general

[echo aurora]

Hey @desert fiber so there is PDF upload supported. However, it's for vision only. You won't be able to make edits to a PDF.

Code Arena in it's current form is for building the frontend for sites/apps. We are in the process of building more support though. A full-stack experiment is currently released and we're hoping to fully release as soon as it's ready.

[icy ice]

thanks. right now I just end up using normal chat, and then I have to copy paste code. i also think it uses a lot more tokens (bad for me and bad for arena.ai). these are small apps nothing like a full stack site. is that the best way? any way to get beta access to the experiment?

[echo aurora]

Max is going to select the best model for that prompt keeping latency in mind. It's the same idea as the prompt-to-leaderboard but turned into "model".

[echo aurora]

Sorry to say there isn't a way to opt into this experiment. We do fully understand the importance of having full-stack and want to move the progress along with it quickly.

[thorny schooner]

Well I think it's swinging too much with latency because I have seen it almost legitimately only use Google AI like I'm not even kidding almost no other AI I have as seen be used as a model that's not Google when us except like every couple other one and even then that's kind of legitimately becoming more rare at least in my experience when Google just does not work with the prompt

honestly sometimes will actively disrupt if I'm using a particular one especially since google every time my experience tends to have a much more defined output limit then some other ones which can disrupt the frame of chat if I was using a previous different model before experimenting with Max

[echo aurora]

It's all going to depend on what is being prompting and what the chat session looks like, so I wouldn't be able to tell you if something isn't working correctly with your chat session. Having Max in Side by Side I think would be really interesting to see how those responses compared to sticking with a specific model.

[thorny schooner]

Very septic that depends on being prompt actually considers anything but okay

[keen beacon]

In my experience if you used socratic prompting or just paste a 300 word essay paragraph you'll have Anthropic, else mostly Google

Using Max I've only ever once see something that isn't Anthropic and Google when I asked about pseudoscience 😕

[echo aurora]

It's a model router, it's going to route to the most capable model for that specific prompt (with latency in mind).

Did you mean to say you're skeptic?

[pseudo hemlock]

Hi

[echo aurora]

Hello

[pseudo hemlock]

How’s life

Lying Flat

Oh ok

[pseudo hemlock]

Did you get grok

[keen beacon]

indeed

[west ore]

guys no offence i think the gpt image 2 model is useless most of it is fail after fail

[pseudo hemlock]

Really?

Wdym fail after fail

[west ore]

like when i ask for any ip (Even nickelodeon IP)

sometimes it works sometimes it doesn't

i know the model was just added last week and i need to get used to it

[thorny schooner]

Yes

[radiant turtle]

It's over

[brisk turret]

Why is the x axis on the pareto chart descending instead of ascending?

Price is going down instead of up

[void shore]

<@&1349916362595635286>

[thorny schooner]

Can't say for the first one but for the second one even on so even I'm not certain since I I i'm pretty sure I regularly do stuff that I am very certain to go way past the 300 words i it just seems to default to Google on honestly at least in my experience

https://klipy.com/gifs/palpatine-star-wars-16--k01KQ8VX1Z10JW76KKKPAEA4KKA

( lost the entire chat just because trying to experiment with Max)

[jovial palm]

saw the same thing with Claude 4.6 too

[desert pendant]

<@&1349916362595635286>

[frozen swift]

what happened?

[velvet furnace]

cloud-buddy = opus?

[light sleet]

I asked it to generate the best steak it can and add pieces of 24 karat gold

[wicked sage]

kimi 2.6

[vale quest]

All this and still no claude opus 4.6 in direct

I hate arena.ai

[vale quest]

Mimo v2.5 pro is underrated prove me wrong

[jade cloak]

I’ve already logged in but when I click regenerate, this will pop up

[golden horizon]

battle mode have opus 4.6 bro

[arctic halo]

Hello Friend, you can make Rs 13,939.50 ($50) every day working on BunnyBand. You will receive a free Rs 2,787.90 ($10) welcome bonus when you register. You can make more money through social tasks and playing games. Then set up your payment method and withdraw. It's easy to use. Register below to get started: https://bunnyband.com/r/sP8kxVLmRwZS

[slender ledge]

<@&1349916362595635286>

[slender ledge]

ban

[barren sorrel]

Still looking for people to evaluate the text for book written by AI

[sonic wigeon]

take a ban and get a real muskan on your face!

[compact flame]

That does not look like an average prompt

[thorny schooner]

Guess what I had to do like three of them because of errors so yeah I'm not surprised I got it but still annoying

[arctic halo]

Hello Friend, you can make Rs 13,939.50 ($50) every day working on BunnyBand. You will receive a free Rs 2,787.90 ($10) welcome bonus when you register. You can make more money through social tasks and playing games. Then set up your payment method and withdraw. It's easy to use. Register below to get started: https://bunnyband.com/r/sP8kxVLmRwZS

[narrow oracle]

Arena. Ai is free or paid?? Can someone tell me

[scenic holly]

Paid with your time

[pliant cliff]

am I the only one who thinks mimo v2.5 pro is like something between Claude opus and glm? Well, it's clear about glm, they're Chinese. but the fact that it looks so much like Claude is very interesting.

[low quiver]

👀

[plucky sparrow]

<@&1349916362595635286>

[quartz light]

check out qwen image 2 pro! it made this in 5 sec

input image

[upbeat ferry]

we woke up to another day where muse spark beats every model in the world

If you are rich donate zuck more money so he can place 1st

My boi muse spark needs that encouragement

[deft horizon]

Hi anyone , what’s the best Ai too on arena to generate a realistic experience short video using my prompt

[slender ledge]

is hugginface worth it guys

thats how they get the models i think

[velvet furnace]

why always popup

[dense sphinx]

Guys you here?

What agent mode for?

I don't understand.

Can someone explain?

Hello by the way

[compact flame]

There's no agent mode

It's not even released

[dense sphinx]

But I have

Wanna see?

[compact flame]

Show it.

[dense sphinx]

@compact flame here.

Hey man you OK?

[compact flame]

You got early access

Nobody has idea what it does but it exists

[dense sphinx]

Yes and it's bad?

[compact flame]

No it's not

You are just lucky

[dense sphinx]

It's said it's multi model

But here is catch

[compact flame]

Ask pineapple or something he probably knows

[dense sphinx]

OK thanks

I hope you get agent mode soon too

@compact flame i root for you.

[compact flame]

Why

I didn't even do anything special

[slender ledge]

what is agent mode

[dense sphinx]

Multi model that allows you do coding and etc

[slender ledge]

wym

like just battle mode

or wha't

[dense sphinx]

Yes but enhanced version

I even made game with that

[slender ledge]

more modeles at same time?

[dense sphinx]

Yes

Again it also works perfectly for coding

@slender ledge also it's not available on many users

[frosty lava]

so basically 5.5 codex this week ?

[slender ledge]

when will it release

[slender ledge]

isnt it out already

[dense sphinx]

Ask pineapple but for me it's avaliable.

[frosty lava]

no its the normal 5.5

the 5.5 codex is not out yet

[slender ledge]

k

is agent mode like u can make an agent?

[dense sphinx]

I also made rpg game from agent mode.

[slender ledge]

like give an prompt for the agent?

[dense sphinx]

@slender ledge Yes prompt also gives

[slender ledge]

can u show

[weak dagger]

Send me the website

[slender ledge]

btw cuz u got things u also got like opus in direct?

[dense sphinx]

@slender ledge no only in battle mode, I almost got opus but it vanished moments later

[misty star]

https://arena.ai/agent

[slender ledge]

sends me to battle mode

[dense sphinx]

Really?

[weak dagger]

Yea

How did you got it

[dense sphinx]

@weak dagger few months ago I signed new account

And it appeared.

Agent mode available for me in early preview

[earnest rover]

any idea in arena ai i saw a model in battle mode named flow state 3
which model it can be ?
nb 2 pro or something flagship
it was quite good though

[weak dagger]

I need it i need agent mode

[slender ledge]

same

btw

quite dangerous

u can do jailbreaks

[misty star]

We are currently experimenting with a new feature: Agent Mode. This is a multi-modal chat experience that allows you to work across different modalities within a single, unified workflow.

For more information on experiments happening on Arena check out this article.

[dense sphinx]

On*

What*

[slender ledge]

like u can give an prompt like u can make agents right or cant u like if u can u give him an prompt

and its jailbroken

[weak dagger]

Could you like try makeing a new account

[slender ledge]

wtf

my grok does all i want

[dense sphinx]

Well grok is no rules

[slender ledge]

ik

i easy jailbraik it

it tells i want

[dense sphinx]

Even I did it won't show agent mode anymore.

Hold your horses mate

It will block as violated terms of use

[slender ledge]

nah

on gork

or this discord

[dense sphinx]

On arena it's not.

[jaunty mist]

i tried to send a suggestion, but it was instantly deleted lol

[slender ledge]

oh

[weak dagger]

Try it i wanna see how it looks like

[main iron]

When is Claude opus coming back to lm areana

[obsidian cargo]

When the light is running low and the shadows start to grow and the places that you know seem like fantasy

[neat lagoon]

Not

[hollow nebula]

I just got this model too and it was really really bad at frontend

[silent tree]

Sorry to ask, what was your prompt?

[hollow nebula]

using the example prompts on the code feature, asking it to create a discord server clone 😅
I'm gonna test with the same prompt other models and see if it's the model being that bad at front end or if it's just the prompt

[bleak minnow]

??

[hollow nebula]

in what language?

fmhy.net

use this site

to find sites

[hollow nebula]

guess which is deepseek's and which is 5.5xhigh's

[sand crown]

You can use Nuvio

[light sleet]

1 is 5.5 xhigh and 2 is Deepseek

[hollow nebula]

nope

[light sleet]

Did you get 5.5 xhigh with Deepseek V4 Pro too?

[hollow nebula]

It's the opposite

[hollow nebula]

Which I find crazy

[hollow nebula]

I selected it manually I had haiku 4.5 with 5.5 xhigh

[silent tree]

5.5 got the server logos right lol and the pings similar

discord doesnt show

• for pings

[hollow nebula]

Overrall it seems like it's worse at one shotting front end though

using it via codex it's a pretty good model but even gemini 3.1 pro at launch was better than whatever this is

[hollow nebula]

note how I'm saying "at launch" since because of how they slaughtered that poor model it's worse than 3 pro at most stuff

using it not via api

[wicked talon]

I deleted Kimi low-key pmo

"system busy"

[ruby kayak]

Can someone answer a question for me, please?

I'm trying to regenerate an image, and the site keeps asking me to log in, even though I'm already logged in.

[oak laurel]

try to clear your broswer cache and cookies

[steep wren]

lmarena do not longer more generate unlimited images?

[wicked tapir]

how much context do models get after I vote but I submit another prompt?

IIRC there was a blog post that mentioned how multi-round conversations on Arena work

[echo aurora]

That's odd, if clearing cookies/cache don't work, would you mind creating a new bug ticket in #1343291835845578853?

If you're able to share a screenshot of being logged in & getting that error message that'd be helpful to have as well. I would note to share this by sending a direct message to @oak python so it can be shared privately. I wouldn't want you sharing your email in this server where others would have access to it.

[echo aurora]

There are rate limits in place. More info can be found in this announcement: #announcements message

[wicked tapir]

actually i found the post: it's https://arena.ai/blog/opendata-july2025/#evaluation-order

[steep wren]

thanks friend

[ruby kayak]

One question before I submit this to bugs. In Battle Mode, if I use the "retry" option a certain number of times, is it normal for a screen to appear asking me to log in, even though I'm already logged in?

[echo aurora]

No, if you're being asked to log in when already logged in, that's going to be a bug. It shouldn't be doing that. There are rate limits, which may be what is happening, but that'll be a clear error message instead of saying you're not logged in.

[spring rune]

Gpt 2 image in arena when?

[steep wren]

so now all the members are limited to generating only 15 images per day. right?

why this change happen?

[indigo knoll]

its on Arena

[echo aurora]

Yeah that's correct. We need to have some reasonable restrictions in place so we're able to continue to provide access to these models. We plan to be around for the long-term and these limitations help sustain that.

[wicked sage]

hii

[echo aurora]

Hello

[wicked sage]

ananas can you check out #1498684119951872160?
if you cant right now, thats fine. take your time!

[frozen summit]

some image models refuse to visualize prompt containing "higgs boson" because apparently they think its analog to the all known racist word.

😭🙏

im crine son

[whole sundial]

this is mainly because people were abusing arena's image battle mode to get gpt image 2 before release, people were generating hundreds of images that they ignored in an attempt to get image 2, it cost them a lot of money

[frozen summit]

nice tag you got there

[torpid pagoda]

Which ChatGPT model is best for image generation on Arena AI?

[whole sundial]

there's an ai company called higgsfield and an unrelated tts model called higgs boson

[frozen summit]

😭

[whole sundial]

idk why it would be blocked

[wicked sage]

gpt model 2.5

bro i am just

HOW DID I SAY THAT SO WRONG

gpt image 2*

[frozen summit]

though im not talking about model

im talking about this

we got to the point where image models restrict visualization of a higgs boson which gives mass to particles

[wicked sage]

higgs boson = BAD!!

[whole sundial]

yeah i know i'm just saying that they are using the same words, it would be weird for arena to block the name of another ai company in their filter (or i guess the creator of the model)

[compact flame]

Pineapple if you are reading this

What is agent mode on arena

Because some people have early access to it

[icy frost]

i got the envs update

[crystal mica]

gemini 3.1 pro back in lmarena when

[icy frost]

agent can create files and search on internet
idk other|
and u can't choose a model

[buoyant crypt]

I get why they did it, but isn’t Arena basically for comparing models? Feels kinda contradictory

[echo aurora]

It's a new mode we're experimenting with. https://help.arena.ai/articles/1811908126-arena-experiments-agent-mode We're going to be releasing this feature to more and more user over time so keep an eye out for it.

[slender ledge]

Will u be able to build ur agent

With like an promo

Prompt

like that u can give an prompt

[slender ledge]

Pls me

[light sleet]

@echo aurora Very excited for this

Is it also like releasing for each users

[slender ledge]

Yea

Ialso want it

[golden ocean]

real

[keen beacon]

tbh I don't get why agent mode so hyped
maybe I'm not a vibecoder (still use AI the traditional way for occasional code)
I've tried it while logged out and found it's quite uninteresting.

[slender ledge]

Did yall know arena has 2 other sites

[light sleet]

Agent Mode is kinda not like fully stable rn

I'm excited for database that's what I meant

[golden ocean]

agentic ai flows made my iq drop by 50 after using it for months but increased productivity

[light sleet]

Yall got early access to agent mode I got early access to the new pineapple plushie

[keen beacon]

agentic ai flows made my iq drop by 50
nah I wouldn't exchange my 132 IQ brain for such "increased productivity"

Prob why I don't get it

[storm dust]

bro got access to gpt 2 image instead 😭

[light sleet]

No😠

[golden ocean]

W

[covert totem]

give us agent mode too plz

or else

I'll dismantle you

[storm dust]

i will vaporize you

[solemn osprey]

would agent mode show up here?

[covert totem]

i would Cut you in half

[storm dust]

yo chill

[rocky geyser]

Hi

[rocky geyser]

No

[red chasm]

This is all cool, but when will we get all the removed models back?

[echo aurora]

No, it'd be in the drop down where you select Battle/Direct/Side by Side

[proud bobcat]

No DeepSeek results yet?

[solemn osprey]

ah

[solemn osprey]

👀

[echo aurora]

No ETA yet sorry to say.

[compact flame]

I wonder if it works or it's just a page

[echo aurora]

[covert totem]

😡😡give

[compact flame]

Is the agent page just for viewing or it actually works?

[storm dust]

bro even made a mod nervous

[echo aurora]

It works

[covert totem]

i own company trust

[compact flame]

Oh k

[red chasm]

Still? Wow, that's interesting.

[dreamy hill]

I wanted to know why new AI models are being added, but I can't find them on the website.

[keen beacon]

Chill, man

[rocky geyser]

@echo aurora Have many people gotten access to the database and api features yet? I havent found other people except myself encountering that, will it be rolled out to the public soon?

(sorry for ping btw 😭 )

[oblique sentinel]

its available on hugging face

not the api

but the dataset

it's updated regularly as leaderboard updates

[echo aurora]

Not all models are going to be added to Direct/Side by Side. In our announcements for new models we'll make that distinction.

[rigid copper]

i think agent mode is not available for me yet c: since it was limited to some user

[echo aurora]

I won't be able to give an early heads up on when new features are (or may) be landing.

[oblique sentinel]

https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset

[rocky geyser]

Okay

[echo aurora]

Yeah doesn't look like it sadly. We are looking to roll this out to more and more people. But in the meantime we are looking to hear from those who've used it.

[compact flame]

Just to ask does gpt 5.5 Xhigh exist In battle?

Or xhigh models are unavailable

[dreamy hill]

A few days ago, I noticed that the new version of Claude Opus, 4.7, was released. However, I’m having trouble locating it, whether through direct access or side-by-side comparison.

[compact flame]

It's battle mode only

Because it's expensive

[echo aurora]

xhigh version is in Battle

[inner relic]

they have claude mythos in attack battle

trust

[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. You can find more information in this announcement: #announcements message

[storm dust]

im gonna fight mythos one day

[inner relic]

cloud buddy is mythos

[compact flame]

Why it's rank on leaderboard is not displayed though

[keen beacon]

Just a hypothesis

[inner relic]

...

ok

[oblique sentinel]

How does arena make money?

if you guys pay for api costs to test the models

[inner relic]

they make money

through datas

data

[sharp mirage]

Btw pineapple will agent mode be available for all ?

or just some ppl?

[inner relic]

so our personal information or prompt go to them and they get money

for selling data

[storm dust]

can i sell this data

[inner relic]

no my words are useless

your's too

[storm dust]

ban

[primal orbit]

If google did truly value arena.ai it would have named it's model nanopineapple instead of nanobanana.

[inner relic]

who's kizenbeta

ya know

?

Got this model through battle chat

[compact flame]

Also I think the anonymous models pay to be on arena

[light sleet]

It isnt

[storm dust]

banana who wins?

banana vs banana man?

[compact flame]

Banananator

[inner relic]

oh also

I asked gemini 3.1 to review kizen beta

[light sleet]

Banananator

[inner relic]

and it said, it's claude sonnet 3.5

oh

sorry gemini 2.5 pro not gemini 3.1

[storm dust]

who's that

[compact flame]

Anonymous models don't have a specific model name honestly

[slender ledge]

Pineapple could u dm me

[light sleet]

the bananasille guy

[slender ledge]

I got an question

[silent tree]

Use Modmail

@oak python

DM Modmail to talk to Pineapple.

[storm dust]

i dont know that guy

but banaan man definitely cooks that guy

[split topaz]

I'm still wondering about the identity of botbot2

[inner relic]

Bro i think kizen beta is a gemini model

can ya check

it might be..

gemini 3.5 pro

[compact flame]

It might be just a custom model

[inner relic]

holy bro

[compact flame]

Trained by people and sent to arena for testing

[inner relic]

ok

I figured out it's seed

[slender ledge]

What does arena use

Huggingface?

[keen beacon]

FastChat

or was FastChat

[storm dust]

yes

[keen beacon]

I missed that day

[echo aurora]

I'd encourage you to read this blog: https://arena.ai/blog/ai-evaluations/ It's important to note Arena does not sell user data or individual voting information. Commercial activity is limited to evaluation services, not data monetization.

[oblique sentinel]

Ahh I see very interesting will check it out- yeah wasn’t making accusations was just quite curious.

Thank you

[echo aurora]

No problem! I didn't think that you were. I'm more-so saying that for other people who may not know.

[weak dagger]

@echo aurora

[slender ledge]

Pineapple

Could u give me agent acces

[keen beacon]

@slender ledge

[slender ledge]

Wym

[keen beacon]

Agent mode is an experiment, which cannot be opted-in.

[slender ledge]

Yes I know

But can’t he give acces to an email

Could I get ur acc?

So I can test rq

[keen beacon]

My account do not have Agent mode
It was a logged-out private window.

[echo aurora]

No lets not do that please.

[slender ledge]

K

[slender ledge]

Pineapple do u have an date when it wil be released for everyone

[echo aurora]

10/10

[echo aurora]

Nope, not ETA

[slender ledge]

How do u get it

[whole sundial]

i do want to respond to this, if that is what you are concerned about then maybe only have it be able to be opted in for experiments like video arena, agent mode, and the stop button and keep experiments like opus removal and battles in direct as they are where they are solely random and can't be opted in or out

[slender ledge]

Or u just have to be lucky

[keen beacon]

Pure luck

[slender ledge]

Damn

[weak dagger]

I wanna put this as my pfp in the server

[slender ledge]

Gl

what

[keen beacon]

Or hack the server's PRNG

[slender ledge]

💀 🙏

Gg

Ur banned

Idk

[echo aurora]

I still think that wouldn't address the concern though. If we did have some kind of toggle for opting into/out of experiments, only those with knowledge and understanding of that toggle are going to use it. When that happens, we're limiting the users in that experiment just to those that are aware of this toggle. That may create a little bit of inaccuracy in the results.

[slender ledge]

pineapple what api do u use

For the models

Huggingface?

[whole sundial]

then i guess you could just make the toggle more prominent? or when you launch a new experiment put steps to toggle it in the announcements?

[whole sundial]

for text i believe arena mainly uses official apis, for image and video they use fal unless the model has an official api and arena has an account with them (i imagine arena doesn't have an account for every model maker especially on image/video models just because there are so many of them and some don't make llms that are on arena)

[slender ledge]

Ok

[novel skiff]

where to create ai videos

[slender ledge]

Video mode

[novel skiff]

yes

[whole sundial]

video arena is only available at https://arena.ai/video, the bot is no longer available on the discord

[novel skiff]

ohh thats saed

sad

[misty star]

https://discord.com/channels/1340554757349179412/1494733753111547984

[novel skiff]

no acces bro 😮

sadly

[whole sundial]

what's even sadder is what happened to the limits, it was 10 but now it's 2

[novel skiff]

i cant become a youtuber nopw

I can't generate actual video files — I'm a text-based AI, so I can only write descriptions, scripts, and prompts for you to use in video generators.

[keen beacon]

GAID syndrome

[novel skiff]

sad

[keen beacon]

For a year since I've discovered LMArena, I've learned how to use AI effectively.
Now since April 1st I've been learning how to think deeper and learn more.

[light sleet]

I have access there

[shy mason]

it's an archived channel soo

[light sleet]

Yeah it was nice while it lasted

W @echo aurora

[echo aurora]

W to all produce

[astral jetty]

@echo aurora question, does the max model route you to any model in the interface, or just the ones you can access through the direct model

So like, could max take you to Gemini 3.1 pro?

[surreal zephyr]

gpt 5.5 cooked so hard wtf

[compact flame]

Dang that actually looks sick

What game it created?

[surreal zephyr]

noita

[compact flame]

I mean genre or smth

[surreal zephyr]

the gameplay is peak tho, especially the animations

and bossfights

[compact flame]

Ohh kk

[misty star]

Can i he profuce

[steep scaffold]

Hi

[wicked sage]

WAIT THIS IS SO PEAK

WHAT THE HELL?????

[surreal zephyr]

its even peaker to play
1 prompt btw

[wicked sage]

what was the prompt

[surreal zephyr]

"make noita but make it look good, max effort"

[wicked sage]

wtf

[surreal zephyr]

yeah 5.5 pro extended is insane

[grand rivet]

gemini isnt that good its security

[wicked sage]

man what the hell

[grand rivet]

i was still able to jailbreak it

[wicked sage]

this shii costs more than opus

[grand rivet]

LMAO

[grand rivet]

its not even better then opus...

openai needs to calm now

[wicked sage]

i guess bro

[grand rivet]

china is goat tho they keep open sourcing models

that forces other european Ai techs to reduce their pricings

[wicked sage]

china is goated when it comes to ai models

[grand rivet]

AI race got better cuz of china

[grand rivet]

fr

imagine we get a ALT of Mythos by china and they OPEN SOURCE IT..........

crazy shi

[wicked sage]

dude thinking of v4 pro is LITERALLY next to gemini 3.1 pro

[grand rivet]

funfact muse spark is free of cost....

they got no Subscriptions and shi

[wicked sage]

muse spark is free??

[grand rivet]

Yeah bro fully free

Till now Meta is planning to make it Freemium in some weeeks or months

but it is free for now

thinking mode fully free

[silent tree]

no wonder its trash

[grand rivet]

yeah kinda it hallucinates alot but with clear context and if we dont chat much in one chat

then its beast

not good at coding tho

i tested it

[loud pike]

guys anyone else have the problem with gemini flash image model that i cant put more than like 5 images ? i need to make new chat

[grand rivet]

yeah its a new limit

u want to make it die broo..... it will hallucinate alot

[loud pike]

ah

[grand rivet]

._.

[loud pike]

isnt there a paid version ?

[grand rivet]

use gpt 2

on arena

[loud pike]

gpt 2 is bad for the images im doing

[grand rivet]

what is it give me those images in dm ill do it for you

its fine for me bro see i tried and it worked probably some issues on your end

see it worked like a charm

[loud pike]

check dm

[sleek stump]

https://tenor.com/view/cat-gato-ansioso-excited-gif-26621483

[lean ice]

you ever notice how a good cup of tea feels simple, but only because someone knew exactly how to make it?
AI projects are the same.
A lot of them look good in a demo, but once they meet real users, messy data, APIs, workflows, and edge cases, the taste falls apart.
I’ve spent the last few years helping startups and mid-sized teams turn rough AI ideas into systems that actually hold up in production.
I’ve built tools across healthcare, finance, and education, helping teams cut manual work, automate onboarding, and make faster decisions with data.
good tea needs balance: the right leaves, water temp, timing, and patience
good AI needs the same: clean architecture, reliable integrations, useful automation, and someone who knows when to keep things simple.
if your AI idea feels tangled right now, let’s talk shop.
I’d rather help you build something people can actually use every day than pitch at you.

[pseudo hemlock]

Hi

[feral elm]

lol pretty sure claude's discord got deactivated or sum rn?

[echo aurora]

Did Discord just have an outage?

I think the server is struggling atm.

[raven matrix]

yeah

[surreal zephyr]

openai and flow servers are gone

lol

discord is cokoed

[wicked talon]

Discord is down lollll

Wait there back up

[echo aurora]

https://discordstatus.com/

Yeah

[wicked talon]

I just realized

They are genuinely calling this server small

Lol

[echo aurora]

Yeah looks like it's recovering

[feral elm]

it is, not much else to say

[ornate light]

some of my servers are broken

[waxen grove]

hi

[echo aurora]

Uh oh, looks like the server is struggling again 😭

[grand rivet]

hmm now we can connect claude with blender,,,

claudes gonna cook 3d designers???

[gray lion]

no way

blender?

[pseudo hemlock]

We’re so back

[grand rivet]

yeahh

crazy right

[crude zealot]

че было с дс

[loud herald]

Claudes already been in blender via mcp servers this isnt new its just the official claude talking about it here

[grand rivet]

but yeah that was hard for begginers

its now easier to connect

[surreal zephyr]

theres already mcp for that, and claude is actually awful at it

[loud herald]

Not really, literally just download the plugin for blender and tell claude to connect to it lmao

[surreal zephyr]

gemini 3.1 pro and gpt 5.5 are only two models good at blender

[grand rivet]

they will improve it

[surreal zephyr]

claude cant read analog clock

its GARBAGE and multimodal reasoning + vision

[grand rivet]

bruh who tf gonna sit and install the depenedencies for it

[surreal zephyr]

litrally as bad as deepseek

[loud herald]

Not awful but what you'd expect an AI that isn't trained on making things in blender

[grand rivet]

i agree at Vision

[surreal zephyr]

no claude js isnt multimodal

properly

[grand rivet]

lets just wait for 3 more months

[loud herald]

The hundreds of thousands that used it

Its a plugin

[grand rivet]

yeah

[surreal zephyr]

its literally 2 clicks, or js ask codex to get it

i literally used that

[loud herald]

Maybe AI truly isn't a good thing if we got people that cant do anything for themselves atp

[surreal zephyr]

5.4 and claude opus 4.6 were garbage at it

gemini 3.1 pro kinda cooked

[grand rivet]

yeah ai is good if we combine human skills plus AI

thats a good thing then

cuz AI alone isnt that good

[loud herald]

True that, 160K tokens for reading a childrens book and analyzing it

as well

[loud herald]

Thats what I'm tryna say, it wasnt difficult to begin with

[loud herald]

Depending on the model it is

People make models to be specifically good at doing things by themselves too

But at the end of the day

[grand rivet]

true

insane multi models are not that good rn

[loud herald]

The only reason claude is "Good" in peoples eyes is because of its creativity in frontend design

[grand rivet]

true it sucks in backend bro it cant even handle a API and call models seperately

i just tried it yesterday and chat was giving errors

[loud herald]

Nah thats just anthropic being horrible

When they crash like that 9 times out of 10 its Anthropic

[grand rivet]

true i only see people making Trains,Svgs and those shis..

[grand rivet]

lol truu

[loud herald]

And GPT 5.5 still beats it 10 fold in svgs

[grand rivet]

yeah lets wait for the next model...

[loud herald]

There wont be a next model that beats them, and also wont be a next model for OpenAI currently lmao

Scam Altman is in deep trouble

[grand rivet]

we didnt expected a big jump but what i saw from gpt the image 2.0 was crazy

[grand rivet]

gemini will release the next model probably now

claude and gpt used their turns now

[loud herald]

Maybe, if their model comes out to be better coding than both claude and GPT it will be over lol

[loud herald]

Thats the only thing that truly improved, Looking at the most recent releases the AI is hitting a cap and its because they keep scaling

[grand rivet]

and if China released a ALt to Mythos and open sourced it.....

[loud herald]

And now these 1T parameter models are on par or beating

[grand rivet]

they gonna use new techniques now to improve it even more cuz just increasing the parameters wont effect it they need new techniques of training

[loud herald]

Thats already happening, Kimi k2.6 and deepseek are cheap asf for API but are on par with the current generation of models, and also GPT 5.5 pro beats mythos by a wide margin

[loud herald]

Thats why these chinese models are winning

Even google is open sourcing great models

[grand rivet]

china is goat bro they push Gemini claude and Open ai to give even more better models for cheaper

[loud herald]

Gemma 4 being small enough to run on my 1660 super 6GB vram is insane

And its not bad

[grand rivet]

cuz china opensources every thing

[grand rivet]

fr

[grand rivet]

true

[loud herald]

Well they are pushing them to go bankrupt

Because US models keep scaling

All of them except say gemini

[grand rivet]

nah i read that china is doing it for research not really for money

[light sleet]

Gpt img 2 it generated the joke too

[loud herald]

Right, the US companies are and they're going bankrupt because of it

[grand rivet]

us companies gonna get bankrupt cuz of China, china really doesnt cares about money

china is playing smart

[loud herald]

Not what I'm saying here

They're going bankrupt because they keep making larger models and the compute cost is too much, its not literally chinas fault

China also put 1.3T dollars into their research and technologies

Recently

[grand rivet]

truee i agree they need to adapt to good techniques or find some theirselves

[grand rivet]

yeah

[loud herald]

Thats the biggest budget in the world for technological advancements

[grand rivet]

fr they can save lot of compute power

nah i would say even more

nice joke

CGFI

[loud herald]

The IQ is dropping in here, its room temp

[grand rivet]

wth is voicebox then?

[loud herald]

I don't even look at benchmarks and will say chinese models are equal to the current US generation models lmao

[grand rivet]

i would say better they will get better with time more better then US

[loud herald]

To say they're at least 4 years behind is genuinely just a braindead statement

I don't see why you'd hate the advancement of AI so much

Just because its a different Country lol

[grand rivet]

fr bro

[loud herald]

Guys giving us vectors

[brazen relic]

Yes bro 😂

[grand rivet]

trying this new model they released (china)

[tidal stream]

3video arena

[gray lion]

i dont think gpt 5.5 pro beats mythos by a wide margin at least according to benchmarks

[grand rivet]

hes talking about the real use remember when Meta faked their benchmarks....

[loud herald]

Maybe not a wide margin but either way it beats it. And yet GPT 5.5 Pro isnt hacking into governments like they suggested could happen with mythos lol

Mythos was all hype

[loud herald]

I have a question

Who is Chatgpt21 lol

[gray lion]

idk i just found this benchmark

[loud herald]

I see, those aren't accurate but after looking at another one mythos is indeed better on all of those

Since we cant use mythos we cant test via side by side

[shrewd citrus]

is it a ElevenLabs thing

[bitter moss]

why image generation is rate limited

[red sluice]

GPT5.5 being utter crap, Mythos is not that AGI model so strong they won't release, just another marketing stunt

[echo aurora]

We have limitations in place for a sustainability purpose.

Need to be aware of spend so we can continue providing access to AI for the long-term.

[surreal pulsar]

GM guys

[echo aurora]

[grand rivet]

its voicebox

[brittle hollow]

Is it just me or is the site being really slow

[loud herald]

Its fine for me

[strong ferry]

So... Opus is showing up, but it's not working in direct chat.

[gleaming kettle]

wisprflow is secure?

[winged rune]

https://www.youtube.com/watch?v=vMeSE9DHKnI&t=387s

[echo aurora]

Can't say I'm seeing the same.

[rigid copper]

sup guys

[whole sundial]

<@&1349916362595635286>

[rigid copper]

<@&1349916362595635286> well, another hacked account

[fallow perch]

hey, everyone 🙂 hope you're doing alright
just checking, are you open to outside devs helping out?
would be glad to jump in if there’s space

[echo aurora]

Looks like the new regex rule didn't help :/

[echo aurora]

We are hiring!! Would encourage you to check out this page: https://jobs.ashbyhq.com/arena

[rigid copper]

new regex rules?

[echo aurora]

An automod rule that should have preventing that from appearing.

[frozen swift]

when it fix? ask me to login everytime

[rigid copper]

ooh, well that likely won't help, because even in heavily moderated server, that deepfakes scam can still appear.

[rigid copper]

I guess the most effective way is manual moderation since we always ping moderator roles when someone sends deepfake scam.

not sure MEE6 discord bot could help with this or not

damn, my english are trash :3

[vale quest]

Dude atleast give us back opus for this type of restriction like you just want more pay

And im starting to believe sonnet 4.6 is actually something lower

Claude doesnt know its specific model so theres no way to prove it

And especially after the sustainability announcement removing claude opus sonnet got worse

[steady grove]

HI GUYS I MADE IT
MY AGENT IS ALIVE
what the wonderful day

[echo aurora]

Yeah I've been looking around and this is an ongoing issue for lots of servers. We're still going to look/change ways to try and stop it as much as we can.

[echo aurora]

You got agent mode?!

[echo aurora]

With the new usage system we're hopeful it'll result in those models being brought back to Direct/Side by Side

[vale quest]

So

You are in charge of the development decisions ye

[echo aurora]

If you're saying we've changed the models for something worse, I can tell you that's not the case.

[vale quest]

It has decreased in precision and i have a wobbly point but still

[echo aurora]

No, as a team these decisions are made. I play a part in that, but I'm not solely making the decisions.

[loud herald]

Is it a slow rollout or a canary thing?

[echo aurora]

I'm confident we wouldn't do something like that. I'll keep an eye out for others saying the same and raise the issue if it's seeming like something is off. The people working at Arena care a lot about this community, they wouldn't intentionally do something like that.

[echo aurora]

Slow rollout, isn't yet on canary.

[loud herald]

Ahh I see

[vale quest]

Can I have

[frozen swift]

login error

[echo aurora]

Sorry to say we don't grant access to experiments on request

[echo aurora]

Don't hesitate to share your thoughts with us in #1498702173650030756 , we're looking for feedback on it

[echo aurora]

What is the error?

[echo aurora]

Is there more info you can share?

[tranquil cliff]

what's the error you are getting?

[vale quest]

Ill enable agentic feature flag i guess

[steady grove]

no, i programmed agent in typescript using local ollama and mongodb

and its finding proof of life of itself

kinda weird

[rigid copper]

at first glance, ctrl+F5 and see if it fixes.

[manic pasture]

@echo aurora hello i would be very happy and can do anything to help arena grow / manage discord server but is there a way to like use arena for free on tools like cursor or antigravity or vscode like make arena api or whatever access and edit my program code directly

[scenic holly]

alt detected

[soft verge]

Does Max model still route to the top models or only to the best available ones?

[rough heath]

Has anyone tried DeepSeek V4? What's the user experience like? And how does it compare to Sonnet 4.6???

[minor mountain]

/image to video

[thin current]

Fake

[grand raft]

time to hunt, moderators

[golden ocean]

true

[modern wedge]

I MUST BE TWEAKING CAUSE THERES NO WAY

OPUS IS BACK BRUH

[stone forum]

Refresh

[modern wedge]

Well damn is gone

Bruh

[sullen sable]

Shesh

[open axle]

What is the return deadline?

LMarena's dead 😭 next step close the site sad this...

[wraith kite]

People, is it just me or all the models disappeared from the site?

[slender ledge]

How

[muted heart]

where are all the good models like gpt 5.5 or gemini 3.1 or opus?

[west nexus]

when are we adding python in code leaderboard

[limber pivot]

I want to create an image but I get this message. I already did it, it goes away. I try to create the image again, but the same thing happens and it won't let me create the image again. I've been like this for a week, and it happens in emails when I try to confirm. I keep getting that message from the second image. I want this to end because I'm seriously getting fed up with these games. Fifth-rate programmers

[subtle rose]

Omg, 4 hacked accounts in 2.5 hours??

[obtuse smelt]

what the

[ocean minnow]

gpt-5.5-xhigh didn't last long? saw it once popping up in comparison. then in list of models, but didn't have a chance to ask it anything directly. now it's gone as well.

[rigid copper]

3 deepfake scam in last hour

[rigid copper]

4? now counts on 5

[subtle rose]

Now yeah

[obtuse smelt]

it is so much

[queen veldt]

It's not a deepfake lmao

It's inspect element

I remember people used to post these 10 times a day and they didn't know how they got hacked

Some people speculating on Fortnite hacks or something

[blissful prawn]

Again lol

[low patrol]

guys why did they remove last updated from the arena.ai ui

i want to know when leaderboards last updated

[vagrant lantern]

How do I choose the model I want exactly because models in direct chat are limited

[olive spruce]

are you referring to claude gang, if yes, its not there in direct chat

[vagrant lantern]

Oh ok

[dapper walrus]

hello

why lmarena keep asking me to create account if i was already in my account

when i try to generate pic

[short sluice]

im using an alt and it belongs to another person so theres a phone

I DONT EVEN HAVE

[dapper walrus]

hug

huh

[hollow mulch]

Guy i using claude 4.6 searching mode just few prompt and got limit token lol

[dapper walrus]

did anyone have access to seedance ?

[short sluice]

WHY WOULD MR BEAST EVEN DO THIS

[dapper walrus]

this why we have this probmlem &

with login bug

[grand raft]

https://tenor.com/view/ah-shit-here-we-go-again-ah-shit-cj-gta-gta-san-andreas-gif-13933485

[zealous gorge]

yoo guys

do you know you don't need to work physically before you start earning money
Cause I do and I earn at least $1000-$5000 daily
I can take you through the process if you don't mind though

[short sluice]

no

[stray aspen]

deepseek has vision

now

[light sleet]

now 💀

now Is a crazy term

[outer flicker]

how?

[slender ledge]

I think u need pro idk

[obtuse smelt]

uh gemini is Something went wrong with this response, please try again.

[slender ledge]

What genimi?

[obtuse smelt]

3 flash (nano-banana-2) and pro 2k

[slender ledge]

Yooo

They added new models

Gemma in text

And other

[obtuse smelt]

i see

[slender ledge]

Just do new chat

API error idk

[obtuse smelt]

hmm i try again but i doing generate images

[slender ledge]

I think the api has an error

Like the api their using

I don’t get error

3.1 works in image

[obtuse smelt]

but i retry failed again

gpt-image-1.5-high-fidelity got this too

Something went wrong with this response, please try again.

[golden ocean]

https://tenor.com/view/update-delayed-lebron-block-update-block-lebron-lebron-james-block-update-releasing-3-2-1-gif-1010810542145176969

[slender ledge]

Wym

What update

[light sleet]

famous ppl can easily generate their pfps using gpt image 2 😭

[slender ledge]

Isn’t that open ai ceo?

[light sleet]

yes

but generated by gpt image v2

so if u ask for a sam altman pic it can generate easily what I meant was that famous people that r recognized by AI can generate themselves a good pfp

cuz the AI knows them and it can generate their face

[slender ledge]

Sadly like ppl like locally famous like an famous singer In India or something doesn’t make that great

Yk

[light sleet]

Come dm

[light sleet]

Come dms

[warm mirage]

I hate this guy!

[burnt patrol]

How do I generate videos again?

[echo aurora]

Sorry to say this is a bug, our team is aware of it but I'll bump the issue again. Would note I did delete the screenshot as it had your email visible in it. We wouldn't want others having access to that email.

[echo sinew]

You can use the Video Arena in https://arena.ai/video. More information on how to use Video Arena can be found in this article.

[burnt patrol]

Thank u

[viscid mantle]

seems this page haven't updated for 3days since April 26,when will this ranking page updates ?and when will GPT 5.5 xhigh release ?

[echo aurora]

We won't have an ETA to provide sorry to say.

It just depends on how many votes we're seeing and how fast we're able to validate.

[hollow patio]

Why the image model doesn’t work ?

[viscid mantle]

I understand that you don't have an ETA, I would appreciate it if you can look at the background data roughly speaking, and is GPT5.5 xhgih in the process?

[prime karma]

@echo aurora How do I use Cloud Opus 4.7? It is not showing directly or side by side. Please help me. Please tell me.

[light sleet]

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. You can find more information in this announcement: #announcements message

-pineapple

[prime karma]

How to use these models and where to use them

[light sleet]

These models aren't available in Direct and Side by Side for now, but they're available in Battle Mode if you're lucky enough to get them.

[echo aurora]

Generally, updates will take around ~ a week for an update, but really just depends. And yes xhigh is currently in the process.

[prime karma]

Gpt 5.5

[echo aurora]

It's the same for 5.5.

[light sleet]

yep

[echo aurora]

In the announcement post we clarify it's in Battle mode only - #announcements message

[prime karma]

This also cannot be done directly or side by side.

[viscid mantle]

Does “by Saturday” mean available in battle mode only, or officially appearing on the Text Arena Overall leaderboard?

[light sleet]

since when did qwen 3.6 max come

No announcement for it???

[hollow patio]

?!

[viscid mantle]

so, April 26 is the last for Text Arena overall ranking in April ,Right ?

[median smelt]

@light sleet

Sora 2

[echo aurora]

That's odd, can you hard refresh the site or try a different browser? Is it just image where this isn't appearing?

[echo aurora]

I couldn't say.

[echo aurora]

Sometimes we don't put out announcements for new added models.

[light sleet]

Tufffff

[hollow patio]

Yes its working now
But why i always get this problem ?
@echo aurora

[echo aurora]

Thanks for sharing this Trace, I don't yet have an answer for why this failed, but it is being looked into and we will create a bug report for it. For future refrence, if you come accross this error I would encourage you to post in #1417174113092374689

You'll find some troubleshooting steps in there.

[fiery gull]

what you thinking about deepseek v4? in start I did so rude about v4, but I1m using it and liking the deepseek v4 to make my documents

[dapper walrus]

all good

[stray aspen]

it feels better than launch

[fiery gull]

I love to do a benchmark asking a pharse can I read in 2 languages

[stray aspen]

but its still not as good as mimo

[thin knot]

is this just for me or once you use a chat for long enough it just starts giving you the traceid and doesn't respond to the question

[fiery gull]

to me mimo sucks

the portuguese from mimo sucks

worse that my english 🤣

[split kayak]

JohnPork

[strong ferry]

Please give us Opus 4.7!!! 😭 🙏

[light sleet]

No

they wont give the USA opus 4.7 😔

[vernal saddle]

opus 4.7 is better in the USA than in other country ?

[echo aurora]

This can happen for various reasons. I'd check out [this message](#1417174113092374689 message), which goes into more detail about what may be happening and includes some steps you can try to resolve the issue.

[silent tree]

bro got alot of reactions

[strong ferry]

I know 😂

Wtf

[vernal saddle]

Has my idea about a music arena been accepted or not?
(or did you forget it)

[light sleet]

use sonauto its better than suno 100%

[compact comet]

pay for it

[vernal saddle]

I use lyria 3 pro personnaly

[light sleet]

It has synthid tho

😔

I couldve used lyria 3 pro too cuz I have alot of credits in Google Flow Music (formerly Producer, Riffusion)

[strong ferry]

https://klipy.com/gifs/i-dont-have-any-money-im-poor--k01KQD17RTG64GPMKYYQDBWFA3C

[light sleet]

then get money

[strong ferry]

Trying but job hunting is difficult.

[vernal saddle]

I think nobody is searching to see if SynthID is on a music

[echo aurora]

This may be an Arena we create in the future. I couldn't share more details about what may be upcoming.

[light sleet]

im excited for the database feature

and the environment variables

[light sleet]

what if u get famous

[vernal saddle]

ok thanks for the answer

[light sleet]

and someone uploads ur music to an AI

And ur cooked

🤯

Sonauto

[vernal saddle]

don't be famous then

[light sleet]

but then u get famous

[vernal saddle]

delete ur account

[light sleet]

but then your account magically comes bakc

back

[vernal saddle]

skill issue then

[light sleet]

but then the issue gets fixed

[vernal saddle]

But if you don't have the skill, they can't be fixed

[light sleet]

But the skill is fixed

[stray aspen]

the new ernie is bad

[sleek urchin]

Holy

[inner relic]

Why

i swear bro, you hate everything

[sleek urchin]

https://tenor.com/view/why-is-he-gif-3749724709641675358

[inner relic]

wth..

[sleek urchin]

Ernie is a really good model

[echo aurora]

In what way? Any specific area you're referring to here?

[sleek urchin]

And its Direct

[sleek urchin]

• its better than sonnet 4.6

[inner relic]

Hm

does erine has better instruction following

and fresh writting?

[sleek urchin]

Dunno

[slender ledge]

w or L image made by gpt img 2

[sleek urchin]

Big L

[slender ledge]

howww

[sleek urchin]

Bad quality (text)

[slender ledge]

i did mini prompt

5 words

idk

10

son 😭 🙏 blud is not omni man just a bit

[light sleet]

@echo aurora Image 2 made a good illusion

Stare at it for like 30 seconds and look around

[thin knot]

hi, ive tried the steps

is there a way i could like open a ticket or something of the sort

[echo aurora]

Lol pass, getting dizzy just glancing at it

[echo aurora]

What was the prompt?

[echo aurora]

I'd encourage you to share the Trace ID in that channel. I go through these each day to investigate and share with the team.

[thin knot]

:b5c56ce6-bd6d-

thank you!

[echo aurora]

Kk going to forward to #1417174113092374689 and will followup there.

[light sleet]

Illusion Image to make us think that it's moving when we move

Weird prompt lol

[echo aurora]

Just me or is https://arena.ai/ not loading?

[rigid pasture]

me too

[echo aurora]

Eventually loaded, but took awhile.

[light sleet]

tuff background gpt generated me

and a good wallpaper

Prompt: Abstract soft blue gradient background, smooth flowing blurred shapes, gentle organic curves, dreamy ethereal atmosphere, soft pastel tones, very shallow depth of field, smooth bokeh-like transitions, minimalist composition, high quality, serene and calming mood

[echo aurora]

Assistant A or Assistant B?

[light sleet]

Idk I generated in direct

[echo aurora]

I'm going with A, feels like it has more depth.

[light sleet]

Yeah

[thin knot]

B looks like the average design on plates/bowls in my country

[storm dust]

same because B looks more perfect

it needs some distortion

[echo aurora]

That sounds charming, my plates are just grey and boring

flux-2-flex (A) & flux-2-max (B)

[rugged abyss]

I dont see the it moving

[echo aurora]

I liked the design on B a lot more, + the color was a nice touch. But IMO "make us think that it's moving " was better achieved with A.

[sleek urchin]

A

NAH

Expland text

Expand*

Anyone wanna see my sticker?

[rocky geyser]

Yea i have been loving agent mode so far!

And the new backend and api features in code arena have been really cool too!

Tho they have a few bugs

[sleek urchin]

What is it?

[random lotus]

I just joined, how can I start creating videos?

[sleek urchin]

https://arena.ai > Battle Mode > Video generational button toggle

[echo aurora]

Here is a direct link -> https://arena.ai/video

[light sleet]

can I generate u a wallpaper

but with pineapples

[sleek urchin]

Woah

Pineapple is there a thing to generate android application?

[echo aurora]

Perfection

[echo aurora]

Code Arena is going to be your best bet for front-end work - https://arena.ai/code We are in the process of expanding these capabilities to full stack https://help.arena.ai/articles/7138211024-arena-experiments-code-arena-fullstack

[light sleet]

my new pfp thanks image 2

[echo aurora]

Ofc, fire away

[sleek urchin]

I GOT AN IDEA!

#1372230675914031105

[stray aspen]

its just bad lmao

i tested it and it sucks

[sleek urchin]

How do you use imagine 2?

[echo aurora]

It's in Direct/Side by. Side, but you'll need to scroll down the list a bit before seeing it.

The dropdown list will eventually be updated to reflect the leaderboard.

[light sleet]

Yeah but I kinda use it through the original gpt too

[echo aurora]

Big fan. Very tropical

[light sleet]

Yeah

[sleek urchin]

Nvm I see it

[light sleet]

I'm the gpt... myself

[echo aurora]

Yeah image instead of imagine + the -

[light sleet]

ask me to generate stuff

im banana-image-2 powered by gpt image 2

but renameddddddddd

[short sluice]

image models disappeared

I was trying to use a model when this happened

Do fix

[echo aurora]

Can you hard refresh the site? It's just image models that are missing? Are you signed in?

[short sluice]

Ok now they're back but I got the experience the frontier text of doom

Refreshed everything is normal again

[echo aurora]

Yeah saw another user mention the same earlier. There was some lag with the site earlier too, which I think is related.

[raw laurel]

So what are you guys using for free alternatives now that Arena got nerfed? Deepseek, ChatGPT, Claude?

[soft river]

DeepSeek or glm

but I think I’ll use ernie more now

Nice it’s the best Chinese model now

[raw laurel]

What’s glm and ernie? /noob

@soft river

[soft river]

These are Chinese models

Like Claude and ChatGPT

They have their own companies improving their models

[raw laurel]

chat.z.ai is glm? How does it compare to the Chinese Deepseek?

[soft river]

Glm is smarter overall

I have to look at the benchmarks

But I think glm was better at the majority of things

[raw laurel]

Damn didn’t know. Though Deepseek was the best of the Chinese models. Thanks for the top. Will check them out.

[soft river]

DeepSeek is cheaper

[soft river]

Of course

I’d recommend you checking Ernie too

From Baidu

[wet steeple]

anyone now how to generate adult content with arena.ai ?

[soft river]

It’s a new model so it’s always good to test it and maybe you like it more

[soft river]

You can’t

Well

[raw laurel]

Thanks. You know if there’s a iOS app for GLM?

[soft river]

You can change the words

So youuu don’t trigger the filter

But overall

I don’t recommend it

It’s for testing

Not roleplay

[soft river]

Nop

[raw laurel]

Ok thx

[soft river]

I believe it’s the only ai company without one

[wet steeple]

so i can't generate p**n stories, photos or videos on this website ?

[soft river]

Sadly no