general

[echo aurora]

It's fine, our site being down like this isn't an okay thing, we are going to do better.

[fierce cove]

Agreed
our feedback is constructive, not a complaint

[glossy umbra]

That’s great. Just don’t be pressured. work at your own pace.

[glossy umbra]

😁

They said they would do better. Let’s be respectful atleast. 😅

[naive rapids]

yo pineapple you've been doing great all day, thank you for being patient with us as well

[echo aurora]

This is an overall issue we're aware on and would like to fix.

Mid-outage though that's going to be much trickier.

[echo aurora]

[viral notch]

okay nvm the site only looks like it works... i can only read the conversations but not add to them or start new ones

[ornate swift]

Bbb

[glossy umbra]

Well sorry if it bothers you. We don’t deserve what’s free. Just know you could, theoretically, not be able to use it anyday. Obviously that’s not going to happen. But be respectful. 😁

[echo aurora]

Honestly, it's the engineers who are working in this that are the true heroes with these outages.

[glossy umbra]

@wanton vortex

[wanton vortex]

the admin is pissed off ig XD

[stuck kindle]

What time will it come back?

[stuck kindle]

What time will it come back?

[echo aurora]

No not at all!

[viral notch]

i suspect the site is going through feature transitions so some code may not cooperate properly

[echo aurora]

I don't have an ETA for you sorry to say.

[minor burrow]

@echo aurora Thank you for your response. I think attaching files is one of the fundamental features. I haven’t seen an admin like you replying to every user’s message. Hats off to you!

[stuck kindle]

The chance to come back today?

[fierce cove]

Google appears free, but its true cost depends on your definition of "free." Lmarena also isn't free; users provide data to improve its models. True free means receiving nothing in return. Nevertheless, I am grateful to Lmarena for this service ♥️

[tight oriole]

is more than 24 hours down?

[viral notch]

im glad its free in terms of no fees or subscriptions

[echo aurora]

Yeah I'd be very surprised if it wasn't back today.

[viral notch]

usually the site isnt out of commission for multiple days

[wanton vortex]

ig yaal have to open a site like status.lmarena.???

[stuck kindle]

That's great, you do a great job, congratulations!

[sullen sand]

What happened

[glossy umbra]

Site is down. Wait until it’s fixed

[sullen sand]

Thanks for saying yo me

[zealous turtle]

Hey all! While the engineering team is working on a fix, please keep chat and tone respectful 🙂

Updates will be posted in #announcements.

Thanks for your patience and understanding! ☘️

[jovial sapphire]

hi guys

have you seen gpt5 image model?

it's pretty bad lol

there's a new model being tested on artificial analysis

it's not gpt1 image

[glossy umbra]

It is ‘bad’ because it can’t edit an element of an image without changing everything else

[jovial sapphire]

it's "gpt5"

[jovial sapphire]

yeah

[glossy umbra]

Nano banana is better at that.

[jovial sapphire]

exactly

[primal orbit]

site is up, but no chat history

[jovial sapphire]

even qwen is better

or seedit

[narrow dawn]

is lma up again?

[glossy umbra]

Yes

[echo aurora]

https://tenor.com/view/futurama-were-back-baby-here-we-go-gif-12141676

[narrow dawn]

ok i have every chat

[hollow ice]

I actually used it 2 hrs ago but now 😭😭

[sullen sand]

Yeah

Baby

[echo aurora]

We're still monitoring

[sullen sand]

Good job bro

Lmarena the best

[primal orbit]

are you logged in to the site?

[narrow dawn]

ye

[wanton vortex]

gg

[primal orbit]

I was not logged in and all previous chats are gone 🙁

[wanton vortex]

love yaal

💕

[echo aurora]

I'm really sorry to hear that. Do you have the sign-in with Google option?

[narrow dawn]

🙁 that should be fixxed in a other way not only logged in thats bad 🙁

[primal orbit]

yes, but I didn't use it. Was fine without it.

[storm ermine]

Where do you guys make your videos?

[torpid locust]

what just happen?

[echo aurora]

Sorry to say the Video Arena bot is still not working.

[narrow dawn]

i can access

[narrow dawn]

i have a question pineapple

[echo aurora]

fire away

[narrow dawn]

how do you get the ai's to work is it with api or how if i can ask or allowed to

[echo aurora]

Team is aware and looking into this btw.

[echo aurora]

Sorry to say we don't have an API that's accessible

[narrow dawn]

huh?

i don't understand so i could do my own website with ai or what?

[glossy umbra]

He's asking how do you get all the AI models on the website to work (e.g. GPT-5-HIGH)

[narrow dawn]

that too

[glossy umbra]

Isnt it pretty straightforward? Just call the API for gpt-5-high or whatever and retreive the LLM response.

[narrow dawn]

better question does it cost money

[glossy umbra]

Yes, it does.

[narrow dawn]

ohhh

[storm ermine]

What is the best way to make AI videos for YouTube shorts for free?

[cedar tide]

@echo aurora qwen 3 max 😶

[neon idol]

Is good or sucks?

[cedar tide]

Very good

[ashen plaza]

im still getting rate limited after a single generation. of course im not logged in. is this intentional now or just a bug left over from everything that was going on? as this happened before everything went down originally

[glossy umbra]

What model are you using

[neon idol]

More intelligent than gpt 5?

[ashen plaza]

its just battle mode, it is image generation but still just one and then it tells me to log in

[neon idol]

Site is working for me

[glossy umbra]

Might be a restriction.

I don't work for LMArena so I can't tell you exactly

[cedar tide]

Now available just the non reasoning version, and yes its more inteligent than GPT 5 No reasoning

it rather competes with Claude opus

[narrow dawn]

for what do you guys use the LMArena?

[ashen plaza]

regular text generations are working normally it seems just image generation is having issues. weird

[unkempt elk]

where did the google login and my chats go after the website down
nvm i refreshed site 3 times and all back

[glossy umbra]

It's highly unlikely that it's an inposed restriction. Nano banana should be free

[glossy umbra]

Your google chats should be saved. Else, it's not confirmed that you'll have the chats after a website outage.

[ashen plaza]

nano is one of the limited ones but never a single at a time. i have no clue whats going on and will assume theres a left over bug still

[unkempt elk]

dw i got them back

[narrow dawn]

for what do you guys use the LMArena?

[glossy umbra]

Research

[narrow dawn]

:0

[green plume]

its back

omg

wait its back ?

or am i trippin ?

[glossy umbra]

Yes it is. Scroll up.

[green plume]

https://tenor.com/view/cat-kitty-small-pet-silly-alone-lonely-gif-4734322154604199486

[viral notch]

glad its back

[mild garnet]

/video

[ashen plaza]

@echo aurora sorry for the ping just had to ask. is the rate limit of only a few generations a bug (as it started before everything began to cash), temporary thing while there was issues or permanent now?

[gray quartz]

hello guys

[regal cairn]

gpt-image-1-high-fidelity 🤔

[crimson escarp]

"high fidelity"

[past marsh]

Hi

[fossil fable]

Actually gonna crash out the new release is CLOSED SHUT

i had a hunch

it was correct

[keen fulcrum]

https://tenor.com/view/soviet-cat-sovicat-soviet-ussr-cat-gif-3531724627123465264

[old halo]

hi

[rigid bay]

cinematic

[compact turret]

Sim

Yes

[thorny kayak]

Hi

How to change images to video

[undone flint]

/video

[turbid verge]

hello

[balmy mist]

how is the new Kimi K2 update?

[formal jungle]

You guys can't see this but I I am seeing the most adorable bunny rabbit

[subtle forum]

Hola

[echo aurora]

Note our Video Arena bot isn't working at the moment, @undone flint

[open mountain]

What is the new gpt image model? where

[echo aurora]

I'm not sure what you're referring to.

[open mountain]

@echo aurora

gpt-image-1-high-fidelity

[tight oriole]

is video working in next 5 -12 hours?

[echo aurora]

Sorry to say I don't have an estimated time for you.

[smoky ice]

/video

[echo aurora]

You can't use the bot in this channel. Check out #1397655624103493813 for more info, but note the bot isn't working at the moment.

[echo aurora]

Let me check.

[mild minnow]

I hate Claude on LMArena. I can’t even make a proper Doom Slayer story without getting a Terms of Service violation warning. 😭😭😭 And Claude is so damn good at making stories too. My life is over

https://tenor.com/view/deadpool-gif-26545415

[polar marlin]

😭

[lucid mural]

/movie

I hate Claude on LMArena. I can’t even make a proper Doom Slayer story without getting a Terms of Service violation warning. 😭😭😭 And Claude is so damn good at making stories too. My life is over

[open mountain]

I hate Claude on LMArena. I can’t even make a proper Doom Slayer story without getting a Terms of Service violation warning. 😭😭😭 And Claude is so damn good at making stories too. My life is over

[willow grail]

https://tenor.com/view/cat-kitty-small-pet-silly-alone-lonely-gif-4734322154604199486

[willow grail]

first of all claude sucks at story making and swe.

next.

[open mountain]

In Russian, Claude swears more than GPT

[barren prairie]

New trend ? 💀

[open mountain]

And Claude responds more humanly than gpt

[willow grail]

use gpt5 high for story making

[barren prairie]

I didn t like it so much..

[open mountain]

Models that deliberate respond less humanly than those that do not think before responding

хорошо

[willow grail]

gpt5 medium to high wins in all fiction novel stuff

[open mountain]

and where to find this version - medium?

[willow grail]

api

are u new ?

use openrouter or openais playground api

[willow grail]

and novelcrafter is the software where u write the story

[open mountain]

where

[latent crest]

Is it that good?

[open mountain]

Where to find her

@regal cairn

[latent crest]

I loved that book!

What does High fidelity have to do with ChatGPT

[regal cairn]

It rarely appears in the lmbattle.

[open mountain]

It's hard to find (

@echo aurora when will gpt image high be available

[tender sphinx]

[open mountain]

say hi

[mild minnow]

Yeah it’s great actually but it can be such an ass sometimes

[regal cairn]

It appears to me twice but it hasn't appeared again, but recently it appeared again to a friend

[latent crest]

Hold on, do we have to login now? I don’t want to

[inner wadi]

Hi when will bot start again

[echo aurora]

Yeah so that's a new model that's only going to be available on Battle mode, you won't be able to find in Direct/Side by Side. cc @open mountain

[echo aurora]

Unclear, I don't have an ETA.

[mild minnow]

https://cdn.discordapp.com/attachments/1161614731036074017/1409457734675595434/image0.gif

[open mountain]

And when will it be public? direct

[echo aurora]

It's not certain if it will, so I can't provide an ETA.

[mild minnow]

Will Arena ever have an App for mobile

[echo aurora]

It's possible.

[open mountain]

perhaps never?... I didn't understand

[mild minnow]

Hmm will it likely have in-app purchases?

I mean there has to be a way to keep apps alive no?

[sage thistle]

Hello, Just Join Here

[mild minnow]

Welcome

[echo aurora]

Correct

[echo aurora]

It'd likely be very similar to our site if we were to make one.

[mild minnow]

Then you would have the most popular app on the appstore

Most apps on the appstore shove in-app purchases down your throat which can be expensive as hell

[ashen plaza]

@echo aurora so is this still just a bug to only be bale to generate about 2 images now? or some kind of safe guard brought up when everything was crashing? or possibly a bug left over from the website going under as i was getting this same thing just before it all started

[fathom jackal]

I also have this doubt, after all the site's proposal was not to require logging in.

[proud hazel]

Yes, I'm a little annoyed by this bad decision. I would like to remain anonymous.

[faint torrent]

Hi everybody!! Why I don't have permission to write in # video-arena-1, 2 or 3? Thanks

[keen ferry]

I pick "or"

[wooden crown]

/video

[orchid orbit]

Why can't I take photos and videos?

[wooden crown]

/list

la

ls

df

[real whale]

What happen

[ashen plaza]

https://tenor.com/view/what-mickey-wha-happen-happen-shocked-gif-16923342

[lusty tinsel]

cant they add option to login with other email instead only gmail? like outlook, etc...

[ornate swift]

Why lmerna dont working??

[ashen plaza]

what issue in specific?

[ornate swift]

@ashen plaza go to nani banana

[ashen plaza]

you do understand you might have diffrent issues right? say what the issue is

[proud hazel]

https://tenor.com/view/anime-what-nani-เอาไงดี-gunkun-gif-8067161216381491543

[devout vault]

@echo aurora nice typing in annoucements

[echo aurora]

yes

[devout vault]

gemini 3 when

[leaden laurel]

was it that anonymous-model-0514 thing?

or what was its name

[whole sundial]

btw @echo aurora you should have made some sort of announcement before you guys decided that it was a good idea to force Google login after 3 image generations

not everybody wants to make a Google account, you know?

[echo aurora]

Which model?

[whole sundial]

just because 99% of the world uses Google doesn't mean that you should ignore the other 1% that doesn't use Google

[whole sundial]

there is a stealth model in the image arena that has been there for a few months called that

[echo aurora]

I agree, we're crafting something atm

[echo aurora]

We are looking to expand User Login to not just Google

[solemn hedge]

Noticed that you added the account feature lets goo!

[leaden laurel]

idk

it was called like that

[gleaming roost]

I knew that with the implementation of login would come higher rate limit fees

[civic spindle]

vro is there a private video generator so i can make free unlimited vids for myself, pfft

[boreal saddle]

Bruh, why say vro

What is this bastardized word?

[timber iris]

What are detailed limits?

[civic spindle]

don't worry bout it unc

[boreal saddle]

Am I really that old?

[willow grail]

i would choose banning all new member who came in in the last few weeks

so i can do more usage in here

D:

[timber iris]

What are detailed limits?

[civic spindle]

does lmarena have unlimited free uncensored private every single chat, image, video model for free with no paywalls

[echo aurora]

The bot can only be used in this server

[civic spindle]

yes but, dont want people to see it

[echo aurora]

We don't have those posted somewhere; however, this was flagged to the team as something we may do going forward.

[timber iris]

ok

[echo aurora]

Sorry to say I don't have a solution for you

[civic spindle]

took me 1 minute and 43 seconds to realize you're a mod here

[potent glacier]

@echo aurora Are the rate limits for direct chat or Battle Mode now? I was only getting rate limited in direct chat even before the login was implemented.

[echo aurora]

Oddly specific

[civic spindle]

what is a rate limit

[echo aurora]

I believe it's just in Direct/Side by Side but will double check

[echo aurora]

Use limit essentially

[potent glacier]

Please do 🙏🏻 As far as I know Battle Mode is supposed to not have a limit

The other models were always rate limited before

Idk why it needed to be implemented if it was already a thing

[civic spindle]

theres limits now? chat limit? photo limit? for what

this used to be unlimited

[glossy umbra]

Be grateful for what you have. We don’t deserve it.

[echo aurora]

Yeah I'm 99% sure Battle doesn't have these limits, as if you run into a limit with a model, it'll sample a different one.

[civic spindle]

thats not what im sayin, im sayin it used to be unlimited, why is it limited now

[glossy umbra]

Because these models are not free

Don’t take them for granted

[potent glacier]

No it was always rate limited before

[glossy umbra]

Be grateful

[echo aurora]

Always has had rate limits. But it depends on the model

[potent glacier]

I’m very confused

Rate limits already existed before

So what exactly was added or changed?

[civic spindle]

lemme guess theres a subscription plan to bypass the limits

49.99 dollars a month

[echo aurora]

There is not

[civic spindle]

but whats changed now if the limits still existed

[potent glacier]

Yeah I’m confused as well

[echo aurora]

Users who are logged in are going to have less rate limits

[magic stag]

"you would ask me to pay to use the technology developed by world wide consumation of millions of hours of labor and research? how dare you"

[civic spindle]

see bro this guy gets it

[potent glacier]

This is why I love local generation 😊

[civic spindle]

mind games

[potent glacier]

I can use it unlimited

[echo aurora]

Both will still have limits, but if logged in you're going to have less. Does that make sense?

[glossy umbra]

Why can’t people just be grateful that there’s a place where you can use LLMs for free? even though there are limits?

[ashen plaza]

as far as I can tell if you arent logged in you get like 2 or 3 generations before being rate limited, even on battle

[potent glacier]

Whoa

Battle Mode isn’t supposed to be rate limited though?

[civic spindle]

oh, less limits for logged in people, oh yee i get it cus i already was logged in with an account on it

[glossy umbra]

Be more specific. What LLM?

[potent glacier]

That was always unlimited

[ashen plaza]

as far as I can tell all of them for image generation.

[echo aurora]

Still double checking on this. There may be a difference for image/text.

[potent glacier]

I think a lot of people are about to get angry lol

[civic spindle]

why cant i send a video game character being mauled to a direct chat to gpt with it constantly being flagged

[gusty loom]

guys what is the best web search model currently

[remote arrow]

After force login, now IP banning. LMArena was a kind a hero, until about... A week ago..

[glossy umbra]

IP banning? elaborate

[ashen plaza]

https://tenor.com/view/jeff-goldblum-gif-12073481

[echo aurora]

Yeah confirmed there are rate limits in Battle mode.

[civic spindle]

lads what happened to story telling lads, cant even send a pic of some game screenshot scenario on there without being flagged, certainly not REAL

[ashen plaza]

the force login as far as it seems atm is just temporary, based on what i picked up from chat earlier there seems to have been a mssive flood of bots on the website and discord and this is to prevent that. of course i could be wrong

[gusty loom]

guys what is the best web search model currently?

[proud hazel]

o3

[gusty loom]

thx

[fathom jackal]

No, it should have stayed the way it was instead of catching users off guard and causing anxiety... if you propose to offer something accessible without the need to log in and get your users used to it, only to remove this feature later, then you should have done that from the beginning.🤷‍♂️

[civic spindle]

admin the censorship thing is so geeked rn

[glossy umbra]

I think the restrictions were necessary though

The site must have went down for a reason

[ashen plaza]

at about the same time the website went down google had a major outage too. if they are linked who knows could just be a conicidence

[whole sundial]

tbh i think one of the main reasons they did this is because of bots that use lmarena, and also to lower the load so more generations from real people go through. I don't like the Google requirement in particular, but they are coming up with other options, hopefully email login.

[glossy umbra]

I doubt they’re linked. Pineapple said the LMArena devs specifically are working to get the site back working. If it was a cloud flare issue then they would just have needed to wait for them to fix it

[echo aurora]

I appreciate the feedback and will be sure to share with the team

[ashen plaza]

yeah could be. my idea was google had a major outage causing the initial issues last night, but it just ended up becoming much worse due to said outage and possibly made a new issue. idk im just throwing out a random theory with litterally 0 backing and im probably completly wrong

https://tenor.com/view/just-a-theory-matpat-fnaf-movie-matthew-robert-patrick-gif-8650789653163891524

[glossy umbra]

We don’t know. Regardless, if there were even bots on the site, there should have been a better way of preventing them.

[lusty tinsel]

didnt they have captcha going on for every generation?

[glossy umbra]

You must refer to cloud flare

[lusty tinsel]

ig

[glossy umbra]

Cloudflare captcha is dependent on other factors. It’s not a fixed rule that it should be every generation.

It can happen every generation. But it’s not fixed. It depends

[lusty tinsel]

i got that a lot before. guess i was unlucky

[whole sundial]

i realized that the most popular lmarena proxy solution is from china, i get constant errors when it's late at night, which also happens to be the time that chinese people are up. hmm...

[ashen plaza]

also accounts really dont stop bots. people can and will make bot accounts after all. it really does seem the main reason accounts were added was for saving stuff and keeping your generations

[neon idol]

@echo aurora What is the limit of generated images for people that have made the login?

[glossy umbra]

Probably not linked. But if I had to guess, the bots would most likely be chinese.

[glossy umbra]

That’s the mask. Underneath it was probably to stop bots.

[ashen plaza]

https://tenor.com/view/plan-conspiracy-theories-gif-27390691

us

[echo aurora]

We don't have this listed somewhere, I'm checking with the team 👍

[whole sundial]

makes sense to me. western ai companies are blocked in china, but they also have to find ways to access those models for benchmarking, comparisons, distillation, etc. so they may be using LMArena for those purposes. Google is also blocked in China, so they'll have to find a way around this.

[neon idol]

let me know please

[glossy umbra]

Good point. I also have some private reasons to believe they’re chinese, maybe using VPNs to mask themselves or whatever.

[potent glacier]

Oof…that’s a big no

Battle Mode isn’t supposed to be rate limited at all

That’s honestly ridiculous and should be reverted to how it was

[misty star]

DH3 is Seedream 4.0

https://x.com/swishfever/status/1964064463717281992

fishy business (@swishfever)

DH3 on the Artificial Analysis Image Arena appears to be Seedream 4.0 from ByteDance

SoTA image generation and editing

X · Sep 5, 2025, 8:33 PM

[high hound]

It's official. LMArena is slowly becoming paid. The one service that was known to be non-biased and provide a good place to use AI for relatively free (with some restrictions), is now getting more and more restricted. Forcing you to make accounts ...

We all know what comes next. Subscription models. Brute-forcing users to login...

[neon idol]

WHAT THE HELL

[misty star]

This is 10 days after nano banana

Lmaoo

💀

Idk

[neon idol]

Where can I try it

[dawn wharf]

Seedream 3 was already very cool

[potent glacier]

I wouldn’t say that unless there’s proof

[glossy umbra]

Nobody knows the devs intention
Just agree for them to sell your data at the cost of using the service for free and call it a day 👍🏽

[limpid schooner]

it's SOTA

[dawn wharf]

but it was bad in prompt following

and complex scenes

had a very good style tho

[golden ocean]

GIVE EXAMPLES

[limpid schooner]

LEFT

left broski

dh3

[neon idol]

How can we try this model?

[limpid schooner]

https://artificialanalysis.ai/text-to-image/arena?tab=arena it's here, you cannot write your own prompts

it will come up randomly often

[neon idol]

CHAT WHAT

I have found a crazy thing

Look at the model

GPT 5?????

[potent glacier]

I will say that having Battle Mode be rate limited now is scummy and shouldn’t have been done

[inner wadi]

Guys anyone using piclumen

[neon idol]

No one cares lol

Ok

[keen beacon]

goodbye free gpt 5

[neon idol]

Fr

[topaz trellis]

hi

[glossy umbra]

I understand LMArena . After all it needs to be monetized

they cant just burn their wallets endlessly

[keen beacon]

no, vc will give them more moni

[potent glacier]

@echo aurora All this stuff definitely goes against the mission statement of LMArena

They shouldn’t say it’s ‘free and open for everyone’

That’s false advertisement

I really don’t think the Battle Mode should be rate limited either because that doesn’t even make sense

[hard quiver]

Down again?

[potent glacier]

They’re just implementing nonsense against us now

All because the website got popular

That’s the only explanation

If you’re rate limiting us, we should get an explanation as to why and also what the rate limits are

Like…Gemini 2.5 Pro itself doesn’t really have rate limits to begin with

So how does that make sense?

There wasn’t an announcement about it or anything, just blindsided your users and expected them to be happy with the changes

[neon idol]

Yeah you are right

[echo aurora]

The team is listening, I do appreciate the feedback.

[potent glacier]

The team shouldn’t have blindsided everyone like that

That is highly unprofessional

A proper announcement is what usually happens in situations where drastic changes are occurring

Yeah the site got popular, because it’s a great site

But the kinds of changes that are trying to be implemented will make a less greater site

The whole rate limiting Battle Mode really leaves a bad taste in my mouth

I know I’m complaining but if people don’t speak their minds and stand up against these things then people get walked all over

And it really does go against what the mission statement is of LMArena

[drifting crow]

Cry about it

[drifting crow]

Yh u should ask for refund

[echo aurora]

Let's be nice, it's fine for users to voice their concerns to us.

[neon idol]

We know something about rate limit for logging user?

[potent glacier]

Cry about it? No, I’m raising valid complaints against a company that clearly wasn’t transparent about their intentions to their users

That’s called ‘holding accountability’

Or do you just let people run roughshod all over you?

[drifting crow]

I don’t typically get upset at free services coz I’m not entitled

[hard quiver]

Where i can use seedream 4.0

[echo aurora]

At the moment, we're not going to share the specific rate limits due to changes. I am passing along the feedback.

[potent glacier]

Has nothing to do with entitlement, has everything to do with their mission statement that is clearly on their website

[drifting crow]

No it doesn’t, it has to do with your interpretation of it

[potent glacier]

No

[neon idol]

Yeah this is my same question but I think that is not actually global

[neon idol]

Ok

[hard quiver]

😔

[neon idol]

Yeah my same reaction

[hard quiver]

And seedream is better than nanobanana or comparable?

[neon idol]

I am very curious to see how good is seedream v4

[neon idol]

Idk, but results are good

[hard quiver]

I hope the use of artificial intelligence never gets monopolized. All this competition in the market is driving the creation of better and better tools.

[drifting crow]

I hope I get the sole ownership of the monopoly

[echo aurora]

Sorry I missed this, is the site down for you?

[tall summit]

i agree

[misty star]

Not out yet

[hard quiver]

Up again

Please, may a billionaire use all their money to create an artificial intelligence that allows image creation and editing without censorship

[empty stump]

grok sorta

[onyx python]

I think I hit my limit using nano banana when using the website. I did not get a message at all though. When does the limit reset?

[echo aurora]

Generally, we're not going to provided details on if/when specific models are going to be added.

So lets keep it mysterious then

But yeah like I said I won't be sharing details about upcoming models until we're ready to announce.

[echo aurora]

About 1 hour after the limit is hit.

[boreal saddle]

Another outage.

[kind merlin]

Hello, excited to start this new journey exploring the world of AI.

[boreal saddle]

Well, you arrived at a pretty bad time.

It keeps disconnecting.

[polar marlin]

😭

[gleaming roost]

and here we go again

[terse shuttle]

what happening?

[boreal saddle]

The website s stability is worse than a DDoS, fr

[topaz sun]

Why do you guys restrict the models so much? Imposing logins? I've been using it since early lm-sys . The restrictions were not that bad. @echo aurora

[echo aurora]

I was seeing the same; however, it's now working. Are you seeing the same?

[crimson escarp]

do we have any ideas about image generation limits? like how many generations am i allowed for gpt-image a day?

[echo aurora]

We've been seeing unexpected traffic and interest in the site, as a result raising the rate limits was necessary.

[neon idol]

I am the only that can go on lmarena.ai without any problem?

[echo aurora]

Sorry to say that info isn't shared; however, it's understandable why that'd be helpful information so the team is aware of this request and are putting thought into it. Would note the rate limit resets ~ an hour after you hit that limit.

[mighty juniper]

What happened to the Copilot leaderboard? it's 99 days old now
Also any plan to get some 🎶 AI Music generation leaderboard at somepoint? 🤔

[crimson escarp]

Thanks

[potent glacier]

I don’t know how it was unexpected when you’re hosting all the major models

That was very well-known

[ashen plaza]

https://tenor.com/view/bbtangela-tangela-bb20-tell-me-everything-tell-me-your-secrets-gif-12997620

[languid wolf]

😫

[echo aurora]

We try our best to get ahead of potential issues as much as possible. It's not always going to be flawless, but we learn from our mistakes and always pay attention to how we can improve.

[echo aurora]

[potent glacier]

I mean it says it all right there in the mission statement…

If the site no longer aligns with that, they should change what it says otherwise that’s misleading

How are models supposed to get adequately tested if they’re getting rate limited in Battle Mode?

[topaz sun]

That sucks. I get that there’s unexpected traffic, but requiring logins feels like a pretty harsh step for what was originally a free, open experience. It comes off less like rate-limiting and more like a step toward monetization, which a lot of users won’t like. There are lighter-weight solutions you could try before forcing logins: temporary IP-based rate limiting, request queuing, etc. Those approaches keep access open without tying usage to accounts. Part of what made lmarena great early on was the frictionless experience. Adding barriers like logins doesn’t just hurt usability, it sends the message that you’re willing to trade off the community’s trust. If you go down that path, don’t be surprised when long-time users start leaving. At that point, you’re not really offering the “open experience” you built your reputation on anymore.

[potent glacier]

Sadly altruism isn’t realistic

[azure vessel]

I agree

[potent glacier]

They probably have good intentions, but they’re going to end up being like all the other companies and charging premium fees for usage

And you’ll still get rate limited even when you’re being charged 🥴

I can see rate limiting direct chat on the image gen models, but on the LLMs?

That’s pretty much a slap in the face

[violet inlet]

Very well said.

[fathom jackal]

The most frustrating thing is these generations that fail but don't give you a refund... Every AI gives you a refund for failed generations except this one... for only 3 generations this is ridiculous

[high hound]

Yeah, failed gens without refunds are rough. and with the new login wall already making things worse, it feels like the whole experience is slipping away from what it used to be.

[glossy umbra]

Lol i have to agree whit you on this one. Its changing from what it used to be even tho it built its reputation on the old system.

[austere charm]

hello

[cinder epoch]

Maaan fk the new system its some bs

This is gonna be a sheetty ai router if it continues like this

@echo aurora Listen to user reviews you guys damn it

We aint sign up for this bs

[lusty tinsel]

hopefully they would listen to what most people been complaining and find a better way... (kinda beginning to look like a cope thought sometimes)

[drifting crow]

How can anyone be angry at pineapple with that pfp

[cinder epoch]

This become paid and i zoom the fk outta here

"Lets offer our users a free experience"
"Proceeds to add a login wall thats gonna be followed by some shetty restrictions soon"

Fk you man

[lusty tinsel]

at least for the battle mode, the original purpose and how it was working should never be touched

[echo aurora]

Thank you for sharing this feedback, it is important to us to hear how we can do better. All of this is being shared with the team. We are taking this seriously.

[echo aurora]

Every AI gives you a refund for failed generations except this one
Yeah this is a really good point. If a generation fails having it count towards the rate limit doesn't make sense.

[ashen plaza]

I actually know some that don't and its stupid.

[high hound]

yeh the money gotta go somewhere after all. but i think there are options. maybe agree to sell user data if the user wants to use it for free, or have some sponsors or do something. after all u brought the communyty on free models and unbiased ratings . if it steeps away youre going too loose the community base.

[ashen plaza]

we are talking ive seen some people complain they have like 70% of their generations used up on errors and didnt get them back. now it was a text based bot not an image generator but still

[frozen aurora]

hi guys can we create 9:16 aspect ratio video?

[ashen plaza]

I dont think we have any control over the aspect ratio

[echo aurora]

Something that has always resonated with me from this blog post (would recommend a read if you haven't already) is this section:

At our core, LMArena will always provide:

• An open, accessible platform for the community to participate in evaluating and comparing models through real-world prompts.
• Transparent, science-driven leaderboard policies that ensure every model is tested fairly, consistently, and with community input.
• Features designed for the community, from a better UI/UX to more ways to engage, vote, and contribute to AI progress.
• Research to push the cutting edge of AI evaluation and reliability.
  I can assure you this will always be the case.

[tall summit]

that can all change

people make empty promises all the time

nobody knows one way or another

maybe lmarena will do better

i have faith

[tall summit]

^

[frozen aurora]

its to bad any one can give me any site for use free veo3 model or very cheap one wit selectable 9:16 aspect ratio?

[ashen plaza]

veo 3? for free? GOOD LUCK! I know of a website with veo 3, cheapest around for 5 dollars. A GENERATION! veo 3 is much too expensive atm to be running for free anywhere outside of testing purposes like in this discord. I can send you a website you can generate videos in 9:16 ratio tho. it has 10+ minute generation times and 5 free priority generations a day tho. other wise unlimited to use

[frozen aurora]

nice i dm you dude

[keen beacon]

I assume there's limits on using the higher models right?

[potent glacier]

Let’s look at the bold words here on the LMArena blog:

It should be very obvious why everyone is angry

These new rules or whatever they are entirely contradict all of that

Battle Mode should NOT be rate limited

It wasn’t before

The only thing getting rate limited was using direct chat

The old model was beyond preferable to whatever mess this new thing is

“Commercial sustainability should never come at the cost of community trust or great science”

Well, I can guarantee you, with these new rules the trust of the community is definitely plummeting

I would hate to see LMArena become another “Open”Ai

Sam Altman really can’t be trusted either

[whole sundial]

"accessible to everyone" well, not everyone wants or has a Google account, Google accounts may be banned in some regions as well. I hope they add other options so more people can make accounts, but I wonder if LMArena will be able to tell which person did a certain prompt or generation, which I imagine they can.

[formal jungle]

What's up with the people as small figurines on a desktop?

[ashen plaza]

something about a trend

[formal jungle]

Ah cool. I enjoyed that action figure trend, I suppose this is similar. Another good one might be to create you and some friends as miniatures in a tabletop war game

[flint river]

i am excited to be here, hope to learn new things here.

[formal jungle]

Welcome! 🙂

[wintry tinsel]

Flushing away good will is a full time job of the modern tech industry

[quiet abyss]

how do i get the image generator on the website to generate 16:9 images?

[jagged vortex]

hello

[latent crest]

Did u try SeeDream 4?

[little narwhal]

New mystery models on OpenRouter

Probably Gemini 3.0

[latent crest]

What’s open router ?

[zinc ore]

No lol

Its not as performant as current top models, let alone upcoming gem 3

[void trout]

hey, if I ask to generate a video of someone that looks like jar jar brinks from star wars, is that forbidden or the error happens sometimes usually?

[little narwhal]

What model would you say its ability is on par with?

[languid crescent]

Man, what's happening? What's with the rate limits? How many are these rate limits per chat?

🙁

[robust yoke]

The rate limits may be due to the fact that the models are being hosted from their respective APIs, and thus rate limits are likely to occur when using the associated model too many times.

[languid crescent]

Is it because of the traffic?

[robust yoke]

I don't think traffic would cause shortened rate limits.

The issue may be arising from excessive use of the associated model in such a short time.

[sullen quest]

2 mill context window? yeah that sounds like it could be gemini 3 dusk could be flash and sky could be pro?

nah, after testing it its just wayy too garbage and doesn't act like gemini does

[little narwhal]

So you think it'll be a smaller open-source model?

I'm praying for open source if it's not a frontier model

[sullen quest]

unless google was massivly cost cutting this wouldn't make much sense, its worse then gemma

[drifting crow]

It’s bard

[sullen quest]

lol

If I had to bet, I think its a perplexity model

[little narwhal]

Perplexity lmao

How do they even get billionaires to invest in them

There is not a single thing Perplexity does better than the frontier companies

[lofty elm]

LMArena not working on brave, keeps repeating

[polar marlin]

😭

[little narwhal]

Is sky or dusk better

[lofty elm]

??

[sullen quest]

sky is better, a little at least

[lofty elm]

after that login email update, eveything got worse

well it was fun while it lasted

[sullen quest]

mm

yeah they kinda messed up their site a bit

it was fine with slower changes

[lofty elm]

ill go back to open router for a while

[fickle dew]

sky is pretty much uncensored

[lofty elm]

what model?

is it llm

[fickle dew]

sonoma sky

I'm guessing they're some kind of XAI model

[sullen quest]

idk grok models like grok 4 a worlds ahead of sonoma sky

[lofty elm]

ohh i see

[spare rune]

People when they have limited free access to a stuff they would usually pay for:

[languid crescent]

I aint the only one

[winged mauve]

lmarena lists image-to-video arena and the veo model, but i don't see it on the website, how can i generate videos from images on the website? anyone?

[languid crescent]

on the discord itself

#video-arena-1

[drifting crow]

Perplexity: “ai”
VCs:

[winged mauve]

I am unable to choose the model

[burnt sinew]

yo chat

[robust yoke]

Greetings, Vova.

[burnt sinew]

heyy

[robust yoke]

Hello.

[burnt sinew]

anyone here used gemini cli

[robust yoke]

I have.

[burnt sinew]

do you like it

[robust yoke]

It's pretty neat.

[burnt sinew]

i know theres better ones

but i want to use it for free

[robust yoke]

That's fair.

[burnt sinew]

have you used openhands

[robust yoke]

I have not.

[burnt sinew]

oh its also cool

[robust yoke]

I see.

[burnt sinew]

its like what cursor had sort of

its in web

[robust yoke]

Ah.

I'll be sure to check it out.

[burnt sinew]

they give lots of free credits too

[robust yoke]

That's nice.

[true sierra]

Can you generate a video from an image you already have?

[burnt sinew]

yea

do /image-to-video prompt:text

[robust yoke]

You sure can.

[burnt sinew]

i think

[robust yoke]

That's correct.

[true sierra]

How?

[true sierra]

Thanks

[burnt sinew]

in #video-arena-3

[true sierra]

Do you paste the image in to the prompt?

[burnt sinew]

nah it will offer you to upload it

[robust yoke]

Well, within #video-arena-1 , if you use the following command, then you can input an image from your camera roll into the prompt, as well as the optional prompt for what you want to happen in the video:
"/image-to-video (image) (prompt)".

[burnt sinew]

@minor fern

smh

[jade egret]

gemini 3 when

);

[burnt sinew]

3 months

its just a guess

actually let me ask gemini when it will come out

[winged mauve]

Can i choose the model for video generation?

[jade egret]

so long

[burnt sinew]

its the best free model rn still

[burnt sinew]

December 10, 2025 it said

[jade egret]

2.5 pro?

[jade egret]

o

[burnt sinew]

yes

[jade egret]

2.5 pro is kinda old ngl

[burnt sinew]

yes its been here since march

[jade egret]

yea

[burnt sinew]

half a year of 2.5!

[jade egret]

how good are you expecting 3.0 be?

[burnt sinew]

idk, if it aligns with the jump from 2.0 to 2.5

then it will be insane

but i doubt it will

[jade egret]

dang

[burnt sinew]

OR if it will be they will make it paid, or other stuff

[jade egret]

i think hopefully they will make flash free

[burnt sinew]

hopefully they keep stuff like it is now

[jade egret]

yea

[burnt sinew]

normal people using the normal gemini website will keep paying

and i will keep having unlimited on ai studio

[jade egret]

but is it gonna be only ultra for like a few weeks or months 🙁

[burnt sinew]

hope not

[jade egret]

yea

[burnt sinew]

like veo 3 its locked behind money 🙁

[jade egret]

🙁

[burnt sinew]

at least they made nano banana unlimited free

[jade egret]

yea

it pretty good

[burnt sinew]

it is REALLY good

[jade egret]

yea

and google anti trust case win

it looking good for google?

[burnt sinew]

i think so

[jade egret]

:0

[burnt sinew]

so far

leading in basically everything rn i think

except coding

[jade egret]

yea

what base model they usuing tho 🤔

[burnt sinew]

wdym

[jade egret]

for 3.0 do they just work on 2.5

[burnt sinew]

wdym

OH

[jade egret]

nvm im dumb lol (:

[burnt sinew]

obviously

i doubt they start from scratch

[jade egret]

oh

yea

anyways gtg

cya

[burnt sinew]

bye bye

[jade egret]

bye

[burnt sinew]

and it said February 25, 2026 avaible in ai studio

i wonder how right it will be tbh

[flat flax]

Hey, came across this Twitter thread summarizing OpenAI’s new paper on LLM hallucinations:
https://x.com/LuozhuZhang/status/1964209351960514778

They highlight that hallucinations aren’t inevitable if you change incentives (e.g. penalize confident errors more than abstentions, reward calibrated uncertainty).

Has anyone here read the paper? Seems like this approach could significantly reduce hallucination rates. Any dataset recommendations to test this kind of “abstain vs. wrong” behavior?

Luozhu (@LuozhuZhang)

This is an excellent study on LLM hallucinations. My takeaways:

1. Treat neural nets less like rigid programs and more like statistical organisms.
From this lens, LLMs behave a bit like the brain: they fill in missing pieces by generalizing patterns, which makes them powerful

X · Sep 6, 2025, 6:09 AM

[burnt sinew]

@sly isle

[earnest rover]

@echo aurora
What's the rate limit
And remember it will be not good setting a rate limit

[thick patio]

is it possible to get desired image size output in Lmarena?

[robust yoke]

No, all of that is random.

[drifting crow]

I’m gonna ask a llm to summarise it for me

Looks interesting tho

[flat flax]

Thank you. I found this Twitter post really well-written.
it resonated with me a lot and I feel I got a lot out of it.

[keen beacon]

Hello LMArena!

[robust yoke]

Greetings, Peter.

[winged mauve]

Can anyone tell me if LMArena gonna be paid or free as it is rn in near future? @robust yoke

[robust yoke]

Well, considering it's a testing benchmark for popular AI models, I don't really see a reason as to why it would be a paid service any time soon.

[winged mauve]

thanks sir, appreciate u replying 😊

[robust yoke]

It's my pleasure. I always try to reply with the most factual information I can.

[winged mauve]

one more doubt, are we gonna see video generation on the website itself in upcoming days?? or will it be limited to discord forever? and will we be ever able to choose the model? just like image generation on the website

[robust yoke]

For the time being, video generation is limited to the Discord server. However, people have requested for video generation to be part of the actual benchmark website itself. In the near future, we are likely to see a video generation feature where you can directly generate videos with popular video generation models like Veo 3, just like how you can generate images on the website with popular image generation models like Imagen.

[winged mauve]

thanks again 😀

[robust yoke]

My pleasure.

[drifting crow]

ohyh twitter summary is good

[flat flax]

I’ve looked through his earlier comments, this man’s thoughts are quite interesting.

[drifting crow]

I can’t see them coz I don’t have twitter account

[unborn ocean]

Funnily, OpenAI is actually the company not implementing their own research ideas from the paper…

I think this shows that anthropic is already aware of it and addressing it.

[robust yoke]

W Anthropic.

[dense sphinx]

So who win?

[robust yoke]

Claude, I'd believe.

[dense sphinx]

Claude sonnet rules.

👍

[robust yoke]

True.

[flat flax]

Oh where is this comes from?

[flat flax]

Who's blog, who wins

But Claude is pretty good though

[kindred pulsar]

How to generate 16:9 images in nano banana

[robust yoke]

That sort of stuff gets randomly decided by nano-banana.

It's not directly adjustable.

[olive knot]

Hi, im new here. Nice to meet all

[robust yoke]

Nice to meet you too, Bimmo.

[keen beacon]

I make a post that we lack valuable insight into how LLMs function because we don't have good benchmarks
Suggest how to improve benchmarks
Nobody cares
OpenAI drops a paper literally with the same conclusion
Everybody and their mama loses their crap

https://tenor.com/view/sayaka-madoka-magica-gif-10367661

[royal pond]

hello

greetings from Borneo

[robust yoke]

Greetings, Tansri.

[whole wagon]

It's an obvious thing, no idea what takes them so long to discover these. All the money and talent going into this and they didn't think of that simple idea to reduce hallucinations, I wonder how much other easy things they miss

[robust yoke]

Let me share my perspective on this. While it may seem straightforward in hindsight, research breakthroughs often appear obvious only after they've been discovered and validated. The challenge isn't just identifying potential solutions, but rigorously testing them, understanding their tradeoffs, and implementing them effectively at scale.

The idea of penalizing confident errors more heavily than abstentions isn't entirely new, it builds on established concepts in machine learning about uncertainty calibration. However, systematically applying this to reduce LLM hallucinations while maintaining model utility requires careful experimental design and validation.

I think it's worth considering that AI labs are often exploring many promising approaches simultaneously. What might seem like an "obvious" missed opportunity could be something they've investigated but found challenging to implement effectively, or that had unexpected downsides that weren't apparent at first glance.

That said, I do agree that sometimes simple yet powerful ideas can be overlooked, especially when teams are focused on more complex approaches. This is one reason why having diverse perspectives and open research discussions in the AI community is so valuable.

What are your thoughts on other "simple" approaches that might help improve LLM reliability?

[whole wagon]

Are you capable of rational thought rather than just pasting LLM output verbatim?

[robust yoke]

Just because it's long and detailed doesn't mean it's an LLM output.

And even if it was, who cares? We're in a server where AI is pretty much normalized, even for creating images and videos.

Therefore, even if my response were generated by an LLM, there wouldn't be any reason to get angry since we're already in a server where that pretty much happens 24/7.

[whole wagon]

It is devoid of any meaningful information or point

[robust yoke]

Just like how many statements tend to be.

[unborn ocean]

Actually from OpenAI 💀:

https://openai.com/index/openai-anthropic-safety-evaluation/

[unborn ocean]

I agree tough, many ideas proposed and implemented are fundamentally incredibly simple (not all of them, but quite a lot).
The hard part is usually scaling and really optimising and making sure they are the right ones though.

[whole wagon]

I still don't think it's that hard ngl

[unborn ocean]

E.g. moe is the most straightforward thing ever, yet it took quite a bit for everyone to adapt it, because while the theory might be simple, implementing it is quite difficultly and requires some experimentation

[whole wagon]

This idea is like reducing hallucinations to half or smth. It should be very visible even with small models

[unborn ocean]

The thing is that stuff you try is small models / experimental papers always ends up being more complicated in bigger ones. That is basically what I am referring to.

[whole wagon]

I'm aware, I still don't think it's that hard. I've scaled ideas like this before it is simply some adjustments made the core idea is the same

Just because you have a big model does not really change the dynamics of optimal data in most cases

Good data for small and large models remains largely similar

I would assume it has been published because they have realised themselves it is an obvious thing

So no point keeping it secret

[trail wagon]

What's up?

[junior jewel]

Hey guys

[spare rune]

thats why i may have said "people"

[obtuse heart]

google has been cooking with their releases, of course they will cook with gemini 3

[latent crest]

Is there a way to pin texts I send ?? So I’ll see the replies later that day?

[manic raft]

hello

[ocean vortex]

This is a difficult balance to get right to be fair, especially with the way current evals are being done. Making it 'refuse' will also inadvertently make it refuse answering what it otherwise would have answered correct.

So it all comes down at what cost are you prepared to tackle hallucinations with, and whether sacrificing overall performance is ok

[ornate agate]

if you're going for max performance/score on benchmarks that don't penalize wrong answer, making the model hyper confidently yolo everything is probably optimal, for that.

I do still think "hallucinations" are a fundamental problem with llms though.

[ocean vortex]

This was for o3, and they did tackle this problem with gpt5 specifically. Though probably not to the extreme levels

https://www.reddit.com/r/singularity/comments/1mm51m5/gpt5_admits_it_doesnt_know_an_answer/

[unborn ocean]

Ik, but I have yet to actually see a review that tried to quantify this o3 vs 5.
They also talked about using a classifier that tries to detect hallucinations as a reward signal throughout the post training phase (of gpt 5).

And I also personally noticed the change. + much of their system card report was focused on reduced hallucinations.

[pseudo glen]

hellooooooo

[elfin harbor]

Helloo

[ocean vortex]

o3 said it doesn't know the answer in 0% of responses
gpt5 did it in 5% of responses (100-(40+55))

small steps

[elfin harbor]

I have a slight question about the rate limit thing? Is this temporary or permanent?

[glossy umbra]

react this message if u want less restrictions

[glossy umbra]

^

[elfin harbor]

Hmm I do agree but I also get it tbh the increase of traffic is expected. I do get the rate limit thing and login thing. But I hoped it would be a temporary thing till a workaround that would be great.

[digital umbra]

it's really low now

only 3 images or so

[elfin harbor]

Oooh I see do you know the break time after that?

[digital umbra]

no

[elfin harbor]

hmm

there was one message that told me to wait 48min but I didn't see any message after that

[glossy umbra]

see? even that isn’t clear. with llms you get “use again after 50 minutes”. with image you don’t receive anything. seems censored to me

[elfin harbor]

I did get message of wait 48min but one time only

I guess waiting is good option for now

[glossy umbra]

i guess. but so much restriction affects user experience.

[potent glacier]

They said they're 'taking feedback' from what we've been saying

It remains to be seen exactly what they do

[tall tulip]

Can't upload image?

Down again?

Oh nvm

[topaz flint]

It seems like

[digital umbra]

i think lmarena will have a paid tier by the end of the year if it continues like this

but of course one can't really complain about it

[potent glacier]

That does NOT align with their mission statement. At all.

[digital umbra]

i'm surprised it has been free for so long

[verbal nimbus]

Funny they didn't include GPT-4.5, no doubt the best model on SimpleQA

[potent glacier]

As per their statement: "LMArena will stay open and accessible to everyone. To do that sustainably, we’re focused on creating long-term value through services that benefit the entire AI ecosystem and serve the larger community."
"At our core, LMArena will always provide:

An open, accessible platform for the community to participate in evaluating and comparing models through real-world prompts."

Limiting people who aren't logged into Google isn't only ridiculous, but just plain not right

[verbal nimbus]

In third party papers, GPT 4.5 scored ~25% higher

[digital umbra]

they raised $100m from investors recently on a valuation of $600m. where does that come from? obviously one part of it is selling the datasets of prompts and responses, but clearly that's not enough anymore.

and yes, nowhere in that mission statement does it say it has to be free

[potent glacier]

I don't care if it's trained on our prompts

Hell, every ai thing is trained on the internet

Things are scraped from all over the place

[trail creek]

Also don't understand how loging in will help the ai ecosystem more??

[potent glacier]

So they can track your usage? Idk...

[verbal nimbus]

It's not mandatory, is it?

[potent glacier]

I think that might be the case

[verbal nimbus]

It was just a requested feature

[trail creek]

the site is for benchmarking not data collecting...And also it has been free and unlimited without any limit for 3 years what changed now exactly?

[digital umbra]

it's mandatory if you want to use the site for more than 10 minutes now lol

but yes, it was originally requested because of the history being wiped every once in a while

[verbal nimbus]

I've been using it for more than 10 mins today, seems fine

I'm not logged in

[trail creek]

I feel like after the damn banana stealth model they got popular and so with that they saw how much money they could make

pure greed...Theres literally nothing else behind this updates but GREED

[digital umbra]

well i got this after 3 prompts today

[ocean vortex]

We can calculate refusal rate for all of those by simply substracting. 4.5 has next to none refusal rate, but interestingly enough 4-turbo seems to be above 5.

[potent glacier]

Logged in users get 'higher rate limits'

[verbal nimbus]

Odd, how many convos have you had in the history? Maybe it's because I've already had lots of convos, so it doesn't flag it.

[digital umbra]

the VC money came long before nano-banana

[ocean vortex]

7% refusal rate for 4-Turbo

[digital umbra]

no it happens both on new and old convos

[verbal nimbus]

Privacy issues side, it's probably because it has better verifiability

[elfin harbor]

does the limit affects chats too or image gen

[potent glacier]

They said it affects Battle Mode as well!!!

And direct chat

[elfin harbor]

Oh well

[trail creek]

If they just not force the login and give the same limit as login in users as anonymous users i would not have any problem (I WOULD)

but they added the rate limit to the battle mode tooo which is the whole purpose of the damn site???? Im i wrong?

[potent glacier]

Exactly.

Enough people need to voice their concerns about it so it can get resolved

[verbal nimbus]

Their paper showed that you need a certain number of conversations in order for adversarial (I forgot the term) vote detection to work

[potent glacier]

Being compalcent won't solve anything

[glossy umbra]

react here if u want less restrictions #general message

[digital umbra]

but yes, based on what i saw with basically every local newspaper in my area, the pathway to monetization is to first force everyone to use an account, and then put up the paywall after a while

[trail creek]

jesus they making me hate capitalism

[potent glacier]

https://tenor.com/view/first-time-meme-movie-james-cameron-western-gif-17538858

[verbal nimbus]

I don't think it's monetization, it's hard to detect vote manipulation for accounts with only a few votes

[trail creek]

everything ...Every fricking thing in this world have to be ruined by money, i mean how much money is too much money

[glossy umbra]

you guys react here to show your opposition to the new update #general message

[verbal nimbus]

They had a paper on it last year

[potent glacier]

This'll get drowned out in the sea of text

[ocean vortex]

Also o1 seems to do very well for it at 14%. That's the main reason it doesn't hallucinate more overall than gpt5 despite lower accuracy

[verbal nimbus]

I personally haven't encountered it yet. Were you using battle or direct mode?

[digital umbra]

battle mode

[verbal nimbus]

Interesting, maybe it's more of a rate limit than an absolute limit

[potent glacier]

The whole point of Battle Mode was to be free and unlimited to compare models. They really s h i t the bed with this one.

[trail creek]

it is the path to monetization...Also now that they have your complete data, be sure that they will sell it to ai companies for bonus benefits on top of the payment they get for putting a model in their site.. Or even worse

[glossy umbra]

react here to vote to have less restrictions #general message

[potent glacier]

Stop

[verbal nimbus]

They said non-Google logins are planned... the data is supposed to be open source btw

Like the data from May or smth is on Hugging Face

[potent glacier]

I'll be completely honest. They're a big company. Don't trust big companies. How do I know? I work for one myself, sadly...

[glossy umbra]

it’s clear people don’t want this new restriction system

[digital umbra]

https://www.prnewswire.com/news-releases/lmarena-secures-100m-in-seed-funding-to-bring-scientific-rigor-to-ai-reliability-302462025.html for those that don't know btw

[trail creek]

sorry to break it to you, but i swear to god that they will not change anything, they do not care, and they don't read these messages... im just talking so i can cope but at the end nothing changes

[digital umbra]

this was in may

[potent glacier]

But this isn't social media. Liking your random thing won't make them change anything.

[verbal nimbus]

Interesting, what website is this?

[glossy umbra]

brings awareness

[glossy umbra]

yea they do care their user base was built on top of the old system. if they change it they know nobody will use their service anymore

[potent glacier]

It almost beggars on farming likes

I could care less about likes or upvotes or w/e

[glossy umbra]

no it’s not like that . just express your unlikings

[potent glacier]

Oh trust me I have

A lot of us have

At the end of the day what'll it get us

They're a company and they're going to do what's in their best interest

Whether that aligns with the mission statement or not

Companies lie. People lie.

That's just the way of the world

[trail creek]

the data they will have access to when they have you google account is a different very different and superior thing than just the data they get from the prompts you are putting on the sites.

[glossy umbra]

it’s not their interest to lose a majority of their user base. they know if they turn it into an ai router then many people will leave

what’s the difference between using lm arena and open router then

[potent glacier]

I have never used open router or heard of it

[glossy umbra]

it is an ai router

[digital umbra]

once the investor money has been deposited, giving them a return on the investment is more important than any mission statement

[verbal nimbus]

On this third party benchmark, GPT-5-Minimal scores worse than GPT-4o: https://github.com/vectara/hallucination-leaderboard

GPT-4.5 beats both

[glossy umbra]

you can’t give them a return on investment if the user base will leave

[potent glacier]

Yes, lying blindly to people. Thank you for proving my point.

You really wanna know how to get results?

Stop using their site entirely

[glossy umbra]

many people will do that if it becomes paid

[verbal nimbus]

What permissions did they ask for? It depends what scopes are granted. If only the OPENID, then all the site gets is a token.

[potent glacier]

If a huge number of people stopped using it, they wouldn't be getting their data for new models

[glossy umbra]

they’ll stop when it’s paid

[potent glacier]

There was that guy a few days ago that brought up payment in their TOS

Let me remind everyone

[verbal nimbus]

It needs email and name scopes to access your name or email

I haven't logged in so idk what permissions it asks for

[potent glacier]

This is what it says in section 8: “FEES AND PURCHASE TERMS. Company currently offers the Service free of charge. However, we retain the right to charge for the Service, or any features or components thereof.”

[trail creek]

their user base now is much larger than it was in the days of the legacy site, now they are so large that the thousand or so people who are in this discord, are just a friction of the actual people who uses the site which is now in the millions.

[glossy umbra]

yup

[potent glacier]

However, we retain the right to charge for the Service, or any features or components thereof.

[trail creek]

i hope im wrong but i don't think im

[glossy umbra]

holds the right to make it paid

[digital umbra]

maybe stealth models will still be free or with higher rate limits, since in that case the provider will pay for the comparison report. just like openrouter

[glossy umbra]

we can’t know unless the site traffic is measured . take a look at the discord server members, we can assume that’s about 5-10% of all users

[potent glacier]

They're gonna have a dumpster fire on their hands soon

I mean they closed the whole chat entirely when the site wasn't working

I thought that was kinda sus to begin with

[verbal nimbus]

I think the allowance nowadays is more generous

[potent glacier]

3 in battle mode? That's generous?

[verbal nimbus]

You didn't use to be able to talk to Claude directly without hitting quotas in like 10 msgs

[glossy umbra]

the nail in the coffin is the fact that they say “the service is free and we’ll continue to have it be free” but they have a rule that says “we can make it non free at any moment”. should be self explanatory

[trail creek]

i agree which in ratio to the profit they would make by keeping this system. is not a bad sacrafice of the 10%

[verbal nimbus]

Or Opus, which is very expensive

[verbal nimbus]

I've had more than 3 battles today, so idk

[potent glacier]

Like I said: Companies can’t be trusted

[ocean vortex]

https://openai.com/safety/evaluations-hub/#hallucination-evaluations

[potent glacier]

This is the whole reason why I am so glad that I am able to gen locally

F censorship, guardrails, corporate handholding, etc..

[verbal nimbus]

Cheers, I linked a third party one above too

[potent glacier]

It’s pretty sad that the only person that we get feedback from is @echo aurora

There hasn’t been any transparency about any of this

[steady rover]

how long does t take to gen a vid caause i have been waiting a hr now

[verbal nimbus]

For ppl who want anonymity, are you clearing your browser history? Otherwise it's not really anonymous, and prompts are logged...

[glossy umbra]

?? if you clear the history the chats are still on their database

[potent glacier]

We keep getting told by @echo aurora that they can’t talk about the rate limit or how much it is

That isn’t transparent at all

If you implement new rules, people should be kept afloat instead of having to play the guessing game

[steady rover]

?

[verbal nimbus]

Yeah, and your userID (it whatever is used to track) would be same across all conversations if you don't clear your cookies iirc

[potent glacier]

Your prompts are already in the system lol Once you type them they’re logged forever

[verbal nimbus]

What permissions do they request from Google?

[potent glacier]

Doesn’t matter if you delete your history

[verbal nimbus]

Yeah exactly

[potent glacier]

Data collection?

Honestly I don’t care. I just get ads targeted at me.

[verbal nimbus]

I mean, does it ask for name, email, etc? Google will tell you the first time you sign up

[verbal nimbus]

I don't think oauth does that

[potent glacier]

I meant about the data collection

[trail creek]

I use tor, clearing history isn't anonymous enough for me

[potent glacier]

Your prompts won’t ever vanish, though

[verbal nimbus]

Yeah, that's the best way for anonymity

[potent glacier]

We’re training models every single day

Our prompts help the models grow

[steady rover]

how long does t take to gen a vid caause i have been waiting a hr now

[glossy umbra]

Lol. This is a classic “technical requirement meets strategic opportunity”. The need of creating accounts to stop the bots is a mask to a restricted system that’ll be followed by paywalls. Not that hard to see what’s going on

[verbal nimbus]

Although if rate limits are an issue, I don't see why can't just clear session and establish a new Tor circuit

It should reset the limits, although I'm pretty sure that's against the policy

[potent glacier]

You can’t use a VPN because it bans you immediately

[trail creek]

i don't care for my prompts because i don't use arena for actual work, i just use it to see what model is best for certain prompts.

Also im ok if the prompts are actually used for training, i want ai models to get better

[potent glacier]

Until you turn it off

[elfin beacon]

hello! Any talk in Spanish?

[verbal nimbus]

I use a VPN, captcha is more annoying but I don't get banned

[steady rover]

how long does t take to gen a vid caause i have been waiting a hr now?

[potent glacier]

We’re only supposed to speak English here

People will find ways to circumvent the rate limits but it won’t work

I mean 3 in Battle Mode is downright insulting

LMArena: “We have made the difficult decision to go Pay For Play.”

[trail creek]

you can just use zen for that, but im really lazy to keep creating new containers every time i want to test a prompt

[verbal nimbus]

In Brave it's just one button

[glossy umbra]

They say it’ll stay free, but the rule that they can flip to paid anytime and their silence on it shouldn’t be ignored. Even if we’re not a massive group, staying vocal shows them that losing core users isn’t worth the risk.

[digital umbra]

free user loyalty doesn't really matter tbh

[verbal nimbus]

Honestly they're offering really expensive models for free, so it's fine imo

Like it wasn't possible to talk directly to such expensive models last year

[potent glacier]

@trail creek @verbal nimbus I’d delete those last chats if I were you. You’re essentially talking about overriding the site limits 😉

[trail creek]

They will not make it paid, but i think they are taking the yupp rout which is the smartet way to keep something "free"

[digital umbra]

user loyalty would have been a concern if they were actually paying for it

[trail creek]

if they make it paid then just use open router

[verbal nimbus]

Hypothetically speaking of course... I want all my chats to be stored 🤣

[potent glacier]

How is open router different?

[trail creek]

they really don't read this i swear

[potent glacier]

Pineapple does

They’ve deleted a lot of things

[digital umbra]

and if too many free users leave the site, the rate limits will have to go up so that the paid datasets remain useful

[potent glacier]

So do other mods

[digital umbra]

so that's kind of self stabilizing i'd say

[glossy umbra]

Sure, the models are expensive, but the value of LMArena isn’t just “free compute”, it’s the community that built the platform. If they cut the free user base down to scraps, they lose the very people whose input and trust make the site credible in the first place.

[trail creek]

i mean if arena turns to be paid, i think open router is the better option, more polished ui and free offers rom time to time.

[potent glacier]

Again: Putting trust in a company is a fool’s game

There’s no such thing as an altruistic company

They all have ulterior motives

[glossy umbra]

Exactly. and that’s why the only leverage we have is to make it clear that if they break the free, open experience, the community walks.

[potent glacier]

What, all three of us?

Yeah that’ll really teach ‘em!

[trail creek]

yes.

[potent glacier]

You'd need to have a massive amount leave

[glossy umbra]

Doesn’t matter if it’s three or three hundred if the core users leave, the platform loses the very people who give it value. We’re only 3 people here. You don’t know the lots that didn’t join the discord server

[potent glacier]

Most of them don't speak English and have no idea what we're even saying

If they did we wouldn't be getting spammed with /image

[glossy umbra]

Maybe, but the admins definitely see it when the vocal part of the community speaks up , that’s why they already reacted to just a handful of complaints.

[potent glacier]

These complaints about the rate limits and stuff definitely go above @echo aurora and the other mods here

If it was a corporate decision, it's gonna be an uphill battle

[glossy umbra]

I know , but pineapple told us he would redirect our opinions to the devs. I’m not saying it goes past an internal chat. But if pineapple really is a man of his word then the devs also know it shouldn’t be a good idea to impose such restrictions

[potent glacier]

Unless they got that sweet, sweet whiff of money

Money makes the world go 'round

[glossy umbra]

We don’t know.

[trail creek]

we've complained to give back the legacy site and they didn't...I cant think of an instance when they actually listened to anything, they just give a shallow excuses that they got from gpt and move on.

[potent glacier]

mmm..I'm willing to bet differently.

[glossy umbra]

Yes you have been complaining but you think they’d actually listen? You need to take action and stop using the service

Many people will stop when it becomes paid

[potent glacier]

Hasn't happened yet, boyo

[neon idol]

How can I try seedream v4?

[trail creek]

i didn't use it for the pat two days. But i don't think 10 people not using it is gonna be of any cost to them

[potent glacier]

Actually I made a mistake about earlier. I can think of one company that's altruistic and has continuously given people free things for years now: Hello Games

[glossy umbra]

Again. 10 users here. off discord are thousands

[potent glacier]

They have continuously given free updates to No Man's Sky while other companies would charge

[trail creek]

and there's some people here that don't even know what we are talking about and login to any site they enter without any care for security.

[glossy umbra]

So a gem to say.

[glossy umbra]

Listen, even if free users wanted to use the service anymore, they couldn’t

when it does become paid

[potent glacier]

But they're most likely not even native English speakers so they don't know what's going on

[glossy umbra]

Yes but they won’t be able to use the service anymore when the time comes

[verbal nimbus]

Looks like rate limits are only for image gen as per #announcements ? That kinda explains why I haven't encountered it.

[trail creek]

those people don't know what boycott is

[glossy umbra]

That is true

[verbal nimbus]

They're having compute issues

[glossy umbra]

Boycott will not be a thing. It will be “i can’t use the service”, not “ i don’t want to”

[potent glacier]

They're not a poor company

They have very generous benefactors and donors

All the major ai companies are represented on LMArena

They're all probably funneling money to them as well to continue training

[trail creek]

they don't actually make models, they just get API from companies.

[potent glacier]

People actually bet ACTUAL money on how models perform

I forget the site

[trail creek]

so how are they having compute problems?

[verbal nimbus]

It could be bandwidth or storage, we don't really know. They're probably saving the outputs.

Compared to text, images and videos are probably massive

[potent glacier]

I believe that stuff gets sent to the respective companies

[verbal nimbus]

They log the data too I think, otherwise votes can't really be verified in the future

[ocean vortex]

I doubt they are saving everything though. When you generate it it is kept on the originating model's server for awhile lol

[potent glacier]

All I know is that we aren't getting any transparency about what's happening or why

[verbal nimbus]

We know they're saving text outputs, I don't see why they wouldn't save images