its free but only 3 answers

[rose jackal]

nah

300k tokens

per day

[cunning heron]

can u send me link if this is real

u can send in dm if u cant here

[rose jackal]

just type google ai studio

thats it

and use 3.1 preview

and uhh go to build thing first

[rose jackal]

you know what

if you using visual studio code

or stuff

whatever

opus is taking lots of context and stuff

its just not necessary for design

but ofc you can

maybe you will get truly best design

but with high cost

[frosty lava]

honestly if i don't use it for frontend then i don't use it at all, i mean for the price of the model and the low usage we get with it, the only thing very good is the front end its doing

[rose jackal]

also they lowered cpu on opus

[frosty lava]

and its a matter of time until other model are that good at front end too so

[rose jackal]

anthropic

[vagrant temple]

Rlx guys...

[rose jackal]

dang ur right actually

[dawn path]

fr?

[rose jackal]

fr

gemini 3.1 sucks

tho

i dont like it personally

[dawn path]

thats cant be real

[dawn path]

for me its best ever ai model

[rose jackal]

for what you use it

[dawn path]

expect claude

[vagrant temple]

BRO calm down

[dawn path]

mostly for coding

[rose jackal]

i mean yeah coding

but where

what language

[dawn path]

cpp and java

[rose jackal]

uhhhh

for java its alright

idk my personally thoughts after month of using

its bad and dumb

[frosty lava]

is there any language where gemini is best ?

[rose jackal]

its awful even in the simpliest luau

[frosty lava]

yeah if we compare it to other ai it's not as good definitly

[willow swift]

even GPT 1 is better than this crap

[frosty lava]

i'd rather use latest open source model or thing like that

than gemini

[rose jackal]

https://tenor.com/view/one-piece-garp-aokiji-kuzan-garp-vs-aokiji-gif-8694662296006827744

[dawn path]

gpt 1 cannot respond normally at all

[willow swift]

unless you are advertising bro

i use it on ai studio , with recommended settings

[rose jackal]

you know the pain of using this

[frosty lava]

yeah that's exagerating but actually i don't think we can be honest if we say gemini is the best

[vagrant temple]

Only GPT Glazers here ahahaha

[rose jackal]

because its good?

top 2 ai

overall

[willow swift]

on gemini the context window is trash

i have no choice but to use ai studio

[vagrant temple]

Why are you defending GPT just like your mom, bro?

[rose jackal]

when it hits 300k context

[dawn path]

Well, maybe not the best AI model, but definitely in third place

[rose jackal]

its so dumb

[willow swift]

still better imo

[rose jackal]

why would i defend it? im just saying its 2nd best model

[frosty lava]

if it was only about price to intelligence then small team that make open source already won lol

[vagrant temple]

and?

[rose jackal]

we're comparing models

[vagrant temple]

imybe in abenchmark

[pure tiger]

guys I just came here after a month have I missed smth? I dont see the video arena😭

[willow swift]

nobody defending anyone

[rose jackal]

benchmarks are useless

ngl

[willow swift]

ai doesn't exist

let's wait for claude mythos

[rose jackal]

gemini is god in benchmark

[frosty lava]

i don't know which ai you already tried ? you should actually try more ai

[rose jackal]

s

[willow swift]

maybe it's ai

[vital mantle]

Claude opus 4.6 1M is the best I think

[rose jackal]

but on practice

[vital mantle]

im happy with it

[vagrant temple]

Yeah, and your top 2 is based on a benchmark dumbass

[rose jackal]

crazy limits

[obsidian cargo]

ai is actually billions of ants they've trained to type

[undone saffron]

Opus mythic is the best actually

[willow swift]

true

[willow swift]

true

[dawn path]

i tried all geminis from 2.5 to 3.1, all gpts and some other models like glm or claude

[frosty lava]

even the latest ?

[vital mantle]

I’ll remove the others AI Claude Has real emotions also

[rose jackal]

benchmarks dont prove anything, gemini 3.1 is higher than gpt and even claude a bit in some ways

[frosty lava]

and you still choose gemini ?

[rose jackal]

but gemini is bad for example

[dawn path]

yes (expect claude mythos)

[rose jackal]

claude mythos will cost like 30$ for 1m

i guess

[vagrant temple]

omg pls stop writing sht

[vital mantle]

and it sends pools plans no ai does that

[rose jackal]

they should lower the prices like kat coder v2 pro

[frosty lava]

that's actually crazy cause why exactly did you choose gemini, being truly honest it's definitly not the smartest one

[rose jackal]

0.20$ for model at gemini lvl

talking about kat coder

[vagrant temple]

@rose jackal how old are u?

14?

[rose jackal]

17

[vagrant temple]

You just said benchmarks are unnecessary, and now you're showing one...

[undone saffron]

Don't get your hopes up that the Claude Mythic will be in Arena, and even if by some miracle they did add it, it would be extremely limited due to its high cost

[rose jackal]

https://tenor.com/view/higuruma-hiromi-higuruma-jujutsu-kaisen-jujutsu-kaisen-meme-jjk-memes-gif-10370496866020481076

i said they're not proving anything

i talk about price

kat coder in benchmarks is one of the top models

with the lowest price

0.20$ for 1m tokens

when claude is 20$ for 1m for example

[vagrant temple]

at this point use step or minimax

[rose jackal]

minimax 2.7 is good tho

[vernal raft]

uhu

[frosty lava]

deepseek still didn't released the new one

that's crazy

[vagrant temple]

ssaaame

[vernal raft]

is meta not working for everyone?

[rose jackal]

v4 will be crazy

they said in end of this month

[frosty lava]

yeah probably the best open source model when it will be out

[willow swift]

for god's sake how do i fix response timeout in glm-5.1

[rose jackal]

and deepseek is cheap

so it will be the best chinese model

BUT

deepseek 3.2 was trained on opus answers

smth like this

copying stuff

so i hope v4 will be originally

[vagrant temple]

like any other...

[rose jackal]

nah

[valid pulsar]

bro what do u mean when u r usign "deepseek" and "the best chinese model" in one sentence

[rose jackal]

im not using it

v 3.2 is bad

[valid pulsar]

yep

[rose jackal]

i said about v4

its gonna be peak

[vagrant temple]

Almost all Chinese models steal data from major U.S. models

[vagrant temple]

.

[frosty lava]

v3.2 was in december

imagine v4

improvement

[rose jackal]

also multi stuff

like good image vision

[valid pulsar]

and still will have a lot of forbidden topics

[rose jackal]

and stuff like watching videos

[rose jackal]

wym

[valid pulsar]

like taiwan

or kinda

[rose jackal]

so what

you will never use taiwan

anyways

[rose jackal]

but you right

its kinda dumb move if they will keep it

[valid pulsar]

ok thats good point actually

[valid pulsar]

but yes

[rose jackal]

what model do you wait for

[obsidian cargo]

gpt-image-2... 🤤

[valid pulsar]

actually i dont really know much about it, recently used opus and thinking of how can i replace it now

and

[rose jackal]

where you used it

on arena

?

[valid pulsar]

yes

[rose jackal]

arena is dying rn

idk

[valid pulsar]

totally agree

[rose jackal]

rn im looking for some better site

at least with free gpt

[spring oar]

Opus 4.6 nerfed ?

[rose jackal]

yes

[valid pulsar]

i can suggest one

[rose jackal]

anthropics fault

[valid pulsar]

thing

[rose jackal]

?

[valid pulsar]

u can buy an invite to a buisness workspace like for a month and it will cost kinda $1-2

5.4

[rose jackal]

not worth it

[valid pulsar]

but a lot cheaper, idk

[rose jackal]

idk

i will wait

like 3 months

when gpt will be free

in every site

opus 4.6 will be default too

i guess

[frosty lava]

maybe we will be able at some point to run model locally that's almost the performance of top 3, for example right now with actual innovation, we can run glm 5.1 which normally require 1.4 T of vram on only 200 go and its about unified ram + vram so..

trust the progress

[valid pulsar]

so what i was tryna say, recently i had like a bug or smth idk, using direct mode on arena and somehow there was a choice of 2 models and there was an opus, i chose it and it was answering me the whole time, when the second choice appeared i chosed skip and still had an opus

[wary robin]

Has arena removed models like claude opus? I don’t see them anymore

[rose jackal]

read the #announcements

and yea

they even removed gpt 5.4 tho

and gemini 3.1

every top models

[orchid olive]

deepseek is free

and it is 💩

[valid pulsar]

maybe anyone know like how to start new chat with a choice of answer? cause i got a bug when generation hadnt stopped and i cant write anything

[rose jackal]

hey

i got this bug

for like

3-4 times

[valid pulsar]

yes its pretty often

[rose jackal]

it will continue

just refresh page

it will be done in 3-4 mins

just trust the progress

if it isnt

= create new chat

[valid pulsar]

i was doing it like for hours)

[rose jackal]

then its bugged\

cuz arena is awful at all

with A LOTS OF BUGS

and they even removed models

[valid pulsar]

yes but how to make arena give me this choice one more time

is it random thing

[rose jackal]

random

[valid pulsar]

damn

[rose jackal]

so

you have not safe chats

and they training ais or stuff

but they wont let use best models

this is literally useless to use arena rn

tbh

[valid pulsar]

dumb thing actually

hahaha

[rose jackal]

just gets collected

gpt-bradar-v1

https://tenor.com/view/hey-brother-hey-brother-what-is-this-what-am-i-looking-at-gif-15927984058571805389

[hybrid obsidian]

Compare these two photos, which show where the GLM 5.1 model is located in the coding

Why does the server say it's in 3rd place, but on the website it's in 7th place?

[echo aurora]

You're looking at Text Arena's Coding Category -> https://arena.ai/leaderboard/text/coding
The Code Arena is a different leaderboard -> https://arena.ai/leaderboard/code

[hybrid obsidian]

oh, thanks for clarifying

[echo aurora]

No problem

[valid pulsar]

maybe anyone know, is claude's and arena's limits of tokens are similar? im talking about sonnet 4.5

[spring oar]

why opus 4.6 is nerfed

[hybrid obsidian]

wtf bro

[echo aurora]

Thanks for the flag

[hoary lynx]

Glm 5.1 is better than sonnet 4.6 in code ?

[thorn hornet]

Yeah

[vagrant temple]

yes

[hybrid obsidian]

I didn't expect such a breakthrough from them

Does anybody know when the next update's comin? Like, not just addin AI models, but like an actual website update?

[honest verge]

Meanwhile deepseek:

Deepseek is basically dead

It's delayed by the may ig

[honest verge]

Atp it just won't come out

Too much time has passed

Most likely they are going to skip this model

[wary robin]

Bro if we agree to give all our personal info away, we expect ALL frontier models. That’s including opus and gpt 5.4 and gemini 3.1 pro

[thorn hornet]

I'd agree on that

[vagrant temple]

same

[wary robin]

They rate limit us anyways, so atleast give it to us a little bit

[tulip parcel]

Guys, what AI model do you think gives the most human like response? Other than Opus.

[wary robin]

Sonnet 4.6

Only Claude seems human to me

[vernal raft]

You should question yourself if your personal info are valuable enough

And what would they do with those

Since they already collect prompta

[stuck vortex]

Hi

[vagrant temple]

hey

[vernal raft]

@echo aurora Do you have any ETA for meta spark?

[light siren]

@echo aurora may we talk

[echo aurora]

No ETA sorry to say

[echo aurora]

I'm about to head into a few meetings so I'll be a bit slow to respond, but yeah what's up?

[light siren]

see anything weird?

[honest verge]

At least give us 2-3 responses limit

[echo aurora]

Yeah that's odd. Are you seeing that now? On all browsers?

[light siren]

not a bug lmao

[echo aurora]

Can't say I'm seeing the same

[light siren]

I brought back LMArena dude

look top left

[echo aurora]

Yeah I see that

[light siren]

I added it as a feature to my extension

any opinions?