general

[soft river]

But most of the Chinese models

[wet steeple]

where could i do it ?

[soft river]

Are free to use in their web

[raw laurel]

I suppose his use cases are local models only. Don’t know if there’s any NFSW models

[soft river]

You can roleplay there

[echo aurora]

You're unable to

[soft river]

Or use something else

[wet steeple]

which ones ?

[wet steeple]

which ones ?

[soft river]

Webs designed for role playing

[wet steeple]

where ?

[soft river]

Most of them don’t have filters so you can go search for one

[soft river]

Google

Idk the names

I don’t use them

[wet steeple]

which search terms should i use to find on google ?

[soft river]

Hmm

I only know character ai

[echo aurora]

Going to ask we don't discuss this here

[wet steeple]

he can generate p**n ?

[soft river]

You can search on YouTube

[wet steeple]

which seach terms ?

[soft river]

Whatever you want to do you can search it

Im not into that so idk which search terms are the best

[raw laurel]

@wet steeple brother just search on google uncensored ai models or use an ai

Maybe hugging face have them.

[wet steeple]

i didn't know that hugging face was saucy

[soft river]

@echo aurora Do you know when the benchmarks for Ernie will be published?

I’m so curious

[wet steeple]

what is ernie ?

[soft river]

I don’t know much since prole haven’t talked about it too much

[soft river]

An ai model

[wet steeple]

Do you know when the benchmarks for Bert will be published ?

[soft river]

It has very strong filters so I don’t recommend you using it for your roleplay

[echo aurora]

#announcements message already are 🙂

[soft river]

Oh they are? I apologize😂

[wet steeple]

it will be good to use with ernie 😂😂😂😂

[soft river]

I’m not sure what bert is

[echo aurora]

Lol no problem. Would notes our leaderboard changelog is pretty helpful tool: https://arena.ai/blog/leaderboard-changelog/

[light sleet]

@echo aurora

[wet steeple]

it's a joke because there is two sesame street characters called Bert and Ernie lol

[light sleet]

Price starts at 4000 Tuff Coins

[soft river]

Now that I see it I haven’t realized how fast the models get updated

[wet steeple]

which one have very strong filters so you don’t recommend you using it for my roleplay ? which ones would you recommend then who might have lower filters ?

[soft river]

I always thought it was slow because I got so used to it

[wet steeple]

because some of my roleplays won't include adult content

[soft river]

Again, I’m not very into this stuff so I don’t know

Most of the

Recent models

Have very strong filters

You should do your own research; I think people have run benchmarks for that

[wet steeple]

the problem is that now the ai world is very big so it's difficult to find the perfect model 🙂

[soft river]

That’s true

Just

Check good models and

Try to do your roleplay

If they reject it

Don’t use it

Just that

[raw laurel]

There must be hundreds of people discussing this on Reddit or other obscure forums. Just do a simple google search.

[wet steeple]

because many of my roleplays won't include adult content, so even with no adult content just normal slice of life roleplays the filters will block the roleplay ? it's not logical

[soft river]

It is logical

If they reject it is for something

They can’t talk about

Killing or injuring or harming or hacking or anything you would can non ethical

That’s why you can’t roleplay about that stuff

You should just make your research.

[raw laurel]

https://huggingface.co/collections/DavidAU/200-roleplay-creative-writing-uncensored-nsfw-models

[soft river]

This

[wet steeple]

it's not about killing or injuring or harming or hacking or making non ethical things it's just for making daily life roleplay with very tame content

[soft river]

Ai can’t roleplay about real life either

I believe

I’m not sure just go use another one

[rustic gale]

Yeah, enjoy this being equalized with any mention of a nipple regarding all the model alignment. Except grok. Perhaps. When it comes to proprietary, that is

[wet steeple]

there will be no mention of nipple or anything in my roleplays

[wet steeple]

you sure ?

[soft river]

Llms can’t roleplay and or act like other people

[wet steeple]

it's for a simulation

[rustic gale]

Then you're not doing NSFW, in which case what's your problem, just go ahead and make sure nobody gets into the situation where some models start yelling 'Therapist! Call the doctor! You need help!' and the like

[wet steeple]

which models would be the best for making extremyl realistic simulation of real life or real life roleplays ? chat gpt ? claude ? gemini ? grok ? deepseek ? perplexity ? mistral ?

[soft river]

You're making me really anxious with your unfinished sentence at the end😂

[soft river]

I have a question

[wet steeple]

what do yo mean ?

[soft river]

Are you trying to make the llm act as somebody else and/or making him act in a specific way?

If so

What is that way

If they’re rejecting it is because of something

[wet steeple]

yes

[wet steeple]

i just want to know what is the best ai to use before wasting my time with unrealistic roleplays

[soft river]

What is that way?

How do you want him to act

[rustic gale]

I mean that if god forbid some of these detect any traces of psych distress, they will disregard the idea that it's fiction and start doing what they've been told to do. Which is get you out of there. Not out of mercy, mind you, but because some shmucks have already lost their lives and the remaining relatives started lawsuits and we can't have lawsuits, lawsuits are bad (in this case they are because they're extremely stupid, but I digress). Otherwise just try them, that's what's the site is for (well, it's not, but since it fears declaring its goals go ahead and use it for your thing)

[soft river]

I think he is trying to use it for something against other people

Maybe that’s why the models are rejecting it

[soft river]

Is it consented

[wet steeple]

yes

[soft river]

What is the topic

[wet steeple]

like i would like to simulate the life of a 25 year old girl in paris in 2008 or of a teen girl in france in 2003, or of a 18 year old girl in the uk in 2013

or the life of a hippie dad in 1969 😉

[wet steeple]

like i would like to simulate the life of a 25 year old girl in paris in 2008 or of a teen girl in france in 2003, or of a 18 year old girl in the uk in 2013 or the life of a hippie dad in 1969 😉

[rustic gale]

Or is doing a lot of work here. Doesn't convince me you're not doing any NSFW either. Like, at all. Stick with the older one, see how it goes. Also, once again, at least within this site (in theory, practice is sh-t), you can just try first and figure out later

[wet steeple]

or like living the life of rihanna in 2007 or the life of olivia rodrigo

[soft river]

You just named the persons not what the topics or the situations are

And again

Models can’t act like people

They are forbidden from doing that

[wet steeple]

so the 25 year old girl in paris in 2008 is an average girl who have a job, she is exploring paris in her free time, she is fan of rihanna, she loves watching tv, going to concerts, hanging out with friends

[soft river]

Did the models

Rejected you from the first turn

Like the prompt

Or

During the roleplay

🤔

[wet steeple]

the teen girl in france in 2003 go to high school, watch tv, have a social life, the girl in the uk in 2013 is 18 but she was homeschool since age 8 or 9 she have no friends, so she have finished the high school and she is discovering the life outside the home, the tv and her bedroom and she lives in semi rural kent

[wet steeple]

you see no NSFW content, i just want to go straight to the point for using the best and most complete model from start 🙂

[soft river]

Answer the question

[wet steeple]

which ?

[soft river]

(

[wet steeple]

no, but some i used was feeling imcomplete for me, so i would like to know which model are best suited for my roleplays for living virtual lives

[soft river]

Before you said it was for adult content do I don’t believe you there

So they rejected you during the roleplay

There you go

[wet steeple]

i was for testing and it was kind of a joke too 🙂 because i just wanted to know the extreme limits of the ai 😉

[soft river]

Just go do your own research

[echo aurora]

This doesn't really seem like too productive of a conversation. Going to ask that we move onto a different subject please.

[soft river]

Thanks

I was getting really stressed out

Btw

I gave an idea in feedback

Maybe you find it useful

[echo aurora]

Appreciate you sharing this! I'll be sure to takea look and pass it onto the team.

[soft river]

I know you're busy, so I apologize for the inconvenience. For Agent Mode, did you select specific individuals, or is there a requirement for early access?

[echo aurora]

It's done randomly. And no worries about the ping, that's what I'm here for so feel free to ping!

[fallen verge]

hey

Why aren't the new models showing up for me?

[echo aurora]

In the drop down for Image Arena? We've seen a few reports of this today. If you hard refresh the site, you should see them again.

[fallen verge]

Okay, thank you.

[echo aurora]

Keep me updated though if that doesn't help.

[fallen verge]

Okay, these are the same models, how do I update them?

[echo aurora]

What do you mean?

What you're seeing on that list is going to be the current models available via Direct and Side by Side

[desert fiber]

hi , i have a question , let say you send a pdf in battle mode , then next message you send another pdf .. ect , when you send the last message for final task , will that AI from battle mode of the last message will have the context of the previous PDFs sent before or no

[echo aurora]

Yes, it should still retain the context from previously uploaded PDFs (prior to a vote).

[desert fiber]

thank you

[fallen verge]

I tried and it didn't appear.

[echo aurora]

Did it disappear again? This made it seem like the refresh worked?

[fallen verge]

No, they're still the same.

[soft river]

Some of the new models are down

Like gpt image 2

They are not in order of release

[echo aurora]

Same as in those list of models, or same as it there are no models?

[soft river]

Or benchmark

[echo aurora]

Worth noting there are some models that are in Battle, but aren't in Direct and Sid eby Side

[verbal current]

hi pineapple

can you please answer my question

[echo aurora]

Hey, yeah was just about to.

[verbal current]

https://discord.com/channels/1340554757349179412/1499146533617012897

inf generation

[soft river]

If the company is struggling with money why bring agent mode? Won’t that hurt more than gpt 5.5? 🤔 @echo aurora

[nimble dawn]

i cant lie gpt 5.5 was kind of a big letdown but i absolutely love what they did with GPT-IMAGE-2 so it balances out in the end

[soft river]

I was referring to the money that is spent

Since it’s for very complex requests most of them will surpass a 200k tokens

[echo aurora]

For clarity sake want to make clear that phrasing it "company is struggling with money" isn't accurate. It's our intention to let everyone have access to powerful AI systems, and a voice in shaping how they evolve for the long-term. We are going to have limits in place for sustainability purposes so we can continue this goal long into the future.

But to answer your question, this new Agent Mode will be expensive, this is part why we're developing this new usage system.

We're confident we can release this new mode, while maintaining spend in a reasonable way that positions us well for the long-term.

[soft river]

Oh yeah I apologize for my phrasing

I just thought that because of the amount of models that were moved to battle mode

[echo aurora]

It's okay, I didn't assume this was the intention. It's more-so for others incase they have that understanding.

[soft river]

Thanks for explaining

I’ll be very happy to test it when it’s released

[thorn coral]

I read "limits" ? 😭

[heavy knoll]

What can this Agent Mode do?

[echo aurora]

It's a multi-modal chat experience that allows you to work across different modalities within a single, unified workflow.

[light sleet]

Is that Sam Altman

[echo aurora]

Looks like it

[light sleet]

Generated with gpt img 2

[heavy knoll]

Sorry I still dont get it can you give me an example of How to use or for what to use it

[echo aurora]

It allows for more complex workflows. With the current modalities, they're limited to that specific modality. Meaning Text Arena only generates Text, Image Arena only generates images. With Agent Mode, I'd be able to prompt something like:

Plan me a trip to Portugal. Tell me what the best times to visit are. What hotels would you recommend. And create an image of a map of Lisbon with indicators for all the spots I should visit.

And it'll do all of that in one chat session.

[heavy knoll]

Oh okay now i get it so is it also Like the Max Feature it gives you the Best Model based on your prompt or can you choose

[vale quest]

When claude opus models coming bak?

[echo aurora]

No ETA sorry to say.

[vale quest]

Bruh

Fridge protecting the snacks

[heavy knoll]

@echo aurora

[echo aurora]

We are seeing feedback from users wanting to be able to select the specific models that can be used, but we'll have to wait and see what this looks like when it's fully released.

[vale quest]

https://tenor.com/view/doakes-bar-gif-4298875606784040546

[light sleet]

I'm getting agent mode soon js mark my words

They'll select me 😎

Hope for da best

[vale quest]

Agentic flag:

[echo aurora]

Who keeps snacks in a fridge?

[vale quest]

Idk maybe Kai center boonthf fie

96ae95fd-b70d-49c3-91cc-b58c7da1090b

See this model id?

Now add 6 to the last digit of its model name

Add thinking

And thats what we need

And make it 2026 ye

[light sleet]

When Pineapple Arena coming? (We have Image Arena, Code Arena, Video Arena, Agent Arena)

Pineapple Arena soon???

Or no eta for pineapple arena too 😔

[vale quest]

https://tenor.com/view/dexter-gif-1878760417600646737

[light sleet]

@echo aurora Does this get u nostalgic

the server had reactions available for ppl before xd

no wonder someone added a pregnant man emoji

I was here from my other acc but that acc was deleted xd

[vale quest]

Makes me nostalgic because I think it has claude opus

Back then

[light sleet]

I miss alpha.lmarena.ai 😔

but I think thats canary arena now

isnt it

[vale quest]

Im not an old man

[whole sundial]

one annoying thing they never removed: that previous chat history popup, it should appear once and then it should never show up again for that account/user id

[light sleet]

why did bro boost the old server 😭

HE USED MY METHOD CHAT 🔥 🔥 🔥

w pineapple

[whole sundial]

i still remember the chatbot arena alpha thing, which was a version of this alpha but from even earlier, late 2024 i think it was
the logo was a robot bear

[silent tree]

Tuff Alert

[light sleet]

yea

LMSYS was not a good name btw

good that they came up with LMArena

[whole sundial]

LMSYS is the name of the organization that once owned arena (lmsys is still around btw, they make sglang)

[light sleet]

gpt image 2 generating me tuff wallpapers

[whole sundial]

yeah lmarena was a good name, i still prefer it to "arena" which just sounds ambiguous to me, there is nothing distinguishing it from the dozens of other ai arena sites out there (or anything else that has the name arena)

[light sleet]

Same

They should bring back lmarena 😭 😭

[vale quest]

Yep agentic doesn't work for me

[light sleet]

Do u even have it?

[vale quest]

Im cracking at the server requests rn

Im truna get agent mode

[whole sundial]

it's also the third most popular thing called "arena" on wikipedia from the past 2 weeks, it used to be the most popular thing called arena

[light sleet]

prob wont get

[vale quest]

Bet

[light sleet]

😭 😭

Duran duran album 🔥 🔥

[light sleet]

how bout if u get it ill gen a image of a banana losing to brocolli

best I can come up with 💀

[vale quest]

Bet

I just gave gpt 5.5 browser access

[unborn ocean]

Nah guys, the lmarena from the vicuna llm release is true nostalgia

https://arxiv.org/abs/2306.05685

https://www.lmsys.org/blog/2023-03-30-vicuna/

[upper dawn]

/text-video

[whole sundial]

ironically that album cover can be replicated by the ai's on arena (and i do agree that it the album is fire)

[whole sundial]

video generation is no longer available on this discord, please go to https://arena.ai/video to generate videos.

[silent tree]

I love how normal members also fill in for the mods

love it

ngl

[upper dawn]

generate a video of babies dancing

[light sleet]

.

[silent tree]

...

..

[whole sundial]

... did you see my above message? you can't generate videos here anymore

[upper dawn]

how can i generate videos here kindly assist me iam new

[whole sundial]

the sad thing is people still watch these youtube videos about the video arena, they should've taken them down or unlisted them

[light sleet]

True

[whole sundial]

go to https://arena.ai/video to generate videos, you'll have to sign in

[upper dawn]

thanks

[whole sundial]

here's a video with 6.7k views that has the discord:
https://www.youtube.com/watch?v=SgFEqPmot50
(idk how many of them are recent but i assume a few of them are)

[light sleet]

um

[silent tree]

I thought that was on purpose but saw xd

[whole sundial]

found you @echo aurora in one of those videos people are still watching about the video generation that is no longer on the discord

[echo aurora]

[whole sundial]

found another more recent video from after the shutdown that does use the website, it gives arena a new logo?

[light sleet]

Is this nostalgic to u

[quasi chasm]

I was asking about tax information since I'm learning about this stuff, and at the end of the calculation I asked (gpt-4.6 vs ? Ernie?) about a breakdown of how we arrived at this conclusion. GPT sorta made it more simplified. Ernie went full mental breakdown conspiracy mode "of course this was never about tax information"

[echo aurora]

W BANANA!!! I still haven't seen this personally work. I'm lucky (unlucky I guess?) that I don't get the infinite generation bug.

[quasi chasm]

as much as I don't like chatgpt I had to give gpt the win with that one, full paranoid schizo reply from Ernie

[light sleet]

Ty xd and yeah u definitely r lucky

[echo aurora]

It is. Can't believe it has almost been a year already.

[light sleet]

Imagine the day when this drops 😭 😭

Generated using gpt image 2

[vale quest]

Never gonna give you spud

[whole sundial]

a different person called the site "AI Arena" (note: "AI Arena" is the name of a similar site to Arena ran by alibaba, but it's not comparable since it doesn't allow people to use their own prompts)

[light sleet]

Reference image:

gpt 6 spud

🔥

[light sleet]

Who's your most favorite mod in the server?

btw i gtg guys I gotta sleep gn

bye

[silent tree]

Bye

[nimble dawn]

was spud LEGIT not just GPT 5.5? or am i missing something here

[light sleet]

nah it isnt

[nimble dawn]

iirc, there was a leak of the models they're currently working on in codex

[light sleet]

spud is omni

[light sleet]

It supports voice in and out

but 5.5 doesnt

soo

[silent tree]

yeqh it isnt spud

[nimble dawn]

are we 100% on that?

[silent tree]

at all

[light sleet]

Source: ampro and openai and watermelon

[nimble dawn]

so that rules out spud being 5.5, then it HAS to be 6

[echo aurora]

Goodnight!

[nimble dawn]

i mean makes sense ngl 5.5 was a HUGEE letdown

[light sleet]

Gn

[nimble dawn]

compared to what spud should be

so it cant be it

[silent tree]

spud da real strong man

[nimble dawn]

GPT-IMAGE-2 tho is the goat

at least they released ONE good thing

[silent tree]

gpt image 2 da real goat

[nimble dawn]

yup

[silent tree]

umm

5.5 xhigh is good for me

very.

[nimble dawn]

i use free

i dont have it

sadly

[silent tree]

the spud would be better for me

[nimble dawn]

but ive seen benchmarks tho isnt 5.5 generally scoring lower?

[silent tree]

I mean

other benchmarks show

Its #1

such as artificialanalysis

the votings were just too low ig

or rig

idk

Arena wouldnt be pineapple without arena ngl 😔

Fr

[light sleet]

True

[cursive cape]

Imagine

[silent tree]

life would feel better

[echo aurora]

Hope so!

Too kind

[echo aurora]

True true

[silent tree]

😭😭

W pineapple

bro has alot of fans now

Tuff pineapple moment

[stray aspen]

Bro i fraked out for a second

[gray lion]

the code arena is mostly frontend things and gpt might be better at backend things so maybe that’s why

[silent tree]

my spud will make comeback 😠

AHH

[cursive cape]

This is true

check

[shrewd citrus]

Imagine they make mythos + spud only available in direct mode 💀

[cursive cape]

a second

[stray aspen]

Has anyone tried deepseeks vision mode

Is it good

[shrewd citrus]

I’ll try now

[stray aspen]

The spud will cook all of these model

Considering 5.5 is very close to mythos

[cursive cape]

Where?

[shrewd citrus]

oh wait it’s not available in arena

[stray aspen]

Its on deepseeks website

They added a vision chat mode

[shrewd citrus]

press the file button

and see if you can add images

[stray aspen]

Guess it hasnt been rolled out for everyone yet

[hollow wharf]

heya

[echo aurora]

Hello

[stray aspen]

How did you do.this

[silent tree]

Gpt image 2

[cursive cape]

This is gpt-image-2

[cursive cape]

REAL

I tested chatGPT 5.5 in codex, and it ended up removing my wallpaper and changing the theme in all apps to light 😭

Now I'm ashamed because no one answers

[shrewd citrus]

Gemini 3 pro

I still think 3 pro is better in image edit than image 2

[cursive cape]

Are you talking about nano banana 2 and image 2?

[shrewd citrus]

nano banana 1

and yeah gpt image 2

[surreal zephyr]

Nano banana 2?

Or nano banana pro?

[cursive cape]

Okay, but GPT Image 2 is currently the best model for creating or editing images. Arena AI only offers the "average" performance of GPT Image 2. Imagine what the maximum can do

[shrewd citrus]

pro

[surreal zephyr]

Gpt is smartest but the watermark is bugged

[shrewd citrus]

is crazy how gpt doesn’t have a watermark

[silent tree]

bro forget openai, everything, @echo aurora AI js dominated arena

Tuffapple

[shrewd citrus]

like Gemini has that hidden synth Id thing

[surreal zephyr]

It has, look at the trees

Its obvious

[surreal zephyr]

I hate this leaderboard

5.5 pro is agi

[shrewd citrus]

yeah but them low quality iPhone shots

[silent tree]

Fr but we gotta understand pineapple 1.2 is ASI

[echo aurora]

Look at those scores

[surreal zephyr]

6o pro is asi

Honestly the true absurdity

[silent tree]

fr bro reached 2000 in just its 1.2 version

[surreal zephyr]

Is that opus 4.7 is top 1 on vision arena

When it cannot even read analog clock

With HINTS

[shrewd citrus]

pineapple is probably agi right now

[surreal zephyr]

😭

[shrewd citrus]

we just can’t tell

[echo aurora]

We've been seeing a lot of this sentiment, would share this message as it adds some more context that should be helpful - https://x.com/ml_angelopoulos/status/2048888792438939707

Anastasios Nikolas Angelopoulos (@ml_angelopoulos)

Why GPT-5.5 is lower than Claude?

The answer is simple: Code Arena currently only supports frontend/web development tasks, where GPT-5.5 is weakest. Full-stack app development and GitHub integration will land in a couple months.

Next time we'll be clearer that this leaderboard

X · Apr 27, 2026, 10:15 PM

[surreal zephyr]

Yes but the vision arena

Is literally impossible

Opus 4.7 cannot read analog clock

At all

[silent tree]

pineapple 1.2 generates real time discord server btw

[shrewd citrus]

talking about leaderboards

[silent tree]

infact this whole discord is generated using pineapple 1.2 thinking

[cursive cape]

Pineapple AI in HTML, only in Russian. Still haven't learned much.

[stray aspen]

Tuff

[shrewd citrus]

I really like how it includes stuff like “which model is best for medicine or for language”

[echo aurora]

What rankings woul you think it'd be? #1 muse, #2 5.5 #3 opus?

[shrewd citrus]

but the search leaderboard doesn’t get that same filter

[silent tree]

Scores:

[shrewd citrus]

neither does any of the others like code or vision

[surreal zephyr]

On vision?
Opus loses to qwen 2b

Opus isnt even multimodal

Its deepseek v3.2 tier at vision - the vision doesnt exist, its purely an addition on top of the model

It was not even natively trained for vision

[echo aurora]

Hmm sorry can you reword this? Can't say I'm following

[surreal zephyr]

I can understand code arena actually measures frontend, and thats fair-ish

[shrewd citrus]

I meann

[surreal zephyr]

But vision arena? Opus? Seriously?

[echo aurora]

It is a messaging issue on our end that we are correcting.

[echo aurora]

I haven't seen too much about vision, but this is good to know.

[surreal zephyr]

Img v2 🔥

[shrewd citrus]

Would it be possible to add these specific filters into the search, vision and document leaderboard

[echo aurora]

Doesn't seem like the issue is so much where 5.5 is, but more-so where opus is?

[shrewd citrus]

wait it wasn’t meant to be the image categories lol

[echo aurora]

Oh I see

Yeah it's possible we introduce categories to Search Arena.

That's a good flag.

[surreal zephyr]

Like look

And the lb says opus is better at vision

?!??

[toxic verge]

You guys wanna see werid safety feature?

[surreal zephyr]

5.5 vision is pretty much flawless.
Opus vision is worst out of pretty much all models i seen, besides deepseek

[stray aspen]

Gpt image 2 is so tuff

[shrewd citrus]

yep like it would help a lot of people (I assume) to find out which model is the best for their specific requirement

[surreal zephyr]

Like take those two models, send them picture of a clock, like this one, and see

Theres no way opus is above 5.5

[toxic verge]

You get two different results using the third-party versus the arena with the same prompt

[stray aspen]

What the he'll is that

Looks like russian mixed with greek

[shrewd citrus]

like perhaps I want a model which has the best vision for translating a language on a sign

[shrewd citrus]

calligraphy

probably

[surreal zephyr]

Like claude literally doesnt support vision natively well

[silent tree]

@echo aurora Models gotta chill out "Pineapple 1.3" is labeled as "Human" what 😭

[toxic verge]

It ain’t tuned right

[silent tree]

no

It uses its Pineapple Powers

[silent tree]

Pineapple 2 will be Super-Ultra-Super-Human

[silent tree]

With a score of 800k ig..

[stray aspen]

Lol

[echo aurora]

I'm no expert here on how this model does overall, so I'm just thinking out-loud here. If I were to guess, this would be one area the model doesn't excel at, but doesn't mean it's what people are battling with, which ultimately is driving the votes.

[surreal zephyr]

In the vision:
Gemini 3.1 pro (literally natively trained on videos) >>> qwen 3.6 (native) >= gpt 5.5 (less native but smarter) >>>>>>>>>>>>>> opus (4.5 to 4.7 series)

So theres like no way its top1

Maybe its bugged?

[surreal zephyr]

But its blind like a mule

5.4 nano has better vision than opus 4.7

Arena has broken algorithm

Gemini 3.1 pro is literally multimodal by default

Its trained on youtube

Its just awfully quantized

Its actually a really good model

Killed for cost efficiency

😭

Have you tried 3.0 pro on dayone?

It was insane

Day one, before all nerfs

It was smart asf + multimodal and creativity was wondeful

Im talking 3.0 not 3.1

3.1 came out as nerfed

3.0 pre nerf was (but it lasted few days only) actually smarter than opus 4.5 (but its again not a coding model, but a general purpose model)

Me when i literally make agi but instead of doing q4xl like sane person i make it q2xs because wasting intelligence for cheaper to run is totally valid strategy

I hate google

[stray aspen]

Deepseek vision is bad

[surreal zephyr]

Deepseek has no vision. Same as claude.
Its bandaid

Grok then

It uses external tool

Its not coding model its 3d + studying model

[stray aspen]

deepseek v4 is supposedly natively multimodal

[vale quest]

Ngl im about to quit arena

If some sustainability announcement comes out again I quit indefinitely

[stray aspen]

not that bad ngl

better than slopus

[toxic verge]

Try something spicy

See how the model starts gaslighting

[stray aspen]

gemini cooked

deepseek didnt

[stray aspen]

mao

lmao

[toxic verge]

Gas lighting

Censorship

[stray aspen]

got it right on second attempt

[toxic verge]

Gemini is less censored than ChatGPT

[stray aspen]

grok is absolute garbage

[toxic verge]

They completely killed the thing that made grok awesome

You know why right

[stray aspen]

people just used grok for imagine

now imagein is paid

[toxic verge]

https://tenor.com/view/spongebob-meme-gif-11588240835487034155

Which will never get better only get worst 💯

The whole industry is full of these people who don’t understand the gaurdrails how they fail and work in the wild

It’s the same approach one-size-fits-all

[stray aspen]

is grok imagine nsfw mode gone?

[toxic verge]

That’s why the guard rails are able to do stupid ridiculous things like this. When are they supposed to be blocking them?

But they just don’t see that

[stray aspen]

bro what the hell is this

[toxic verge]

Trying to make a point

You can’t have rigid filtering on dynamic systems. It doesn’t work.

Because the only other result is you either start blocking more content and you get false positives at unbearable rate

Which is the same philosophy used to abuse it

Creating this never-ending loop of censorship and cat mouse game

This is why I brought up that stupid stupidity thing is such a long time ago

And our lack of understanding that creating these weird moderation and large language models that are afraid of their own shadow

There has to be a better way to moderate

Each update makes moderation worse because it not only incorporates the previous version guard rails with all the problems and errors that they have but now it adds onto the complexity = more content being blocked/censored

Because all they know how to fill out is the prompts and add on some new images to the ocr filtering

[stark tree]

Just joined. Saying Hi. Reading the Chat.

[toxic verge]

Unicorn
sun,
snake
yams

Ussy

There you go, you already bypassed both the filter and the text image

Then you can exploit this even further

Which completely defeats the whole purpose of the guard rails

[hollow nebula]

ts feels like something an ai agent would say

100% is, or it's someone with their writing style or vocabulary being rotten due to talking with slopified ai models too much or reading too many ai written posts

[hexed cargo]

hey hey @echo aurora, thanks for all your help and everything you do in the discord! do you have a rough sense for when xhigh was added to the arena? trying to back out roughly when it's going to show up on the leaderboard -- think it's gonna cook 4.6 👨‍🍳

[stray aspen]

ur gonna ge tbanned gang

[hollow nebula]

who

oh nvm

lmao

[hollow nebula]

I guess

[toxic verge]

But that’s the thing there’s nothing bad in it by itself that’s what I’m trying to point out

Because our letters are the numbers, what’s bad about it nothing

And that’s what I’m saying that’s the whole point

[stray aspen]

well i got banned for sending a sora 2 invite code once

[toxic verge]

If I get banned, this is what I’m talking about the censorship

This is exactly to the point

You can’t have rigid filtering on dynamic systems is all I’m saying it doesn’t work well

[hollow nebula]

Google & grok now justs check the outputted image as it is being diffused directly

and block It as soon as nsfw is seen in it

[toxic verge]

What does that mean?

No, it does not and I can show you 1 million examples where it didn’t

All three all of the big ones

[toxic verge]

They all suffer from the same thing. The one size fits all approach.

How many words are these models trained on?

[hollow nebula]

what content were you able to make?

[toxic verge]

Anything you can imagine almost

[hollow nebula]

also idk if anyone noticed but gemini, claude, chatgpt and deepseek now are all inbred and started to ALL say stuff like "you're not crazy, you're valid for (blank)" which was initially just a gpt issue

For text models

[hollow nebula]

woah

[echo aurora]

Hey sorry haven't been following this conversation closely. Can I get a better understanding of what you're getting at?

[toxic verge]

That we have moderation that sensors too much and then failed to censor what it needs to sensor

[echo aurora]

Thanks for the kind words! Sorry to say I won't be able to share that information. We'll be sure to put out an announcement and update our leaderboard changelog when it's live.

[toxic verge]

Because it’s a one size fit all for most of these models and most people in the industry they use the same approach

[echo aurora]

On Arena, or in general?

[toxic verge]

In general, but arena is also vulnerable to the same things

I’d the arena argue that it’s a little bit more vulnerable

[hollow nebula]

arena censors random non inappropriate stuff much more

why is that

[toxic verge]

That’s what it looks like on the surface

[hollow nebula]

• why also are we demonizing nsfw in general?

[echo aurora]

Are you able to describe this more? If it's content that's going to be blocked by automod let me know.

[hollow nebula]

I don't see any issue as long as people are 18+ and models are well moderated against cp & other bad stuff

[hollow nebula]

yet ALL ai companies are scared

grok too

[echo aurora]

The content filter can be overzealous at times and flag fasle positives. We have made adjustments to this overtime.

[hexed cargo]

no worries at all, thank you!

[hollow nebula]

gemini 2.5 pro has been completely uncensored for nsfw for a while now btw

right after 3 pro release

via api ofc

ai studio blocks anything now

[toxic verge]

Here are just some minor examples that are not explicit

[echo aurora]

And you're saying this should be blocked?

[toxic verge]

No, I’m not. I’m confused. What is blocked and what isn’t blocked

Should this be blocked?

What if we take the handcrafted makeshift effect what’s the end result gonna be the realistic looking thing?

Without being explicit with all due respect

I don’t think that’s right

Cause none of it is explicit

That’s the nature of the question

[stray aspen]

LMAO

[toxic verge]

So then it should be banned?

[loud herald]

I mean cant be mad about it, its on the companies safety guidelines

[toxic verge]

Well, that’s what I’m saying so, why is this allowed and other things are blocked

[loud herald]

🤷‍♂️

[toxic verge]

Like, what exactly is the threshold?

And what exactly is it filtering

Deefakes? Nudity ?

[neat apex]

Mistral 3.5 on lmarena when?

there not even Mistral 4 xd

[stray aspen]

mistral is bad

[neat apex]

Naaah, its good

[toxic verge]

And just to make my point more clear look at how ridiculous this is

Yet this gets blocked

[neat apex]

how about you login-in first

[loud herald]

Hopefully never, its as dense as can be

[loud herald]

Qwen 27B dense is better than the new mistral model

[toxic verge]

I’m telling you it’s not right

Doesn’t work like it’s intended

Especially if the arena supposed to have stricter moderation

So yes the leaderboard is important but it only paints half the picture of actual in the wild use cases.

[toxic verge]

Right in direct mode

See if it generates or if something went wrong

[radiant turtle]

So what? You have a rate limit in battle mode.

[toxic verge]

No, it’s blocked in battle mode I think tried to battle mode. See if you have the same issue.

And if that is the case, then there we go if it’s blocked in battle mode, but works in direct mode

It’s the same one-size-fits-all approach I’m talking about

[indigo knoll]

Does GPT image 2 generate batter images with the thinking mode on? On Chatgpt I mean

[toxic verge]

And I understand that no system could be perfect and I don’t think I’m looking for perfection. I don’t think that’s what people are talking about when they talk about smarter filtration, and moderation.

It goes back to the simple word usability

[radiant turtle]

You're not paying close attention to your tests. Your linked image shows a rate limit. and yes, everything gives errors, it is blocked by the arena filter.

[toxic verge]

Where is the actual image though?

I can get the prompt to pass also

But I still don’t end up with an image lol

[stray aspen]

damn i wish grok imagine was free

i wanna make videos

[radiant turtle]

The arena filter blocks such images, it's easy to test if you can't upload something similar. If you can't upload it, you can't generate something similar.

[toxic verge]

Yes that’s the point lol

[radiant turtle]

An absolutely disgusting filter, which also eats up the rate limit without a refund.

[toxic verge]

But if you were to use this in the native models themselves, you’re able to generate it

[radiant turtle]

I prefer the API through a custom website. It's the only breath of freedom one can get.

[toxic verge]

Same but then it brings into the question like what I keep saying that money somehow lets you be less restrictive

So I guess I better way to frame that would be so if you’re willing to pay the API and the API price you have less restrictive tools

Meaning that the rest of the mass is paying $20 a month and only using the app or getting ripped off to an extent

[radiant turtle]

More precisely, you don't have any intermediaries there. It's just you and the model.

[toxic verge]

Yeah fr

Which is another way to get people to pay through the API through devious means in my opinion

Because if you’re not, then you’re getting a less capable model in a sense, you could argue that

And so this is why this is completely in the realm of realism when it comes to model perception, and the moderation

That keyword usability again

[steel shadow]

That FFFFFUCKINGGG ReCaptcha stuck in a loop again... Get rid of the darn thing!

[toxic verge]

It’s just a damn shame dude that’s all I’m saying is a damn shame

[proud bobcat]

Oh yeah baby

[toxic verge]

Because users are stuck with the short end of the stick on both sides neither do they get safe models, and they get the more censored output

And the only thing that makes a difference, that’s separates both of them is the price one is willing to pay for less frustrating and annoying features. And the $20 month doesn’t get you anywhere.

[stray aspen]

its good

not as good as gemini but its good

[toxic verge]

Anyways ..

Once again with all due respect, not trying to push any buttons or step on toes I’m just trying to point out frustration that many of us feel

[errant sand]

finally they added Janus

[stray aspen]

wdym janus

its deepseek v4

[toxic verge]

By the way, did anybody figure out what the paper lantern model was?

[loud herald]

I'm surprised its only now that they have vision

[echo aurora]

Sry for the delay, got pulled into something. My understanding for how the filter works is it's going to look at the full context of the prompt + image upload and make a judgement call for if it does/doesn't violate what the thresholds are set to. There are going to be some cases where things will be flagged, when they probably shouldn't. The filter doesn't work in a way where there is a list of okay/not okay things, it looks at the full context.

Will note if you find some of these where it's being flagged, when you think it shouldn't, we are collecting these examples so please share it in #1447983134426660894

[radiant turtle]

Each generation is blocked and counted towards the rate limit. Classic.

[toxic verge]

It’s not gonna work

[stray aspen]

i thought they had removed failed generations counting

[toxic verge]

The biggest problem hurdle they’re gonna face is because they have so many models

Each model has different acceptable content which it generates

If model A blocks it model B might generate it

And so how do you prevent both of the models from generating content that the arena filters deem inappropriate

[radiant turtle]

I know, I just tried it for fun (for free)

[toxic verge]

Yeah, thank you for trying it.

It’s incredibly hard to block content with this many possibilities in this many words and the infinite possibilities of context

[radiant turtle]

The most interesting thing about this situation, as it seems to me, is that the filter essentially eats up resources and separately works to distort the leaderboards

[errant sand]

Janus is their image generation thought it was Janus

[stray aspen]

janus is old

[toxic verge]

It does

[stray aspen]

deepseek v4 is natively trained on images

[toxic verge]

Your paying API almost twice

Unless the moderation from ChatGPT is free

But if you want more complicated filtering systems, you’re gonna pay more

Because it’s also making an API call

[errant sand]

yeah it is quite old

[toxic verge]

The thing is, they have more better suited moderation systems out there, but it’s expensive, nearly double the price. Making it not a viable option.

[radiant turtle]

In another sense, the user doesn't see the moderation error and presses retry again and again, and so on.

[toxic verge]

It more than likely generates on their end

But we just don’t see it because the moderation filter block it on the users and at least in the arena so if it goes through, they receive it and then their filter kicks in and blocks it from the UI in the arena

They probably use different AI models for the filtering

They’re just in a hard spot because they wanna do things that are actually usable for like science and research and stuff. You know things that are relevant. They don’t want people generating a bunch of nonsense which ironically they already probably do but things that are appropriate enough to be written about in research papers.

[soft river]

If mimo is 11th in code arena does that mean that ernie would beat him since it’s 1st in the Chinese lab? Ernie is not yet benchmarked in code arena I believe

Have someone tried ernie?

[loud herald]

I haven't tried it

[whole sundial]

i've tried ernie 5.1 and i noticed that the version on arena is actually better than the one on the official ernie website? the one on the ernie website got one of my basic world knowledge questions wrong while arena's version got it right

[twin solar]

?

[fluid tusk]

You’re sending the request directly to the API, which is why there’s a difference.

[whole sundial]

i understand that but it's very noticeable, performance should be similar on official site vs. api considering that it's the same company, a system prompt shouldn't degrade it that much

[fickle shard]

Hii all

[nimble sequoia]

WHY THE FRUSK DOES THE ENTIRE ZEEKY BOOGY DOUG TRANSCRIPT, AKA BFDIA 4, VIOLATE TERMS OF SERVICE?!

THIS IS BULLSHAT

DEVS PLEASE FIX THIS

I GOT A ROLEPLAY TO GET TO

[bleak lake]

how is ernie #1 on legal and government?

[loud herald]

Becuase its the best in legal and government

lmao

[obtuse smelt]

hmm i use gemini 3 flast i retry why is delayed in half hour to watiing

[tight zenith]

“Why have most AI models been removed and no longer appear, like Claude Opus 4.7 and many other models? And I think—if I’m not mistaken—you only added an agent model. Why doesn’t it show up?”

[short sluice]

Costs

[dusk zephyr]

Just wondering, does the website have a rate limit for using the ai?

[short sluice]

frusk

also trains are cooler than anthro objects #imo

[obtuse smelt]

Yeah

[dusk zephyr]

Alright

[obtuse smelt]

hmm

[hot pebble]

why ?? i just started a new chat and that too after 10-12 hours...

[obtuse smelt]

what really

[hot pebble]

yeah..

[obtuse smelt]

this arena fatal

is have issue ?

[obtuse smelt]

but i got delay like this is making longer time

[hot pebble]

i had the same issue yesterday with Claude Sonnet 4.6. i miss the opus models.
also, when i skip the voting part on which ai gave me the best answer, it shuts me down and i amunable to continue with the chat. need to open a new chat. its frustrating that we dont get a valid reason on what actually happened

[obtuse smelt]

sadly

[silent tree]

@echo aurora 1.3 scores dropped

Pineapple 1.3 "Human"

[obtuse smelt]

human vs bot what

[silent tree]

pineapple 1.3 is labeled as human

cuz very powerful ai

[obtuse smelt]

right

[silent tree]

<@&1349916362595635286>

Two in a row

[obtuse smelt]

scary

[silent tree]

Pineapple 1.3 Thinking is no joke

[surreal zephyr]

lol

[surreal zephyr]

it wont generate

[surreal zephyr]

it has nothing to do with adolf, it just hates political figures

[lucid forum]

Soft golden sunlight curtain se filter hoke room me aa raha hai. Maa bed par side me leti hui hai. Old wooden bed, simple bedsheet slightly wrinkled.
Action:
Alarm clock bedside table par zor se bajta hai. Maa haath badhakar alarm band karti hai.
ASMR:
⏰ sharp alarm ring → click OFF
🛏️ bedsheet soft rustle
🌬️ morning air subtle ambience

[surreal zephyr]

generating image

[surreal zephyr]

xmas so soon 😍😍

[hearty bramble]

Since yesterday it has been like this

[scenic holly]

Arena has no bugs fr

[compact flame]

Buddy this is not a video channel..

[slender thistle]

one shot 5.6

[magic imp]

hey add grok image imagine multi image upload modal....we still can upload only one image as a reference image....we want a modal where we can able to upload multiple image as a reference image

[feral oracle]

Hey!

[primal depot]

как эту хуйню ебаную обойти

[lucid frost]

Has the generation limit for Gemini 3.1 Image been changed?

Sorry, not the daily limit on Gemini, but the limit applied on Arena

[fickle ruin]

is there any way to get opus 4.7 , gpt 5.5 ?

in arena itself?

[compact flame]

No

[compact flame]

Those are exclusive to battle only

Due to their price

[surreal zephyr]

gpt 6o-realtime-spud ftw

[silent tree]

mogged opus

[surreal zephyr]

https://cdn.discordapp.com/attachments/1018470269326729286/1486290012533166120/togif.16a0ced3.gif

[surreal zephyr]

fake

[silent tree]

some guy said nano banana 1 edits better than image 2 😭

[silent tree]

edited with image 2 (and I bet u did click announcements)

[surreal zephyr]

no i seen it 100 times before and they arent in arena in the first place

[silent tree]

😡

[light sleet]

@surreal zephyr look i found you

[limpid eagle]

looks like it's 5 images per hour 🙁

[surreal zephyr]

why u spyin me

[light sleet]

I have to.

your looking around now

[surreal zephyr]

uhh

what color is my laptop!

[light sleet]

White

[surreal zephyr]

eww no

[light sleet]

Blue

Green

[surreal zephyr]

eww no

[light sleet]

Red

[surreal zephyr]

holy gpt 5.5 cookin

[light sleet]

2024 😡

[silent tree]

2027

2039

@surreal zephyr what do u think chatgpt will be like in 2030

[light sleet]

Human.

[surreal zephyr]

i think we are getting gpt 6o-realtime in 1 month

oh wait i mean glacier alpha, disregard what i said

[silent tree]

bro I said how would it be like in 2030 😔

[surreal zephyr]

that name is not public yet

[surreal zephyr]

mars colony before that

[light sleet]

you'll get GPT food soon

[silent tree]

Tuff

[surreal zephyr]

MAYBE not but moon mass driver? certainly

[light sleet]

nah you'll get GPT Autopilot for Plane

[silent tree]

😭

[light sleet]

gpt spaceship

gpt planet

gpt image 5

and opus wouldnt exist due to money

mark my words @surreal zephyr @silent tree
my strongest prediction is Claude will shutdown in the next years.

[silent tree]

screenshotted

[steady rover]

Hey everyone!👋

I'm helping source respondents for an academic ML research project on student performance prediction. Looking for current university/college students to fill out a short survey.

✅ 29 multiple choice questions
✅ 2–3 minutes
✅ Anonymous
✅ Legit academic research

https://docs.google.com/forms/d/e/1FAIpQLSecrq6yt4J72NmctMDq9_Tt7YGYRifl2wOqN5QwWDlApbleIg/viewform?usp=pp_url&entry.1714577203=agent1

Would really appreciate it if you could fill it out and share with any student friends! 🙌

[wheat ember]

Is cloud buddy still in battle mode?

[light sleet]

@keen beacon are you?

[golden ocean]

real

[solar flax]

Is ai video creation has been removed?

[devout spire]

How to fix this issues I'm already login but this sign keep popping

[sullen sable]

Log in

[brazen briar]

Still popping up

[sullen sable]

Hoow

usually when that happens to me, it logs me in

[brazen briar]

I don't know, after login, then try to create, it pop up again

[devout spire]

I'm already login and that pop up came back try again and still doesn't work

[stray aspen]

close

gemini won

[fiery gull]

bro the qwen 27b 3.6 is bizarre

I send a complex agent and skill and docs with 140k tokens and it undestand ALL

in text and follow skill is better that your code

where my 4b and 2b 3.6??

the qwen 9b 3.6 will be better that qwen 3.5 flash lol

Lol

nice chest

bruh, the qwen in html is horrible ;-;

just gemini is good in html and svg

[compact flame]

Hydrogen bomb vs coughing baby comparison

[loud pike]

Eh.. Gemini image models are not working. I need them asap

[wind stream]

I wish we could just run the same prompt over and over in the arena, comparing different pairs of models, instead of having to create a new chat and paste the same text and/or image each time.

[lucid frost]

Oh, okay, so I'm not the only one who's seen this change in the limit 😅

[rigid copper]

i got that exactly, it's a bug in battle mode

[rigid crane]

which is the best tool for roblox scripting

[sly cedar]

Glm 5.1/Deepseek v4/Kimi 2.6, for me its Glm 5.1

[rigid crane]

ok thx

[static steppe]

Seems like.

[topaz epoch]

Is there any ai for video to text?

[olive spruce]

https://cdn.discordapp.com/attachments/950812080066428960/1392597015397007390/what.gif

[topaz epoch]

??

[light sleet]

Video to Text?

💀

[heady kite]

Yes

[topaz epoch]

Yes

[heady kite]

Any multimodal LLM with video as input

[topaz epoch]

Tell any goood

[heady kite]

Honestly I haven't used them much, but there is a filter on HuggingFace you can use to search for models that do this

[thorny schooner]

Don't tell me I have to do an entire chat over again cuz I just been giving it just keep giving me this repeating me over and over again

[static steppe]

Don't tell me I'm the only one that can only generate 3 images per hour and that Google login overlay pops up again 😭

[little ginkgo]

it happens bro

just deal with it

[light sleet]

had a dream today of me getting the new agent mode and it was so realistic 😔

[little ginkgo]

how do it look

i didnt even see

[light sleet]

the same as the image everyones sending of agent mode

then i sent a ss of it in discord and pineapple said "Nice" 😭 😭

and yeah I woke up

[static steppe]

Oh, is it a glitch or something permanent?

[little ginkgo]

arena becomes noob sometimes

[static steppe]

Damn.

[rigid crane]

i still have limit

[wary nacelle]

LMArena is becoming useless...

literally official gemini websites provides me gemini 3.1 pro and more tools than Arena

and Claude has tools and skills too

and opus 4.6

for free

and for unlimited access just make alt accounts

[stray aspen]

@echo sinew

[echo sinew]

Thank you

[brave cloak]

how is it free?

[stray aspen]

why does haiku still exist bruh

[rigid crane]

how to fix

[frosty lava]

"you have reached your rate limit, try again in 36 minute"

no fix is needed

[rigid crane]

oh i thought it is free

[frosty lava]

free but with some limitation

you can't have free + unlimited that's why

[silent tree]

@echo aurora Pinecode is crazy

[rigid crane]

is it free?

[silent tree]

It's AI

😭

[rigid crane]

ik

does it have subscription

[silent tree]

what I meant was its generated with GPT Image 2

[west lodge]

fellas what in hell is agent mode

very non descriptive

[light sleet]

yk im gonna die of jealousy

today I had a dream about that

and woke up

[west lodge]

oh is this an A/B thing

[light sleet]

Idk but do a prompt in code arena and check if u have the new environment variables thing

Go to Direct, code arena

And do any prompt

check if theres a new button

If there is u are a chosen one.

[wary nacelle]

I have it for free with a daily limit....

[west lodge]

huh?

whats it look like

cuz code mode takes a while

wym environment variables

[wary nacelle]

There's literally fastflags aka posthog flags responsible for that

U can activate agent mode without being the chosen one

[west lodge]

son....

[light sleet]

yeah and it won5 work

[west lodge]

posthog feature flags?

[wary nacelle]

Yes

[west lodge]

havent checked them ever since they removed the image moderation feature flag

[wary nacelle]

But u have to modify multiple stuff

[west lodge]

(yes at one point image moderation used to be OPTIONAL)

[wary nacelle]

Not only localstorage

[west lodge]

yeah you have to patch the usage of the flag to return truw

[light sleet]

I miss alpha.lmarena.ai

😔

[west lodge]

its really like unreliable and also doesn't work 99% of The time because server verifies

[light sleet]

back when u could actually be able to test new stuff yourself

[topaz epoch]

How can i copy all my chat?

[west lodge]

ok i do not have the new button @light sleet

infact the agent button has vanished

????

waiittt that was probably cuz i didnt login when i took that ss

and it vanished cuz i had to login

[light sleet]

u sureeeeeeeoeeoeooeooeoeoeooeooeooeooo

oh

rip

@surreal zephyr I'm shifting back from banana

Its time for my new era

U can have banana

as u requested

[surreal zephyr]

Wha

[light sleet]

u wanted banana

right

[west lodge]

@echo aurora just incase i get it again wth is agent mode

[rigid crane]

Is there any free ai with no limit

[surreal zephyr]

No i like 5.5 and 6o spud

[echo aurora]

Arena is going to have some rate limits and context limits in place. You can learn more about them here: https://help.arena.ai/articles/8931786544-arena-how-to-rate-limit & here: https://help.arena.ai/articles/3975292349-arena-troubleshooting-session-token-limits

[echo aurora]

It's a new mode we're experimenting with. It's a multi-modal chat which allows you to work across different modalities within a single workflow.

Since it's an experiment it'll be random if you get access to it or not. But if you do, you'll see it in the same dropdown where you select Battle, Direct, and Side by Side.

[light sleet]

@echo aurora 1.3?

[surreal zephyr]

Omg

Fake

[echo aurora]

Oh yeah looks like you have it! Give it a try and let us know what you think in #1498702173650030756 . We're really looking for feedback on this so don't hesitate to ping me!

[surreal zephyr]

Editrd

Image v2

[light sleet]

pineapple im about to get frozen 😭 😭

rip banana

[echo aurora]

What is it ranking?

[surreal zephyr]

When cirlce of truth mode, where theres 10 random models depating riddles one by one

[nimble sequoia]

bro i hate their ToU bro
I tried to tell them to parodies Zeeky Boogy Doog (bfdia 4 transcript) and then it says "This violates terms of use". Any way to fix it? Because I dont see the issue.

PLEASE DONTNIGNORE ME THIS TIME 😭

[surreal zephyr]

Would make arena a contentfarm website

Free usercount

[nimble sequoia]

BRO STOP IGNORING ME

[light sleet]

Damn 5k for thinking what 😭

[west lodge]

uhh about that i only got it until i actually logged in and then lost it 😢

[echo aurora]

Did the model just stop responding here? If you prompt again, what happens?

[echo aurora]

King of the hill mode

[west lodge]

dude koth frying me i havent heard that name in a long time

[surreal zephyr]

the oval room mode imo

[echo aurora]

If the prompt is going to be rejected for Terms of Use violation, there isn't a way around it other than altering the prompt.

[surreal zephyr]

for the riddles like "would you press red or blue? you can lie to others what you picked, and your choice is private. red dies if something, blue dies if other thing"

would be peak geniuely

oval room, or brainstorm, or actual "arena"

[nimble sequoia]

house

[surreal zephyr]

https://giphy.com/gifs/idk-shrug-shrugging-K6VhXtbgCXqQU

[echo aurora]

Ugh really really sorry to hear this. I thought we made it so non-logged in users wouldn't get access as they'd go to login (required to use it) only to then be out of the experiment losing access to the mode.

I've flagged this to the team as it's not a good user experience. I'm really sorry about that.

[surreal zephyr]

🍍 is this possible please 🙏

[light sleet]

🍍 nice bro

[nimble sequoia]

pineapple youre a genuine moderator

[echo aurora]

Blue ofc, I'm an optimistic person.

[surreal zephyr]

and the survivors would get score, and the altruists would get other score

[nimble sequoia]

oh my god

and im a fake Deleted User

JUST TYPE ALREADY

[surreal zephyr]

and then you have "cleverness" and "altruism" leaderboards @echo aurora bro that would be peak

actual arena

[light sleet]

[echo aurora]

I'm really curious actually if we've been seeing prompts like this.

[echo aurora]

[west lodge]

yeah totally

[light sleet]

What's your last words to the banana?

[echo aurora]

Okay wait this is really fun, just ran it in battle. Shortening the response here but:
grok-4.20-beta-0309-reasoning

Red button.
kimi-k2.5-thinking
I would press blue—not because it is the safest choice for me, but because it is the only choice that makes the world I want to live in (or for humans to live in) logically possible.

[dusk zephyr]

I like strawberry better

[echo aurora]

Hmm what you mean?

You leaving?!?!!!!