general

[robust yoke]

Its knowledge cutoff is pretty early on, like 2024, for instance, so it won't know what model it exactly is, and as such, will hallucinate.

[jade cloak]

yes yes

[robust yoke]

Ah, I see.

[cedar estuary]

So, video arena is working normally? And no, I'm just generating normal video within guideline

[robust yoke]

Yes, it's working normally.

[wooden gust]

I think so.For example, if you go to the original GPT website and ask what your model or version is, what answer do you get, and if you ask from LMarina, what answer do you get? But there is a difference here.

[robust yoke]

Well, you should never be too sure about that.

Maybe there's something in your prompt that falsely triggers that.

[cedar estuary]

Ok, maybe I'll try to leave and rejoin the server to see if works

[robust yoke]

Alright, see ya.

[wooden gust]

You can explain it well, yes, you are right.🥳

[robust yoke]

I am right because I have evidence.

[robust yoke]

And that evidence lies here.

[slim ivy]

Adorable kitty couple rocking the wedding floor – bride in traditional saree with crown, groom in sherwani & turban, full of cute dance vibes 🎶🐾💃🕺”

[robust yoke]

Wrong channel.

Go to #video-arena-1 for that.

[wooden gust]

Bro, you hit the wrong spot.👀

[cedar estuary]

@robust yoke same. What's slow mo enabled?

[robust yoke]

Slow-mo?

You mean Slow Mode?

That's just the delay at which you can send messages.

That way, nobody ends up spamming a chatroom at a rapid pace.

[cedar estuary]

Yes

[wooden gust]

Goodbye everyone 🥳 InshaAllah we will meet again, that means we are talking.

[robust yoke]

Inshallah, brother.

Farewell.

Allah blesses you.

[wooden gust]

Wait a minute, are you Muslim?!

[jade cloak]

lol

[robust yoke]

No, I just know a bit about Muslim culture.

Like how you guys fast for Ramadan.

[jade cloak]

look at this

[robust yoke]

And you can't eat meat because it's haram.

You have to eat halal meat.

[wooden gust]

You're absolutely right.I wanted to explain the difference.The models on LM Arena lie, but if you actually ask a question on the website, the answer is correct!

[wooden gust]

Wow, it's good that you know these things.

[robust yoke]

It certainly is.

I wouldn't want to upset any Muslims after all.

[jade cloak]

where do you live btw

[robust yoke]

Washington.

You?

[jade cloak]

Oh

India

[robust yoke]

Nice.

[jade cloak]

@wooden gust what about you

[robust yoke]

I'm willing to guess Bangladesh.

[jade cloak]

Pakistan, UAE, Saudi Arabia, Bangladesh?

[wooden gust]

Thank you so much for your beautiful mind. Hehe🤗

[robust yoke]

My pleasure.

Your mind is also quite beautiful in itself.

[wooden gust]

Habibi come to Dubai no no no Bangladesh ✅

I am just kidding no seriously please
I am Bangladeshi guy 🤗

[robust yoke]

Figured.

If he speaks Bengali, then surely he's from Bangladesh.

It'd line up after all.

[wooden gust]

Huh🤔 yeah👍

[wooden gust]

Actually, I'm a bit of a fickle person, I talk a little too much.I apologize to everyone for talking so much for so long 🤐 and I'm leaving 👋

[jade cloak]

Are you using translator?

[robust yoke]

I don't think you're necessarily a "fickle" person just because you talk too much.

I also talk quite a bit too.

It's okay to be passionate.

And you don't have to apologize for every single thing that you do, as kind as it is.

[wooden gust]

If you don't lie, I'm using yes.🥹I don't even understand what the translation means.🫤

[jade cloak]

Yea i see, i guessed it from your way of typing

[robust yoke]

I also happen to see that your translator doesn't handle spaces between sentences very well, heh heh.

[wooden gust]

Oh! But I meant excited or emotional.🤗

[robust yoke]

Ah, I see.

Your translator worded it differently.

[wooden gust]

Ah, right Guess✅

[wooden gust]

Thank you again for saying it's beautiful.

[robust yoke]

It's really my pleasure.

You complimented me, so it's only fair that I compliment you back.

[jade cloak]

Yea

[ocean vortex]

😇

[wooden gust]

Hey bro, I think I got caught.🙃

[robust yoke]

I think so too.

[jade cloak]

LMArena should add a new feature, where it shows that what the model is thinking, like the thinking proccess.

[robust yoke]

I'm sure that wouldn't be much of a hassle.

[ocean vortex]

they could but then you would know if it's a reasoning model or not

[wooden gust]

Thank you.

[ocean vortex]

They could only add it to direct chat I suppose

[robust yoke]

All they'd have to do is add a little drop-down that shows what the model is typing in "<thinking>" tags, then just put that text within the drop-down.

[jade cloak]

Thats im saying

[robust yoke]

My pleasure.

You deserve it.

[wooden gust]

Good think

[jade cloak]

Yea..

Have yall made any website guys? If yes can i see them.

[robust yoke]

It'll be hidden in Battle mode for anonymity purposes, but revealed within a Direct chat since the model is chosen.

[robust yoke]

Sure have.

[jade cloak]

Yes

[jade cloak]

Please let me see it

[robust yoke]

Qwen Max (the latest version) made this: #ai-creations message.

It's a recreation of Snake in a 1900s style.

[jade cloak]

Damn nice

I liked the ui also

[robust yoke]

Yeah, it's pretty creative.

It likes to take some creative liberties, unlike other models like GPT-5.

[jade cloak]

Is Qwen Max reasoning or non reasoning

[robust yoke]

Non, unfortunately.

[jade cloak]

Its not reasoning?

[robust yoke]

Yeah.

[jade cloak]

Then its good

[robust yoke]

I suppose so.

[jade cloak]

It can create such good things in less time

[robust yoke]

Yeah.

[jade cloak]

Btw i was also working on a autoclicker

python

[robust yoke]

Then again, thinking models are also good since they can think out a good, accurate solution, even if they take longer.

[robust yoke]

Oh nice.

You could probably use Qwen for that then.

It's very good at coding.

[jade cloak]

Here is it

[left iris]

Does anyone know is there a possibility to fix the typos in AI generated images smoothly ? A specific prompt maybe that it would not mess up again when inserting the same image for it to fix ? I have this cool design for logo but there is one letter missing instead of nocturnal it says noctural

[robust yoke]

Moleh moleh.

That looks impressive.

[jade cloak]

I used Gemini 2.5 Pro in it.

[clever cairn]

Bhai kisko pata hy abhi cinema wala Jo anime character log shooting Jaise live action sach mein ho raha hai aisa kaise karte hain

[robust yoke]

I'd recommend using nano-banana for a quick, easy fix.

It uses the same font while fixing the text.

[jade cloak]

Kya matlab

[robust yoke]

Breh.

[clever cairn]

Apako pata hy

[robust yoke]

Willing to guess Indonesian.

[jade cloak]

Samjha nhi mein

[jade cloak]

He is Indian

[robust yoke]

Figured.

[clever cairn]

Yesa

[jade cloak]

Yeh kya hai?

Thoda explain kro bhai dhang se

[clever cairn]

Aisa video live action character

[jade cloak]

Video ya photo

[jade cloak]

Yea, i myself didnt expected that we can create such good ui/ux designs in Python. Because all python apps i have seen in my life does looks like 1990's.

[robust yoke]

Well, not just that, but also because the UI looks too good for a Gemini creation.

[robust yoke]

Typically, the UIs that Gemini creates aren't really all that good.

[jade cloak]

Fr.

[left iris]

K. Thanks mate ! Will try it later when i get home in the evening from work.

[jade cloak]

He did the same thing. He made a trash ui,

Then i explained him well.

[robust yoke]

Alrighty.

Have fun.

[robust yoke]

Fair.

[jade cloak]

I think giving screenshots to Gemini make its work better.

[robust yoke]

Yeah.

[jade cloak]

90% of the time it makes good ui when we provide him a image.

[robust yoke]

That's true.

[jade cloak]

Did you made more websites?

I saw the Snake one.

Is there any other?

[robust yoke]

There is, actually.

Specifically, one that I made a while ago.

[jade cloak]

Can i see it

[robust yoke]

Here ya go: #ai-creations message.

[jade cloak]

Its made using GPT5-high

ig

[robust yoke]

It's a writing tone changer that changes the tone of your text using Gemini 2.5 Flash Lite, Google's smallest and fastest model.

[clever cairn]

Bhai yesa

[robust yoke]

It was partly made using GPT-5, yes.

The other half was Claude Opus 4.0.

[jade cloak]

White mode looks good.

[robust yoke]

I'm glad.

The drop-down menu is a bit buggy, but overall, I think it did pretty well for a UI.

[clever cairn]

Tnx

[jade cloak]

Kya thanks

[jade cloak]

Yes

Dark mode looks very buggy. White looks perfect,

[robust yoke]

Ah, interesting.

I thought the dark mode looked pretty good, in my opinion.

[jade cloak]

LOL

[robust yoke]

Breh.

[jade cloak]

💀

[clever cairn]

Yahai

[robust yoke]

Breh.

[clever cairn]

Bhai ai toll batao best

[jade cloak]

Bro is legendary

[robust yoke]

Minor coding issue.
The "minor coding issue" in question:

[jade cloak]

Bro is proffesional coder tho

[robust yoke]

True.

[jade cloak]

Btw

Did you used LMArena to make this

[robust yoke]

I sure did.

[jade cloak]

Nice Nice

[robust yoke]

I suppose that's why it's not exactly up to par.

[jade cloak]

This is how it ends.

[robust yoke]

Yeah.

It uses Reinforcement Learning and learns from its mistakes over time.

Hence the counter on the top-right.

[jade cloak]

https://goflowmap.netlify.app/

Check this

my website

[robust yoke]

This looks nice.

[jade cloak]

Yea

its made using Gemini 2.5 Pro also

I use it because it have long context length

[robust yoke]

Fair.

[jade cloak]

Do you use Comet Browser

[robust yoke]

I sure do.

You were likely able to tell because it shows up as my current activity.

[jade cloak]

Is it free?

[jade cloak]

Yea!

[robust yoke]

It requires a waitlist to join, but otherwise yes.

[jade cloak]

Whats special in it

What it does

[robust yoke]

However, I can gift you it.

It has an assistant that can temporarily take control of your browser and get stuff done for you.

[jade cloak]

Hmm.

Alright!

Nice to talk with you.

I got to go.

[robust yoke]

Ah, alrighty.

It was fun talking with you as well.

Farewell, and have a splendid day.

[jade cloak]

..

Ok

[robust yoke]

Hm?

[jade cloak]

Sorry, i mean thanks.

[robust yoke]

Ah, it's my pleasure.

I was simply curious.

[turbid dome]

hello , comparing video generator for my job

[robust yoke]

Well, you'll wanna go to this channel then: #video-arena-1.

[fiery gull]

when the lmarena update the learboards ?

[true compass]

Hi all

[robust yoke]

Pretty soon, I assume.

[fiery gull]

hi

[robust yoke]

Given that they tend to update it on a daily basis.

[robust yoke]

Greetings.

[lusty light]

Hi I'm Mo, I'm here for the videos

[random fjord]

help it is stuck on this and sometimes it says failed

[robust yoke]

Greetings, Mo.

That just means that it's thinking.

[random fjord]

its been like that a long time now

[robust yoke]

Well, it tends to do that for a while.

[random fjord]

btw

[robust yoke]

Hm?

[random fjord]

is staff amas a special role?

bc i dont see anyone else with it

[robust yoke]

Staff AMAs?

[random fjord]

ys

[robust yoke]

You mean the mod role?

[random fjord]

look at my roles

[robust yoke]

Ah, that.

I don't think it's necessarily "special", per se.

[random fjord]

bro its been like that for a while but before when it was typing and before it was done it says something went wrong

[robust yoke]

Maybe the Cloudflare session expired then, but who knows?

[fiery gull]

dawn bro, I'm testing the qwen 3 now, whyyyyyy ;-;

I need a therapyst

[robust yoke]

Er--

[fiery gull]

or just use the lmarena ?

[robust yoke]

How come you need therapy?

[fiery gull]

because I'm testing Ai, and not working

[robust yoke]

Ah.

[fiery gull]

I'm have soo many task

[robust yoke]

Lom's also got the same issue.

Except with GPT-5 High.

[random fjord]

WHAT cant it just stay open?

[robust yoke]

Unfortunately, doesn't seem like it.

[vocal breach]

Hello everyone, just joined

[random fjord]

so i gotta somehow do this done ok

[jade cloak]

No

[robust yoke]

Every two minutes or so, you'll have to refresh your page to re-verify.

[jade cloak]

If its stucked

then its a bug

Just restart your browser

and try agian

[random fjord]

okay

[full badger]

hello

[jade cloak]

If it still didnt worked. Then have to create a new chat.

[jade cloak]

Hi

[robust yoke]

Greetings.

[jade cloak]

Are you programmed to say "hello" only?

Heh

[fiery gull]

hi bot

[fiery gull]

I think some peorple only have urge to say hello, or is bot from discord to start chats in void chats

[robust yoke]

Or perhaps they're actual humans that simply only like to socialize one time.

[fiery gull]

What are your expectations for the new 2.5 flash? I think it will be 8 or 80, or it will surpass all except 2.5 pro or up 10 points

[robust yoke]

I've tested it out, and it's safe to say that it's practically the same as the old one.

[jade cloak]

I would like to have file upload feature in LMArena.

It will be useful

[robust yoke]

Yeah, would be pretty nice.

[fiery gull]

Idk for sure, but I think they didn't put it either because their website wouldn't be able to handle it or because they can't reach an agreement with the AI ​​companies

[analog crypt]

hello

[jade cloak]

Hi

Aren't you "hello" only?

Yes you are.

[robust yoke]

Oh well.

[fiery gull]

just my hypothesis

sources: voices from my head

[jade cloak]

lmao

[scarlet cape]

Hello every body.

I'm new here.

[jade cloak]

Hi

Welcome!

[robust yoke]

Fair enough.

[robust yoke]

Greetings.

Welcome to the server.

[obtuse nest]

Can u create nsfw?

[jade cloak]

@robust yoke Bro your profile is really dark that im unable to see it.

[jade cloak]

Browtf

peace out

[obtuse nest]

Im js askin can u answer?

[jade cloak]

no we cant

[obtuse nest]

Wat if u do

[jade cloak]

we just cant

[obtuse nest]

How

[jade cloak]

give reason

why you want to create

[queen veldt]

You can

But not on lmarena

[jade cloak]

This is LMArena's server

so it should be only LMArena

[queen veldt]

Depends how much nsfw you want it

[knotty fable]

I can only give my own reason for 'NFSW' creation.
I'm doing a scifi story with a planet with a rather hedonistic culture- this does not mean people are humping each other in the streets ofc. 😹
It's rather mild, but some scenes could not be done with the insane level of consoring we've seen.
It's changing though - and expected this to happen. Just in from 'Moovieflow'
These models are designed to support NSFW (Not Safe For Work) content generation, giving creators more freedom to explore mature themes in their projects.

This while the two AI's I use most got rather mild censoring in one case, and barely none in the other clip making AI.

[wide briar]

Hello there. Sorry for interrupting a bit. :]

[knotty fable]

No worries.

[robust yoke]

It's not dark, it's transparent.

[wide briar]

Oh yeah, just posted a problem at the bugs forum. I'll be glad if you check it out.

[knotty fable]

Seen it, no clue.

[robust yoke]

There we go, answered it.

[remote arrow]

Hmm.. World of problems..

[amber verge]

hellooo

[robust yoke]

Greetings.

[amber verge]

do you know the difference between all the ai models lsited on the website?

[robust yoke]

The difference between them was that they were all made by different companies.

😃

[amber verge]

in features

like gpt 5 and 4o and all of them

[robust yoke]

Well, both support image uploads.

[shut jay]

／

[robust yoke]

https://tenor.com/view/the-simpsons-lenny-leonard-listening-interested-attentive-gif-4574713

[oak cosmos]

Hello

[fiery gull]

hi

[fiery gull]

...

You want to know more about AI?

[tall tulip]

How much image can generate in one hour?

[fiery gull]

5 per day only

Idk if the lmarena improve the limits

[tall tulip]

U mean per hour?

[fiery gull]

no, day

[robust yoke]

Greetings.

[fiery gull]

the last time I look this its just 5 per day

[tall tulip]

Wdym, i use it like hour ago

[fiery gull]

hmmm

You tried to create a another account?

[tall tulip]

Nah

Too lazy

[fiery gull]

use a another google account

ah

[tall tulip]

Im using ai because I'm a lazy ass person

Making another account is too much effort

[fiery gull]

too, but I use a various acconts 🥵

[tall tulip]

Ah

So we can refresh using new account

[fiery gull]

yep

[tall tulip]

Got it, but still too lazy

Thanks tho

[fiery gull]

ok

[cunning hawk]

this in lmarena?

[wide briar]

Hmm, wait, wat? Is this AI Generated or..., what?

[cunning hawk]

yeah i have that bug too

[robust yoke]

It sure is.

[cunning hawk]

bottom right?

pls be observative

[wide briar]

Well. I'll be damned. That looks neat tho.

[wide briar]

Ah ye. My bad. Sorry. :']

[tall tulip]

Is that a bug?

[cunning hawk]

yes

it will never let you generate again

[robust yoke]

This is Hunyuan 3.0, Tencent's latest image generator that even beats nano-banana, I believe.

[tall tulip]

Dang

[cunning hawk]

youll account is like corrupted

[tall tulip]

Really?

[robust yoke]

Even GPT-Image and Seedream 4.

[robust yoke]

Sure is.

[cunning hawk]

doubt it

its very obvious its ai image

[tall tulip]

Send me some image

[robust yoke]

The scores speak for themselves.

[tall tulip]

Of qiuyan or whatever that is

[cunning hawk]

china number one

[robust yoke]

https://cdn.discordapp.com/attachments/1421712716866912256/1421712717223432313/gsb.png?ex=68da08a5&is=68d8b725&hm=4d71eac65bfce1160d0e614ebc860f23674cd1c25b1ac745cabfe55473538e01&

[frozen osprey]

The image certainly doesn't

[tall tulip]

Example of the image broder

Im dumb with statistics

[robust yoke]

The image may not, but this leaderboard does.

[frozen osprey]

Ah another benchmaxxed leaderboard with hyped latest ai model

[robust yoke]

True.

Just like every other image generator nowadays.

[cunning hawk]

why do you place dots

at the end of ur mesage

[frozen osprey]

At the end of the day, it's you who decide Whats better for your use case. Numbers are just numbers

[robust yoke]

Well, it's normal, isn't it?

[cunning hawk]

no its not

discord is not that serious

[wide briar]

Hmm, so like, i can image an realistic Modern Talking dudes photoshoot with this model then? If yes, that's interesting.

[robust yoke]

And neither is typing.

Or having that argument.

[frozen osprey]

Bro is mad at someone making a semantically correct sentence 😂

[fiery gull]

hhmmm, really? I'm skeptical about these results

[robust yoke]

Well duh.

[frozen osprey]

You know you've spent too much time online if you do this

[robust yoke]

https://tenor.com/view/ishowspeed-speed-trying-to-not-laugh-speed-holding-laugh-speed-laughing-speed-laugh-gif-242565803459391828

[fiery gull]

ok -_-

[fallen adder]

s s sorry, but . I . I I

I want to be your light. @robust yoke

🙃

[robust yoke]

https://tenor.com/view/joe-biden-presidential-debate-huh-confused-gif-9508832355999336631

Yeah, you don't say.

[wide briar]

Heheh...

[quasi atlas]

Hey! Just a quick reminder: please keep conversations aligned with the purpose of this server. Topics not related to it should be avoided. Thanks for understanding!

[robust yoke]

Yeah, Zoro.

Gotta keep this chat filtered. :troll:

[quasi atlas]

Thanks!

[robust yoke]

Our pleasure.

[cunning hawk]

okay sorry

[robust yoke]

https://tenor.com/view/spongebob-spongebob-squarepants-gif-26050971

[half mist]

Hey, I was wondering if there’s a limit to how many times a video can be generated? I saw a YouTube video saying it’s free forever, which is pretty amazing! I’m just curious if that’s true.

[robust yoke]

Not true, unfortunately.

There is a limit to how many videos a day you can generate.

Apparently, if memory serves me right, it's either 5 or 8.

[half mist]

Dang, I am just trying to find a good video generator for free

[fallen adder]

By the way, my apologies.
Previously, I didn't really understand much about AI and its various models.
As far as I know, this site serves as a platform to test different models and understand their limitations? (Correct me if I'm wrong). And it's free of charge?
Once again, I'm sorry if this topic is inappropriate.. I'm just curious how this site is able to sustain itself.
I looked around for a donation button or something similar, but couldn't find one.
Is it not necessary?

[robust yoke]

Yeah, I get that.

Maybe soon.

[trail eagle]

Hi

[half mist]

Hi

[robust yoke]

Actually, there is a donation button.

Notice the little hand with the jar next to the Send button?

On the LM Arena chat input box?

[robust yoke]

Also, it's completely fine. No need to apologize.

[still marten]

hi

[robust yoke]

And yes, that is spot on.

[robust yoke]

Greetings.

[tidal arrow]

How i can download veo3?

[half mist]

Hey, is there a way to create videos using a specific model instead of just trying the arena mode?

[robust yoke]

Short answer: you can't.

[robust yoke]

Unfortunately, no.

[tidal arrow]

How i can use it?

[half mist]

Long answer: yyyyyoooooouuuu cccccccaaaaaannnnn’’’’’ttttt

[robust yoke]

It's up to random chance.

[robust yoke]

Via Google AI Studio.

[half mist]

Not for free though, it requires Google AI Ultra

[robust yoke]

That's true.

[half mist]

It’s about $250 dollars per month

[robust yoke]

Or, you could use it via the “Build” section.

[half mist]

Still requires Google AI Ultra. Just tried it a few days ago

[robust yoke]

Since the APIs are hosted locally on the server and don't require API keys.

[robust yoke]

Hm.

Could try on Flow, then.

I'm not an Ultra subscriber and was seemingly able to use Veo 3 on there.

“Seemingly” because I never actually tested.

Simply saw that I could.

[half mist]

Nvm, it actually requires an API key not in the free tier

[robust yoke]

Ah.

[half mist]

The build mode used to work, but Google decided to move it to a paid tier

[robust yoke]

Well, darn…

[languid crest]

I was just curious on how lmarena affords to run all these models and that too for free? I mean yea our user feedback data is valuable but does that cover all of their costs?

[half mist]

You can however use Veo 2 using Google AI Studio although, it’s limited

[robust yoke]

They probably get their revenue somehow.

Though, not even we're sure.

[robust yoke]

That's true.

Not as good, but not bad either.

[half mist]

You should probably get the free trial for Google AI pro to get Veo 3 because you get around 13 daily video generations for free and around 50 for using credits

That’s what I did

[languid crest]

do they mostly host the models (ones that are openly released) themselves or just use apis?

[robust yoke]

Usually, they use the APIs of open-source models.

However, there does seem to be a way for them to use closed-source models too.

[half mist]

Do you guys have the Grok Imagine API?

[languid crest]

umm, how so?

[robust yoke]

For instance, when nano-banana was still undercover and not publicly available.

[languid crest]

hmm interesting

[robust yoke]

Google gave them access to the API for it.

And since it's only $0.03 per generation, it's dirt cheap.

[half mist]

How much does Veo 3 generations cost?

[robust yoke]

Hmm…

Not sure.

[half mist]

For the veo 3 fast one not the quality one. The veo 3 quality is usually really expensive

[robust yoke]

Apparently, $0.75.

[half mist]

That’s not that much at all, but will add up the more people use it

[robust yoke]

A 30-second video would cost around $22.50.

[half mist]

What? You can do 30 second videos? How? I couldn’t get that to work

[robust yoke]

Via the API, I believe.

[half mist]

There was a server I was in that supported Veo 3 with no limits, but stopped because it didn’t have enough money to keep going

[robust yoke]

The cost of Veo 3 video generation depends on the access method:

• When using Google’s Gemini API or Vertex AI directly, Veo 3 is priced at approximately $0.75 per second of generated video and audio output. For example, an 8-second video would cost $6.00, and a 30-second video would cost $22.50 at this rate.

• For subscription plans via consumer products like Google AI Ultra, the Ultra plan costs $249.99/month and offers 12,500 credits, which allows for about 125 high-quality Veo 3 videos (each approximately 100 credits). The Pro plan costs $19.99/month and includes 1,000 credits, allowing for about 10 high-quality videos per month.

• On third-party platforms, Veo 3 videos in Fast Mode cost about 20 credits each, while high-quality Veo 3 videos cost about 100 credits each.

In summary, the pay-as-you-go API rate is about $0.75 per second of video, while subscription plans offer monthly credits that translate to multiple videos depending on quality and length, starting from about $20/month for limited use up to $250/month for more extensive usage.

[robust yoke]

Ah, dang.

If only I was in that server while that was happening.

[half mist]

Who is /

[robust yoke]

Me.

The slash-looking thing is just an italicized “I”.

[half mist]

Oh, well here is what it said exactly: "Due to the high costs I have made the decision to only host ALBERT, the other AI's including video gen were racking up big bills and I ain't for that lol." Albert is just an LLM AI

[robust yoke]

I see.

[half mist]

I would do the same if I was paying

[robust yoke]

Fair enough, heh heh.

I wouldn't wanna be paying for millions of users just so they can use a video generation model.

Seems like a waste of money to me.

Could be put to better use.

[half mist]

The person in the server even said that it was trying to find a free Veo 3, but I said that doesn’t exist at all legally

[robust yoke]

For obvious reasons.

After all, I'm not very sure any celebrities would want deepfakes of them doing heinous things from millions of random people.

[half mist]

Wait, I wonder if you can get the Apple Intelligence Cloud model onto LMArena

[half mist]

Yeah

[robust yoke]

Well, considering what these guys are able to do, it's likely possible.

[half mist]

The only way to access it is through the Shortcuts app, and other than that, there is no other way to access it*

[robust yoke]

Hmm…

I'm sure they'll find a way.

[half mist]

From what you can do with Shortcuts, it wouldn’t be too hard to make

But, the Apple Intelligence Cloud model always seems to tap out when mentioning copyrighted characters or NSFW content for obvious reasons

[half mist]

Also a bonus, you don’t have to pay anything for this model since it’s free

[robust yoke]

That's true.

[half mist]

The only thing the Cloud model doesn’t have is Thinking longer which is a crucial thing for an AI model

[robust yoke]

Indeed.

[half mist]

What I like about the Cloud model is you can change how the output is presented as.

[mighty bison]

Hello

[half mist]

Ello

Also, holy god these are terrifying

@robust yoke Why does it always ask me to verify I am a human? I understand one time, but it’s everytime I reveal which models are which

[flint sandal]

What is better for coding? GPT-5 Pro or GPT-5 Codex. Both from openrouter

[half mist]

Test them on LMArena and find out

[flint sandal]

On lmarena there is no gpt-5 pro

Oh on openrouter too

I think that gpt-5 codex is better because its literally for coding.

[half mist]

Yeah probably

[peak lagoon]

Why audio sometimes is not generated?

[half mist]

For Veo 3?

[peak lagoon]

Yes

[half mist]

Google says that audio is beta and may not generate

[peak lagoon]

Ok, thanks!!

[acoustic lark]

Hi everyone, can you explain why I can't use this program? I want to log into my Google account, but it shows what's in the photo. Also, when I try to create an image, it shows me a safety warning sign. I click it, but then it gives me an error and nothing works.

[quasi atlas]

@sharp quest you would like to check #1397655624103493813 to learn how to use the bot.

[languid crest]

on the website, is there a way to top the text generation mid way? I can't seem to do so

[half mist]

It wouldn’t be that hard to make either, you just need to make some get and post requests, and you’re good

[graceful marsh]

#hello just checking new updates - thanks

[fiery gull]

don't talk clanker, or U can be banned

[waxen hinge]

hi

[warm kettle]

Checking out the updates from LLM's - all in once place.

[golden ocean]

https://tenor.com/view/get-blown-up-dog-dog-explosion-sticky-notes-gif-17071604041793579031

[lapis imp]

so my chat is stuck on the genrating for the past hour is there any fix ?i was doing research for my school project i cant start over lol

[regal sundial]

What is this codename Rainbow?

[wicked mason]

https://tenor.com/view/remember-the-alamo-jon-mud-niche-tiktok-brainrot-gif-6900259674673832484

[chrome gorge]

https://www.linkedin.com/posts/godhand_ai-promptengineering-futureofwork-activity-7378095823580061696-Ib1V

[golden ocean]

The ai research: https://tenor.com/view/remember-the-alamo-jon-mud-niche-tiktok-brainrot-gif-6900259674673832484

[polar niche]

gpt-5-high is down

[oak sierra]

where is my video

[hushed terrace]

Hi! Please check https://discordapp.com/channels/1340554757349179412/1397655624103493813 to create videos on the video arena channels. You´ll receive a direct message when your video is ready. Please be aware it can take up to 10 minutes.

[tiny lily]

Helooo

[fiery gull]

hello

[remote arrow]

So... #leaderboards is now a new HI and HELLO room and #ai-creations is another prompt dumpster? I don't see an #announcements about that..

[neon idol]

Yuppie my account is back

[remote arrow]

Is that true? What was our last talk in this channel?

[neon idol]

Absloute banana

Memes on banana

And puppie and kittie on food

[ocean vortex]

It is safe with me, I have it.

[remote arrow]

Give it, Dom. He might cries..

[remote arrow]

What do I call Nano Banana and Seedream?

[stoic parcel]

hello

I want to know the platform

[neon idol]

Nani banani and SodaDream

[glossy umbra]

👀🍌

[topaz bay]

They don't have the qwen-image2509?

I've heard it's as good if not better than nano banana, so i think they're using the wrong one on llmarena?

[static prairie]

hello

[robust yoke]

Greetings.

[remote arrow]

Read #1397655624103493813
Put your kid's picture here with that prompt will never gives you any video..

[robust yoke]

Wrong channel.

[robust yoke]

“Comprlains”.

[remote arrow]

Blame Banani

[robust yoke]

Yeah.

[remote arrow]

Soda made it worst

[remote arrow]

Scene 1, take 13
#ai-creations message

[robust yoke]

Aye aye aye…

[neon idol]

Hello man

How is your day going?

[remote arrow]

And, yet, hundreds of videos on Mytube glorifying its "precise" text recognition..

[robust yoke]

Greetings, Greggorious Prime.

[neon idol]

What model dis you use? A chinese one i think

[neon idol]

🤫🧏

[robust yoke]

Doing good today, and you?

[neon idol]

Me too thx

[remote arrow]

That's Hunyuan, I think..

[robust yoke]

My absolutely pleasure.

3.0.

[neon idol]

.

[robust yoke]

Hunyuan 3.0

[neon idol]

Wth?

[robust yoke]

It generates very good images.

[neon idol]

since when has it existed?

[robust yoke]

Since I found out yesterday.

[remote arrow]

Note that it isn't a pure image model, it's a language model with image output, like GPT-4o or gemini-2.5-flash-image-preview ('nano banana').

[neon idol]

Where?

[neon idol]

Bro is not nano banana

Its nani banani

https://tenor.com/tnpM3b5fU9J.gif

[robust yoke]

Skadoosh: https://hunyuan.tencent.com/visual

[remote arrow]

I copied that one from Bluedit

[robust yoke]

Chat, I clearly don't think this guy knows his lingo.

[remote arrow]

Hmm.. What should I call it after Nana-chan and Egg-chan.. 🤔

[robust yoke]

Ah.

I thought it was Nano-Chan.

[neon idol]

Lol

[robust yoke]

Anyway, Hun-Chan.

[remote arrow]

I should re-check for consistency. Lol..

[neon idol]

Hunywan 3.0 is better than soda dream?

I am trying Hunychan now

[robust yoke]

In terms of output quality, I'd say “Yes”.

In terms of character recognition…

no.

[neon idol]

What???

[remote arrow]

For me, can't say yet. Since Qwen just claims to be new banani killer too..

[neon idol]

Thats a joke?

[robust yoke]

Rage baiter detected!!!1!1!11!1 😵😵😵 (Joke, by the way.)

[remote arrow]

Who? Wolverine?

[neon idol]

Screen?

Send screenshot for proofs

[remote arrow]

Downloadable to run locally, will eats you alive because it demands for 320GB of VRAM..

[robust yoke]

The tweet.

[neon idol]

OF THE ANNOUNCEMENT

[robust yoke]

True.

[neon idol]

Bro in my opinion your brain isnt for real braining 🥀

[neon idol]

In my opinion no one image generator shoukd be able to beat sodadream for still 1 month

[proud hazel]

Written 15h ago? Why is nobody talking about this.

[neon idol]

Yeah...

[robust yoke]

Yeah, guys, I have the tweet right here!!!1!1!1!1!1!

[neon idol]

💀

Clickbate announcement

No one announcement for 3.0

[robust yoke]

IT WAS NEVER REAL!!!1!1!11!1!

https://tenor.com/view/venom-gif-27487650

[remote arrow]

What do you expect from 3 bananas in a row anyway?

[robust yoke]

A “wave” of bananas… ba dump tss

[neon idol]

A better model of nano banana?

[remote arrow]

I've been expecting for Gepetto-Image-2 that never comes..

[twin otter]

#generate video of a football player running in the wood wearing red short black tee shirt and golen shoes

[proud hazel]

At least we have this, but still just speculations.

[remote arrow]

Wrong channel #1397655624103493813

[robust yoke]

https://tenor.com/view/dean-cain-running-away-football-gif-15211313

No way, it worked?!?!!11!1!1?!!11!

[neon idol]

🧏🧏🧏

Let him mewing

[robust yoke]

Let's let him mewing, fellas.

[remote arrow]

👏 tadaaaa... Wizard!

[robust yoke]

True.

Generated a video out of thin air.

[robust yoke]

I didn't even think it was possible.

[remote arrow]

LMArena should mirroring grok.com here..

[robust yoke]

Suddenly just remove the system prompts for all of the models and let them just go absolutely haywire.

https://tenor.com/view/me-atrapaste-es-cine-its-cinema-cinema-esto-es-cine-gif-12869046600151364058

[remote arrow]

Dangerous mind detected..

[robust yoke]

You see, Peter… I'm something of a psychopath myself…

[neon idol]

Chat i need a prompt really difficult for text to text model

[remote arrow]

Go to https://grok.com/imagine

[robust yoke]

I got you, my Sigma bro.

[neon idol]

😏

New rizz unlocked

[neon idol]

Why?

[remote arrow]

The most difficult prompting in the 7 realms

[neon idol]

Im not understanding

https://tenor.com/nQcrsOBldoL.gif

[rigid portal]

Hello everyone

[robust yoke]

Greetings.

[remote arrow]

Hi someone

[neon idol]

Pls a prompt 🥀

[robust yoke]

It cooked.

[rigid portal]

Don't call me someone bro

My name is jude

[neon idol]

Bro is being mad

🚨🚨🚨

[rigid portal]

Dang

[remote arrow]

What's your name? Verify you're a human

[rigid portal]

I don't get time for verification bro

[neon idol]

Sodadream

[neon idol]

Sodadream Chan

[formal cosmos]

Hiii, sorry to bother you guys, could someone tell me if there is a limit to how many videos can be generated? ☺️

[rigid portal]

Have you tried it many times but is not working

[robust yoke]

There sure is.

Also, greetings.

[rigid portal]

6

[remote arrow]

5

[rigid portal]

6

[formal cosmos]

oh, thank you, have a great day<3

[robust yoke]

I can assure you that it's indeed 5, by the way, Miss.

[neon idol]

These are 10 questions?

[rigid portal]

Same here brother

[neon idol]

Or only 1?

[robust yoke]

Yes.

[remote arrow]

It was 8

[neon idol]

What are the answers?

[rigid portal]

What are you all talking about 😒

[robust yoke]

“Was”.

[robust yoke]

Lemme see…

[rigid portal]

Lol

[neon idol]

https://tenor.com/bimC7.gif

[rigid portal]

😊😊

[remote arrow]

Any chance to a bargain? Maybe 9?

[rigid portal]

It's 10 brother

[neon idol]

https://tenor.com/qrWbPfExDJY.gif

[rigid portal]

It's enough no more answers

[formal cosmos]

wow, is it 8, 6 or 5? I'm getting confused 😵‍💫

[remote arrow]

Could be 10. Don't worry..

[rigid portal]

Yoo it's 6

[rigid portal]

Anyways

[robust yoke]

It's five.

[remote arrow]

Ah.. You're bargaining!

[rigid portal]

What are you taking about

[robust yoke]

I sure am.

[robust yoke]

I'm bargaining with him.

[rigid portal]

@robust yoke what are you talking about

[robust yoke]

It's very common in many places.

[remote arrow]

Prompt trader

[formal cosmos]

I'm going crazy, there are so many voices XD

[neon idol]

Answers? 🥀

[robust yoke]

Ah, right.

[neon idol]

Thxxxx

[rigid portal]

The answers is 10

[remote arrow]

I muted my mic. Chill..

[robust yoke]

Well, you're gonna wanna get used to that here.

It gets pretty noisy.

[rigid portal]

I have a question

[formal cosmos]

you are so sweet, thank you hahah

[robust yoke]

Yes?

[rigid portal]

How old is everyone

Because all of you're taking like 6 months old baby haha

[remote arrow]

Today or 17 years ago? Or 17 years from now?

[rigid portal]

I don't get you

[echo aurora]

Let’s keep conversation focused on AI please

[rigid portal]

That's your concern

[rigid portal]

Who is ai?

[neon idol]

Greg why you dont have a bit of umorism 🥀

[remote arrow]

My newest AI creation

[robust yoke]

'Cause he's the fun police.

[rigid portal]

Wow
How were you able to do that

[robust yoke]

He's not supposed to.

[neon idol]

💀

[formal cosmos]

Please, someone tell me there are 12, just so I can test something😇

[remote arrow]

Long story and history, dude..

[remote arrow]

5

[rigid portal]

It's easy am just testing your advice lol

[rigid portal]

Think so

[remote arrow]

It was started by @robust yoke talking about Copacabana in Brazilian..

[robust yoke]

He's part of the WWNFPD (World-Wide No Fun Police Department).

[neon idol]

Answers bruh 🥀🥀🥀🥀

[remote arrow]

What was the question? The video limit is 5 x 2

[neon idol]

Lmao

@robust yoke 🥺

Answers? 🥺

[neon idol]

YESSS

[robust yoke]

The video limit is actually…

[remote arrow]

Really? I thought that the most difficult prompt in video generation could be "Mouse dance"

[robust yoke]

\frac{300}{\pi^2} \int_{0}^{1} \int_{0}^{1} \frac{1}{1-xy} dy dx = 50

[neon idol]

Wth is that?

[robust yoke]

The answers, just like you asked.

[remote arrow]

His processor is overloaded..

[neon idol]

Uhm ok?

I have to upload all of 10 questions?

[robust yoke]

Well, I assumed that's what you were gonna do anyway, yes?

[neon idol]

Yes

[robust yoke]

There you go.

[rigid portal]

Everyone is cooked

[robust yoke]

I sautéed them. 🗣️ 🗣️ 🗣️

[remote arrow]

Welcome to Bananiland, the gate to the Sodaland..

[robust yoke]

“Welcome to the jungle!!1!1!1” 🗣️ 🗣️ 🗣️ 🔥 🔥 🔥

[remote arrow]

Watchout!

[rigid portal]

Am not taking to you bro

[rigid portal]

Don't talk to me and also don't reply to me again

[remote arrow]

Yeah. Me too..

[robust yoke]

Sticky banana substance… 😩

[remote arrow]

Be careful, young one, that could be "stick" to your middle name..

[rigid portal]

Can you say it clear pls

I don't get you

[neon idol]

#rules

[lean wigeon]

Hi everyone, new here.

[icy sage]

Hi

[fiery gull]

bruh, the new 2.5 flash still halucinate, it talk the zero soda has sugar

I hope that when it out of preview will have less halucinations

[golden ocean]

real

[true phoenix]

Why doesn't a sound come out in the videos

[echo aurora]

It’s going to be random if your video has sound or not.

Not all models have audio support.

[true phoenix]

There is no solution to this problem

[echo aurora]

No, as the purpose of Video Arena is to evaluate how models perform so it’s going to be in Battle mode, where you can’t select specific models.

[true phoenix]

You have another free site, use

[fiery gull]

why the direct exist ?

is to we use the models for free with nothing?

[echo aurora]

Sometimes getting to the desired outcome requires specific instructions for each new generation a model provides. If you want a specific outcome from X and Y model, that may require different instructions with each turn to get to that desired outcome, so giving our users a way to do that is a benefit. With video that’s going to be a lot more $$$ to do.

We may incorporate ways feedback can be provided via a direct chat method in the future.

[whole sundial]

please look at #1397655624103493813

[heady spear]

Thanks buddy

[untold bronze]

Hi Looking to test different LLM for coding.

[willow anchor]

hello

[trail eagle]

Hi

[echo aurora]

[trail eagle]

How to use seedream?

[wide briar]

Hello there..., again.

[spark vault]

Hi again

[feral basin]

Please remove the 'LOTTE WORLD' watermark from this photo.

[weak wave]

hello

[warped elk]

Greatings everyone ...new in this world of AI

[delicate jungle]

wow this is cool haha

[golden ocean]

https://tenor.com/view/get-blown-up-dog-dog-explosion-sticky-notes-gif-17071604041793579031

[stiff kernel]

Please, check out ⁠ to learn how #1397655624103493813 roperly prompt the bot.

[sour pawn]

I am looking at a video on Youtube abouct LMArena

[wintry tinsel]

Wow I am so amazed at how stagnant and boring frontier AI model space has been for past several months

GPT5 was disappointing and boring

And that was months ago!

When will we get actually good decent meaningful Sota model?

Gemini 3 pro must save us

[stiff kernel]

Hi! Just a quick reminder that messages in this channel should be in English only, as per community rules. Thanks for understanding!

[exotic nebula]

why do all the mods have weird profiles. Like bee, pineapple, JackSparrow, and so on

[exotic nebula]

hell yeah it is damn cool. welcome to the club

[echo aurora]

how so?

[exotic nebula]

hey, no hate, you guys are so chill and one of the best mod committees out there. Most of the mods I know maintain insane professionalism

[granite heron]

Captain*, Captain JackSparrow

[exotic nebula]

Yes my lord forgive this lowly peasant for such a mistake

[granite heron]

https://tenor.com/view/jack-sparrow-hat-tip-thank-you-gif-27250758

[echo aurora]

https://tenor.com/view/ban-bongo-cat-gif-18085186045616341050

[remote arrow]

☕ finally..

[shy basin]

Hey there. Is there any way of increasing the chances of getting a video with sound. I would really love a video with sound, but every time that I try to generate a video, I’m not getting it. The videos themselves are really, really good though,I’m very impressed.

[echo aurora]

Sorry to say there is not

It's going to be random, even if you include in the prompt ~"use X model"

[shy basin]

Darn. Why did they put all of the video models in the discord server and then put the other stuff in their website. They should’ve just streamlined everything so we could have full control over what video model we’re using. I’m just curious.

I’m also trying to make a short movie with the videos I’ve been generating since they have all been super good and there’s always the one model that’s better than the other that I can use. Is there any way to increase the chances of something generating something that has already been generated, but with a different image. I like to use image to video a lot, and then put different types of things in those images as if they were in the roomor seen that I took the photo.

Like I generated a video of a certain type of dinosaur in a college campus space and I want to know if I can get the same thing but in a different image

[echo aurora]

Well we are treating it as an experiment as it is the first time we've put a battle mode for a modality in the Discord server compared to the site. Reason being with videos others could find inspiration and discuss what others are creating. It's possible we move Video Arena onto the site in the future.

[warped elk]

👍

[past roost]

@echo aurora can u please remove limited image generation for seedream 4

[echo aurora]

Sorry to say that's unlikely we remove rate limits

[past roost]

@echo aurora Please remove the restrictions for Seedream 4 so that we can generate unlimited images.

[echo ibex]

hi

[whole sundial]

idk if they will but i'm pretty sure you can use nano-banana more than seedream even though seedream is a bit cheaper. they are probably putting limits on seedream to try to stop them draining money (LMArena's newfound popularity is costing them a lot of money, also why video gen was reduced) I also don't like how they secretly downgraded gpt-image-1 to the low preset to save money. That is not fair to the model or OpenAI. They also removed the flux kontext model for the same reason. I have a feeling they are intentionally being vague and secretive with the removals and limits because they don't want people finding out that they are draining too much money now with hundreds of thousands of people using it and they are doing whatever they can to stop, same reason why the google requirement was put in place. I still highly doubt that they will ever add other login options, yet more lies from LMArena.

idk if they will ever be profitable, yet another example of the internet ruining something cool for everybody. their next step will probably be either trying to add payment or force the AI companies to foot the entire bill instead of LMArena paying for it themselves. It was fine when it was still a university project, companies donated inference to them in exchange for feedback but since they are a proper company now, it's just discounts for stealth models, they pay for all of it and image gen is expensive, they should just admit they are doing these things for cost cutting instead of being vague about it. gpt-image-1 is the best example, high quality images like what they were using cost 0.24 each but now they are using the much cheaper low option that is like 3 or 4 cents each. I brought it up to them many times but they never told me the truth: it's cost cutting, not a mistake, not OpenAI making sneaky changes. just them trying to stop the bleed.

[simple robin]

hello all

[whole sundial]

I used to enjoy LMArena and its image gen when it was not that popular. But now with many people using it and the platform being nerfed because of it, I simply don't like it anymore. I still remember the terrible Gradio interface, but I went through it fine and when the new site came out earlier this year, I really liked it and I heavily used gpt-image-1 in high quality on the site with no problems. But now I can't anymore, and I don't have the money to get ChatGPT Plus (which is the reason why I used LMArena for it in the first place), so I've just stopped using that model.

the text portion is still fine, but I feel like chopping that up is next. don't be surprised if future models have worse limits on there. They messed up video gen, then they messed up image gen, text mode is next.

and they will probably force Google login on there as well soon enough. China uses LMArena proxies a lot so I'm surprised they haven't done it there yet (Google isn't available in China, so those users will have to find ways around the limits)

surprised they said Google wasn't giving them money for login exclusivity, but companies lie all the time so you can never truly trust them, especially when their business is centered around giving your data to other people (including Google, and Google makes money the same way!) and they also illegally voided US users' 7th Amendment right to a trial by jury by forcing people into arbitration without their knowledge. only way to opt out is by sending them an email (that info is hidden deep in their ToS), but who knows if they'll listen?

[ivory schooner]

bug

kimi-k2-0905-preview and kimi-k2-0711-preview dying?

[whole sundial]

yeah they don't work for me either

Moonshot API is fine though

maybe they ran out of credits, they are becoming too vague with their errors now. Back in the Gradio days, they would pass over whatever error they got so you can see it (like a rate limit, for example) instead of being vague and saying "Something went wrong." (That's not not helpful to anybody, especially the LMArena staff who is supposed to deal with these errors! Would be helpful if they just gave you the error they got so LMArena immediately knows what's wrong instead of having to look through logs. And besides, "Something" could mean anything! Credits ran out, rate limit exceeded, provider down, dangerous content, model discontinued, or maybe the LMArena cat ran over an employee's keyboard and is causing these problems. Nobody knows what's wrong in a vague system like this!)

also guys, please do not blame pineapple for these problems! he's just an innocent messager, he has little power in LMArena's decisions, all he does is manage this server and relay feedback, bugs, new models, etc. to their team. it is not his fault that any of this is happening.

[hazy blaze]

Some of the video models are horrific

[viscid cloak]

Could anyone check the status of image gen? models started to read past images and prompts and editted based on them. It would occur even if I deleted previous tabs. This incident once took place when multi-turn edit was introduced couple weeks ago and now it came back.

[hazy blaze]

Only veo3 is the goat

[viscid cloak]

Plus, BATTLE would pull up results from two same models now.

[ivory schooner]

Is it me having too many conversations with Kimi or is it a system issue?

[whole sundial]

it's a problem on their side

[ivory schooner]

Ah ah, when can we fix Kimi

[whole sundial]

whenever lmarena gets around to fixing whatever problem they have, i think their credits ran out

just a guess though, I wish LMArena shared more info about errors like they used to

[glad root]

Hello 👋

[tulip stone]

try out my new game in the ai creations

[tawny kettle]

pause, birds chirping)
Every day, I see the sun rise…
and I hear the birds sing.

[hardy fractal]

hey helo

[tardy rose]

Hello 👐

[zinc hearth]

Hello Peeps - looking forward meeting some of you

[fading rover]

Hello all is there anyone working on seedream 4 ?? I need help with something as I see it does not really convey the natural tones when go into art mode and make things really vibrant unnecessary and also not imitating the drawing style unlike nano banana

[subtle pecan]

Hello I enjoy using AI

[leaden sun]

"profitable" is not only a matter of technology greatness but a matter of how to run a business that is preferably sustainable like a circular economy, and given the fact that the current political environment, mostly in the socalled collective west, is pretty unstable (civil unrest, incoming wars, regime changes, collapsing economies and a looming financial crisis like a great depression etc.) it's a turbulent time not only for AI businesses, most companies are bracing for a global downturn...

[fading rover]

Seedream 4 seem not generating giving error "Something went wrong with this response, please try again" tried 3 different login on my mobile all same.

[cunning vale]

hello

[tropic ether]

How can I use AI models in server?

[woven quartz]

Hi

[fiery gull]

this server you just can use video generator #1397655624103493813

[random wolf]

when do they gonna put, cancel generating? it's so frustrating

[fiery gull]

this is to avoid bots

don't repeat this so many times, or you can be banned

[random wolf]

why? I didn't know about that, why I can be ban? and I'm just saying a problem brom

[dusky ravine]

It's on progress, it's kinda frustating for me too expecially with generating prompt being stuck. But they're doing something about it idk, probably adding edit button or fix the bug

[fiery gull]

the anti bot may suspect that you are a bot, and ban your IP

[hollow ivy]

How good is the new Deepseek in coding and roleplaying?

[fiery gull]

see in lmarena leadboard, creative write and coding

but now is the 3.2v

[fiery gull]

roleplay

[fiery gull]

coding

[dusky ravine]

wdym roleplay? you can use this model for roleplay?

[hollow ivy]

yeah, just prompt it, to be a gamemaster (GM)

it's actually fun

(with some [advanced] models)

eg. Gemini

(for coding, Gemini is too weak, though)

[dusky ravine]

how? mind if you give me roleplay prompt example?

[fiery gull]

I use it for teacher

[hollow ivy]

yes, wait..

..where internal consistency, realism and character-persistence was paramount
also NPC-intelligence and humanlikeness
and where the LLM had to play a really intelligent and impartial gamemaster who made the game immersive and fun to play```
(taken verbatim from another discord)

[dusky ravine]

for learning what is it?

[fiery gull]

I am using it to learn english, pedagogy, pharmacy and prompt

[dusky ravine]

oh like a chatbot

[hollow ivy]

just reword that, to prime the LLM to be[come] that entity

[dusky ravine]

ok thanks!

[hollow ivy]

yeah, these are excellent to practice english

(and learning english actually makes fun with them)

[fiery gull]

I'm using the learning it gemini app, its soo op

I recommend it

[hollow ivy]

yeah, Gemini has an awesome language understanding

feels, as if it could pass the (long-version) Turing test with its next version (v3.0)

[fiery gull]

gemini guided learning

[hollow ivy]

i wonder what gemini 3 pro will be able to do..

what do you think?

will it be better in coding than GPT5-high and Claude Opus 4.1 Thinking?

[fiery gull]

Idk about this, I don't code

[hollow ivy]

gemini 3 flash preview seems to be available already (?)

[fiery gull]

but in the rest I think the 2.5 pro is better than 4.1 and 5

in my use ok

[fiery gull]

hell nah, just now they release the 2.5 preview

first they to need release the gemma 4 to release the flash/pro 3.0

[hollow ivy]

gemini flash latest > gemini flash 2.5

2.5 flash existed already

[fiery gull]

I'm using it

[hollow ivy]

this new one is in upgrade

i think it is oceanstone (in battle mode)

[fiery gull]

but sometime it halucinate

[hollow ivy]

yeah :/

and sometimes it's mixing up things

[fiery gull]

I thought this oceanstore was gemma 4, but I don't know, I still have hope

[hollow ivy]

if it was gemma 4, that would be astonishing

because then new gemini would be even stronger

gemma is supposed to suck (in comparison to gemini)

[fiery gull]

ocenstone +20b, oceanreef ~10b

I think this

[hollow ivy]

and skytrail?

[fiery gull]

nope, the gemma 3 27b is better than flash and pro 2.0

[hollow ivy]

pro 2.0 was a dud

way below 2.5 pro

[fiery gull]

shytrails Idk what it is it could be

[hollow spade]

hello

[fiery gull]

hi

[hollow ivy]

bonsoir monsieur

[hollow spade]

i want to use them all in free and build something effective and help to the people

[hollow ivy]

great goal, good luck

[fiery gull]

use what?

[hollow spade]

use all model

[fiery gull]

ah okay 🙂

[hollow ivy]

currently the best coding models are: GPT5-high and Claude Opus 4.1 Thinking

[fiery gull]

the 5 codex don't is the better?

[hollow ivy]

place 3 are gemini 2.5 pro and Grok 4

codex, ah idk about that critter

isnt that a GUI?

[fiery gull]

nope, is a new model

[hollow ivy]

ah right

but only a small incremental change, i guess?

[fiery gull]

I test a one time it, I liked it making a svg images

[hollow ivy]

(..compared to GPT5-high)

is codex in battle mode?

[fiery gull]

idk

[fiery gull]

webdev arena

web.lmarena.ai

[hollow ivy]

have you ever encountered sorting-hat?

or route66?

[crude quarry]

HIU

[hollow ivy]

or nightride?

[fiery gull]

Hi

idk, I don't use the codex, just a one time

hell nah, I don't make a ideia what the route66, sorting hat or nightride is this

I just think the nightride is chinese

[hollow ivy]

route66 should be a chatgpt version (?)

[fiery gull]

why?

[hollow ivy]

wait, i have it in my text-files..

[hollow ivy]

therefore:

it answers like chatGPT

and sorting-hat answered like Gemini

but the latter one is hard to get

[fiery gull]

hell no, deepseek in the past it said that it is the gpt

[hollow ivy]

hm, but not consistently

it said that it was several

[fiery gull]

HMMMMMMMMM

sure

[hollow ivy]

deepseek identified as:

• claude
• chatgpt
• gemini

iirc, it was by random

[fiery gull]

is a new oss?

okay, make sense now

[hollow ivy]

but never as grok (iirc)

[fiery gull]

what meen iirc?

[hollow ivy]

i heard, Grok 4 can be good for roleplaying, too

[fiery gull]

🙃

[hollow ivy]

if i recall correctly

[fiery gull]

okay

[hollow ivy]

they already plan grok 5

maybe next quarter or so?

[fiery gull]

I'm excited for 3.0 pro too, but I think it will be paid for on aistudio 🥲

[fiery gull]

Idk, the xai release the grok 4 fast now

[hollow ivy]

really?
but i heard, gemini 3.0 flash will be better than 2.5 pro

[fiery gull]

the 3.0 flash lite need to be better than 2.5 pro

[sullen quest]

?