general

[bright temple]

& i don't have a vpn on

like wtfffff i can't even log in

[hard plume]

https://tenor.com/view/captcha-lumi-madara-pro-ban-gif-25000717

Hate this capta

[wicked sage]

hello! :D

[torn mantle]

you can use ZWSP injection to bypass it, ive already implemented that in my code to reduce false positives but if its annoying you can try that already

[torn mantle]

disable adblocker

it can block some cookies that are necessary

which contains the auth session

if your adblocker cant differentiate between essential cookies and analytics cookies then you probably should replace it

it just screams = im a dumb adblocker, i will block anything i see

but again it depends on your adblocker settings too

also you are using two

also you need to switch to another browser if you really care about privacy

it really doesnt matter if you have adblocker or not if you are using chrome

telemetry/analytics are being sent anyway and with their introduction of manifest v3 it literally killed most of adblockers efficiency

[surreal creek]

petition to rename #general to #complaining

😂

[jolly garden]

Why should we use ⁠General to remind ourselves how beautiful and functional it once was? The site is currently unusable. If you prefer, we can use it for alternatives to Lmareana.

[icy burrow]

I was completely taken aback by how kind and respectful the moderators are on the server. I truly enjoy how much they are able to explain things when I need them, and they never take a step that shouldn't be taken. I am really happy to finally be part of a discord server where things are fair. Thank you to amazing dedication staff members. Happy New Year's as well

[echo aurora]

Hey @jolly garden I am sorry you're having problems with captcha right now. We did recently make a change that we are monitering closely along with community feedback. We are asking for those that get this issue to please share the Eval ID for the chat sessions in this thread: #1451574502369656842 .

I am going to ask though that we keep feedback a bit more respectful. We do want to hear from the community when things aren't working, but do ask it's being done in a helpful way.

[icy burrow]

I completely agree, I think he should be punished for his behaviour in the server. LMArena is being offered completely for free, and he has no right to complain about what is given to him. There is a saying, and I almost forgot it, but beggars cannot be choosers. Thank you again for your dedication staff members, I truly respect the work that you do, especially considering the lack of compensation that most of these positions have.

[golden ocean]

plus if you REALLY care about privacy
uninstall urban vpn extension(it got deleted from chrome webstore a week ago or something because they logged all ur ai chats on websites lmaooo)

@bright temple

[jolly garden]

@echo auroraDon't take this personally, but whoever chooses to make the changes you're referring to clearly has no respect for LMarena users. What you did by choosing the Google CAPTCHA is proof of that. Out of respect, since you're deleting my messages, I'm speechless.

[golden ocean]

fr, choosing google cpatcha must become illegal

[icy burrow]

you should have no right to speak this way. If it were my decision you would be gone already. Why would you speak of censorship given the vile things that come out of your keyboard

[jolly garden]

It's a good thing no one asked your opinion

[echo aurora]

No one is being punished.

[echo aurora]

This isn't helpful, please stop.

[icy burrow]

I'm glad the server staff are taking appropriate action given the situation. It's not my place to intervene, so I'll leave you to it

[jolly garden]

I'll stop here, I just ask those who have to carry our messages to the developers to report the users' discontent.

[echo aurora]

Everyone that works at LMArena care about the users. Getting the captcha system working that's best for the company and best for the users is something we're striving for,

[golden ocean]

lmaooooo

[echo aurora]

Will want to remind, this is the best way to report the problem in #1451574502369656842

I would like to keep this channel open for general conversation about AI.

[jagged crest]

help

@desert abyss

[golden ocean]

https://cdn.discordapp.com/attachments/1339675158201503884/1363950454379384904/bounce.gif

[echo aurora]

Can you clear your cache/cookies and hard refresh the site?

[jagged crest]

wait

[green kindle]

[jagged crest]

@echo aurora its working but i cant login

[fierce kelp]

#1343291835845578853

[echo aurora]

are you logging in, then being sent back to the home page but not logged in?

[jagged crest]

yes yes

[echo aurora]

Yeah known bug sorry to say #1451836386293448725

[desert abyss]

Hello! @cobalt nimbus Please check ⁠how-to-video-bot to learn how to generate videos.

[icy burrow]

Happy New Year's to the lovely staff in this server

[lofty frigate]

Bro why do i keep getting this captcha thing everytime i retry now its annoying

It didnt happen yesterday

[calm lagoon]

I don't have it. That's strange

[lofty frigate]

So lucky

[meager harbor]

so 1500 score wasn't reached in 2025 in lm arena

how much will the sota model reach end of 2026 ?

i'm sure it will not be over 1600

iirc max score end of 2024 was 1407. so 83 elo diff between sota end of 2024 and sota end of 2025

things are getting slower

[queen veldt]

Oh it's 31 December today I didn't even realize

[weak crane]

hello

[topaz bay]

Lmao "3x usage of the claude pro plan"

Anyone who's used GLM for coding?

I've heard it does little mistakes n stuff, and that it's pretty good, especially considering price

[weak crane]

yeah

[topaz bay]

is it coming to an ide

🙏

[fickle venture]

Welcome to Lmarena

[fickle venture]

What IDE?

[proud bobcat]

its so peak

[hollow ivy]

"There is no way in hell that ..".. GPT-5.2 is on par with Gemini 3 pro.
-# (butchering LeCun's nice sentence for this one ^^)

[proud bobcat]

Well these are all the final scores after tallying all the benchmarks together

And 5.2 is defo benchmaxxed

[ocean vortex]

I understand why OpenAI benchmaxxed. But I still don't like this 🧐

[proud bobcat]

I think we need unbenchmaxxable benchmarks

I’m really thinking of making a roleplay benchmark of my own

[ocean vortex]

not very xhigh when you look at this image

[proud bobcat]

Wait so it didn’t even reason?

Oh wait it did

Oh

[ocean vortex]

It kinda didn't. Either because of contamination or xhigh literally just works in extreme edge cases

I mean relative to earlier models

reasoning is not longer

but yeah it did reason in general lol

[topaz bay]

idk

[ocean vortex]

And when it's overfitted/contaminated it is unlikely to spend excessive time with the filler already knowing the answer

[weary galleon]

Claude > g3p > g3f > grok > gpt is the only correct answer. All other ones are WRONG!

[topaz bay]

hmmmm

kimi > zlm > g3p > grok

grok 4.1 is goated

benchmarks are sleeping on it

[weary galleon]

Kimi and ZLM are trained on Claude and Gemini outputs.

That's why their outputs are similar.

[opaque maple]

pls update it that we can upload files😭🙏

[meager harbor]

since claude is very expensive and gemini 3 pro is absically free on ai studio, answer is easy

[hollow ivy]

In this poll, "better than" means pure performance/success rate, without taking into account these factors: price, speed, token usage, availability and energy cost

[proud bobcat]

I am so fr

It is probably a modified version of O3

O3 trained on new data

[ocean vortex]

it isn't

gpt5 is a new pretrain

similar size for all of them though. gpt4o, o1, o3, gpt5...

[meager harbor]

if we cannot use the models or so litlle how can we judge ?

[proud bobcat]

There’s no way the price is that much higher on the old architecture

And the hallucinations

It leads me to believe it’s two different architectures

[hollow ivy]

All models are usable - via battlemode.
-# (if you are smart, patient and methodical)

[wind shore]

Lmao AI is not ending the human race.

[obsidian cargo]

AI will end the human race because we'll stop being humans

[novel crater]

):

[queen veldt]

Who's bro lying to

[astral bloom]

Same prompt
NBP vs gpt 1.5
opinions!

[proud bobcat]

It’s only because of Grok Code Fast

Which

WHY HASNT HE MADE GROK CODE FAST 2

It’s infinite money

Literally

[queen veldt]

Nah gemini 3 is creeping

[green yacht]

real 🙏

[dusky pier]

Because it's slop

[proud bobcat]

Grok code fast is NOT slop

[dusky pier]

Sure

[proud bobcat]

It is the most successful model to date

Quite literally

[dusky pier]

On creating bugs maybe

[proud bobcat]

That’s almost every ai model

[dusky pier]

Never use grok for coding

[proud bobcat]

If people continuously use it it has to be doing something good

https://tenor.com/view/lovestayfile-stray-kids-jeongin-skz-gif-9500357924908482314

Ew

[dusky pier]

.

[green yacht]

yeah but its pretty affordable, preformance is eh but debugging was pretty good in my experience

and as you can tell from the name it is really fast

[proud bobcat]

Whatever you say Chara from the hit indie game Undertale

[dusky pier]

Chara one shots Deltarune btw

[proud bobcat]

Roaring knight when they see a 7 year old kid

https://tenor.com/view/взгляд-2000-ярдов-война-war-soldier-gif-3632617944134077161

(They’re so cooked)

[dusky pier]

I wish there was Chara in Deltarune

I miss her

[proud bobcat]

CHARA IS A GIRL???

WHAT

[dusky pier]

I've been a fan of the game for over 7 years, and I always saw Chara as a girl, so

[loud leaf]

anyone still getting nebulaphase in arena?

[opaque maple]

srry

why does everyone like Grok?

[proud bobcat]

I thought they were nonbinary

Idk

[proud bobcat]

It’s the exact opposite

Everyone hates grok

[dusky pier]

Canonically androgynous

[proud bobcat]

I see

[opaque maple]

oh good

everyone loves Gemini

[proud bobcat]

I think grok is fine

It’s the budget king of American models at the moment

And delivers pretty good performance

Only downside is

Elon Musk

[obtuse smelt]

Elon musk should be president after donald trump ?

[queen veldt]

Broken clock is right twice a day ahh

[dusky pier]

We don't need a Jewish rightist president

That's it

[old garden]

Veo 3.1 fast

[inland quest]

Hey guys

[gaunt spade]

peak

[proud bobcat]

Literally

[thorny schooner]

Does anyone know any actual long-term fixes for the verification because this thing is killing me if not does anyone know any good alternatives you know to this

[dusky ravine]

@gaunt spade epste..

[gaunt spade]

happy christmas

[proud bobcat]

I love how the only thing saving gpt in the market share (openrouter) is OSS 120b

WAIT WHAT

WHAT

DEEPSEEK BETS XAI AND ANTHROPIC????

[gaunt spade]

rate my new pfp

[safe sleet]

Idk if this server has a @staff ping or similar, but

[daring rock]

@safe sleet thank you

[serene roost]

wouldve put that shi in a vm or sandbox before downloading very riska stay vigilent

[safe sleet]

Don't worry

Seems like your average token stealer

[fiery gull]

Wow, I felt the deepseek 3.2v better that gork 4.1 fast in my work

Maybe is some people using the speciale and it explode the tokens use

[fiery gull]

Xiaomi? 👀

[proud bobcat]

Mimo Flash V2

Peak model

[proud bobcat]

No no

In fact it’s all people using normal 3.2

DeepSeek brought it home with 3.2

It’s just such a peak model

[fiery gull]

I want my v4 exp until end 2025 deepseek 🥺

Hell no, I know, but leave me dream

[proud bobcat]

I’m so excited for V4

If they update the dataset with the current architecture?

We could have a juggernaut

DeepSeek already goes head to head with gpt with maybe half the parameters

[terse shuttle]

🙂

[zenith burrow]

A high-detail 2D cartoon medical-style infographic showing a Kurkure snack packet cut open from the side. The inner white layer of the packet is clearly highlighted and labeled as a protective barrier. Inside the packet, Kurkure snacks look fresh and crispy. Outside elements like sunlight rays, air particles, and moisture droplets are trying to enter the packet but are being blocked by the white inner layer. Use shield icons and medical-style arrows to show protection. Clean white background with light pastel colors, flat yet detailed medical infographic style, sharp outlines, smooth gradients, educational visual clarity, no realism, no blur, pure 2D cartoon explainer look, perfect for facts video.

[next surge]

For many founders and teams, the biggest obstacle in project development isn’t the idea,it’s finding the right engineers to execute it. Success demands engineers who are technically sharp, communicate clearly, deliver on time, operate well across global time zones, and understand the importance of SEO and influence.
I won’t claim perfection, but I continuously work to meet these expectations. If you’re building something now, I’d be excited to work with you and help drive it forward.

[hearty prawn]

Insane! GLM-4.7 now the # 1 opensource model on Artificial Analysis.https://x.com/Zai_org/status/2005102251908759807?s=20

Z.ai (@Zai_org)

GLM-4.7 is featured on Artificial Analysis Intelligence Index, positioned as a leading open-source model.

X · Dec 28, 2025, 2:23 AM

[brave orbit]

sorry gemini 4 my englishs just bad

[brave orbit]

Gemini 4 just need to know that since i fast typed

[drifting crow]

Guys I have it good authority that pineapple is actually AGI

[thorny schooner]

Damn not being able to use this website at all is killing me

[high aurora]

Please help. Why does EVERY MESSAGE require me to confirm I'm not a robot (reCAPTCHA)? This makes me incredibly sick and angry. How can I make her go away and stop bothering me?

I'm sitting there without a VPN.

And other helpful programs

At the slightest mishap, I immediately start looking for hydrants with ladders. This is really driving me crazy. Very annoying.

What to do?

It seems you're not alone with this problem.

[obsidian hawk]

It’s a new update

I think

I’ve had it a few times too

English

[granite heron]

@upbeat ridge hey there!, this is not the right place to send prompts, if you wanna know how to create content, please head over here: #1397655624103493813 message, you will find a detailed guide on where and how use the LMArena bot.

[surreal creek]

anonymous-1222 still not providing responses 😅

[bleak pier]

hi everyone! this is not the right place to send prompts, if you wanna know how to create content, please head over her

[surreal creek]

what graph is this?

[surreal creek]

wow, and on the one prompt it actually does reply it beats Gemini 3 Pro!!

an interesting model, would love to get to talk with it more, just need it to work 😂

[surreal creek]

annnd ofc back to not replying 😂

so pesky, you are, anonymous-1222!

[echo aurora]

This was flagged yesterday, sorry to hear it's still having issues.

[weary galleon]

👋

[burnt sinew]

Are there any good benchmarks for TTS anyone know?

[opal geode]

Wondering why I cant copy paste in chat screenshots for claude 4.5 but it only accepts screenshot links when its uploaded,while gemini 3 pro accepts screenshots directly,its some bug or ? Thanks for answer

[echo aurora]

This is a problem on our end sorry to say. Something we've been wanting to fix. I can bump the issue on Monday.

[opal geode]

Ok thanks, would be nice to get it fixed, have a nice day!

[echo aurora]

+1, you too

[rigid copper]

sup guys

idk i don't see any much differences between gemini-3-pro-image-preview and gemini-3-pro-image-preview-2k.

[echo aurora]

Got any side by side comparisons you can share?

[still ibex]

Hey @echo aurora

Got a quick question

More like a suggestion

[dusky ravine]

#1372230675914031105 here

[still ibex]

This isn't a feedback

Genuine question actually

So do you self host open source models or are the hosted on their own servers like qwen from Alibaba servers?

If it isn't self hosting what's your source of income, making most models completely free like gpt 5.2 is expensive

Like in general what's the actual source of income

How it it feasible to give these models for free

@echo aurora

[echo aurora]

Hello, that's a good question. Enterprise and research partnerships generate revenue to sustain the platform, while all public leaderboards and evaluation data remain freely accessible. We do not sell user data or individual voting information. Commercial activity is limited to evaluation services, not data monetization. You can learn more here: https://news.lmarena.ai/ai-evaluations/

[still ibex]

Ooh ok

Thanks for the reply man

Appreciate it and appreciate the work you put into this

[fast skiff]

Wide 2D animated night city shot, dense anonymous apartment buildings, dim yellow windows, empty streets, slight haze, slow camera drift, muted blue and grey tones, cinematic lighting, noir atmosphere, subtle film grain, quiet tension, adult serious animation style

[coral birch]

The output resolution is different, with 2K you can see more detail )). Though, with asking the AI to upscale an image I don't think there is a difference (as far as I have tried). You can try asking both model to generate an image and then zoom in, note how with the 2K model you can zoom in slightly more before it start to get blurry with the pixels

[formal salmon]

when will video ai like sora be added into the website?

[queen veldt]

Some people already have the videos in their lmarena

It's random

They have a video button next to the code one

But it's very rare

[gaunt spade]

damn

i want that

[west lodge]

it might be random + region

cuz i can never get it where i live ill try on a vpn at some point to prove

[hushed kernel]

I am trying to log into LMArena with my google connected gmail and am not able to get in. I have even tried logging in by my gmail and cannot get in with the Email Confirmation Token. I get an error message saying that the page can't be reached. Sometimes I can get in, sometimes I cannot

Also, this seems to be related to Google or Gmail, because I can create a new account with a non google email and not have any issues.

[gaunt spade]

new feature

[formal salmon]

idk how you got it lmao

I dont got it

[gaunt spade]

it doesnt work tho

it bugged my site

[queen veldt]

Is this an ad for them or? I never heard of Hugston before

https://www.reddit.com/r/Hugston/s/1vhibmjpwj

[fickle venture]

Dam ig they still not sure to post it OR it's only for developers

[gaunt spade]

yeah its gonna cost alot, cuz there is sora 2 pro

[fickle venture]

Correct

[obtuse smelt]

oh yeah this website delay ?

[opal geode]

also after some use of claude opus or any other model, it gives red text Something went wrong with this response, please try again. and its over, cant refresh or revive it. Is it suppose to be like it?

[subtle frost]

What model produces the most accurate results from searching the web
I use AI to learn a lot in med school, for example I might ask:

How does DNA replication termination occur in eukaryotes?

I think Gemini is really bad at this and Grok is really good, I think Chatgpt is kind of in between

[fierce kelp]

What do people think about multilingual AI?

[fiery gull]

? even 350m model today is 100% multilingual today

if a model don't is 100% multilingual, the model is bad, just it

[fierce kelp]

I mean using mutliple languages at once, such as for language learning. Obviously I know basically every model is multilingual in general.

[obtuse smelt]

hmm gemini is delay but not bad and need fix this website

[astral bloom]

<@&1349916362595635286>

[neon idol]

<@&1349916362595635286>

[quasi atlas]

Thanks!

[obtuse smelt]

is normal again ?

[fiery gull]

Ahh yes, great question, I tried to use english and the result came out another language (portuguese), but it didn't work very well, better to use in the default language, about the trust the AI are 100% fluent in all languages

[jovial star]

May I know what llm are used on vid animation pls ?

In the discord bot

[glass pagoda]

Man these recaptchas suck cant even send a message before it gets invalidated

[gaunt spade]

fr

[fierce kelp]

Send us the post and someone might vote on it for you

[fickle venture]

What website?

[fierce kelp]

LMArena.AI

[vagrant garnet]

How can I create a clip in aspect ratio 9:16?

[left lodge]

New Changes detected, a new automatically detect modality button appeared.

[fierce kelp]

Have you tried putting it in the prompt? You might have to manually put in the resolution.

[fierce kelp]

I have had this a few days.

[vagrant garnet]

Yes I put in the prompt but no difference

Always got a 16:9 ratio video

[gaunt spade]

what does it do

[vagrant ferry]

hello

[left lodge]

A feedback section also appears after the generated ai responses (not on all , but this one appeared on image gen)

[left lodge]

as the name suggests, it tries to automatically select chat modality according to your prompt like image, search, code or just chat.

[gaunt spade]

ohh, i thought it was useless

[sacred ember]

Error

[tulip kindle]

any info when the captcha will be fixed or deleted?

[quaint whale]

hello

[hard plume]

I hope it's deleted

[tulip kindle]

I'm also

It's so stupid system

[fleet lintel]

question : why bother with GLM 4.7 when Gemini 3 flash is about same cost and more reliable and better?

[bright temple]

is nano banana down?

[next ivy]

What’s the difference between LMarena and Yupp?

[cloud zinc]

one is better

[next ivy]

which tho..

[fickle venture]

@quasi atlas sorry for ping but is this allowed to say?

[sour spear]

<@&1349916362595635286> 🔨

[gaunt spade]

nothing

both sites are designed to rate models and test them

[sullen sequoia]

does anyone have a list of which models support image input? I keep gtting "Selected models do not support image input"

[sour spear]

Nano Banana Pro just blew my inner child's mind. That's so accurate, and it generated all 10 pages in full resolution.
ChatGPT on the other hand... 🤷‍♂️

It's hard to read in the screenshots, but all text is integrated into the images without the slightest mistake.

<@&1349916362595635286>

[queen veldt]

👋

[steady lion]

I was wondering — where's the best place to submit a feature request for Chat Session?

I'd love to be able to download my chat history in the future. Is there already an option for that, or am I stuck with copy-pasting everything manually?

[proud bobcat]

i hate gpt

i just

despise

gpt

[novel crater]

gemini 3 flash is like a caffeinated depressed frustrated 9-5 software engineer doing cocaine

[surreal creek]

having an account is a good place to start

[steady lion]

I already have an account using gmail email...

[surreal creek]

oh coolio!

[steady lion]

I can see only an option to delete chat session but not to download it...

[surreal creek]

they may add the feature in the future?

[steady lion]

so I am stuck in copy-paste method

[steady lion]

How can inform that? is there a way to submit a feature (kind of form)?

[regal lion]

Is it worth to buy gpt full year subscription in 18k rupees INR

[queen veldt]

Um no

You get basically 2 months free only

And you're locked in for gpt for a year

Google might improve their ui and add new features

They're just getting started

[queen veldt]

Oh also

Don't use google play for gpt sub

Go on got website

[regal lion]

Why

[queen veldt]

Since play takes their cut and they will make you pay more expensive for a sub

I think it's like 25 on playstore and 20 on their website $

Check the website first

[regal lion]

In website only show monthly subscription it's in web in india 1999 INR

I mean in website sorry for weak english

[queen veldt]

There's no reason of buying full year

They have a promo rn i think

They'll give a month for free

To anyone

So you're paying in advance

[regal lion]

Yes that's why I want suggestion

[queen veldt]

Something that you will still pay in the end

If you get yearly you're locking yourself

You might switch to some other ai in 2 months

You don't know

Don't give them free cash

That's the point of yearly

You will eventually not use it for a month or two and they're in profit

That's how annual plans work

It's a bad deal

$40 cheaper only

[regal lion]

I think you are right bro

[queen veldt]

Instead of 240 you're paying 200

[regal lion]

“This free offer is not visible in India.”

[queen veldt]

Get a vpn

https://chatgpt.com/#pricing

Did you try this one?

Just make sure you click personal in the top

They're trying to trick people

To buy business which is $3 more expensive

Lmao they're so greedy

[novel crater]

is a resting heart rate of 115 normal

[regal lion]

“Thanks, brother I got confused because I was saving about ₹5,000 INR with the yearly plan instead of paying every month.”

[queen veldt]

5000 is nothing

You might switch sometime to some better ai

5000 is spent in 5-10 days

I used to love chatgpt but i switched to gemini

And + i heard gemini is getting UI rework on their app in a month or something

[regal lion]

Thanks for advise bro I will think about it as per your suggestion."

[regal lion]

"Bro, 5000 might be a bit too hefty for my humble wallet right now. I'd probably take a full 100 days just to spend it, and even then I'd be stressing

[west lodge]

gemini's grounding tool syntax?

this could actually just be the tool call syntax

but this happened after a long convo with 3 flash

idk why

[finite flame]

Hey is there any way to use the bot for unlimited time ?

[west lodge]

clear your cookies

gg!

you lose the chat history tho

[queen veldt]

Bruuuuuuh

This app just tried to scam meeeee

DONT DOWNLOAD THIS

It asked me to confirm my card it didn't say any payment or anything

I pressed continue

And it tried to charge me $144

I pressed this agree and subscribe and it instantly tried to charge me it didn't even asked me to charge me that amount

I usually get a pop-up

[zealous sparrow]

A/B test goin on in AI Studio for gemini 3 pro

Either a new checkpoint or an old one

[obsidian cargo]

oooo

[zealous sparrow]

the A/B test is rare so gl

[gleaming roost]

For me, A/B testing always gives me bad or broken answers, it never gives me anything good, most of the time I just skip it.

[zealous sparrow]

Could be different, not sure how often you use AI studio rn

[gleaming roost]

I have a PRO subscription every day.

Simply regenerating some responses a few times reveals that the A/B test appears quite frequently.

[frigid quiver]

short films

[magic ravine]

Seedream is not working

[sterile tartan]

Any news for Gemini 3 GA?

[rustic dock]

I accidentally deleted my LMarena account. When I try to log in again with the same Google account, the site loops to the homepage without any error. Could you check if my account can be restored or fully detached from my Google login?

[queen veldt]

no its stuck

you probably had video button

you need to clear cookies and login with new account

[cloud zinc]

he is trying with his deleted acc

[hexed dirge]

Hi I'm new here

[queen veldt]

doesn't matter, he got the video button when he deleted it

and binded that feature to his before deleted account

and now it's stuck until lmarena devs do something about that bug

[fickle venture]

Send console logs and report on #1343291835845578853

[zealous sparrow]

new AI studio A/B test, so maybe

[gaunt spade]

what are the outputs

[gaunt spade]

also how often do u encounter A/B tests?

in AI studio

[zealous sparrow]

rare

very rare

[gaunt spade]

are the outputs good atleast

or just the same as the current models

[zealous sparrow]

i didn't test coding, text wise i guess itsokay

[gaunt spade]

well writing isn't a very good benchmark test

can u do some html stuff

lets see how it does in UI design

[zealous sparrow]

ran out of my usage

on AI Studio

[gaunt spade]

damn,

u can use other accs

yknow

[sterile tartan]

Good to know

[quartz pike]

meh imo i just use antigravity

[zealous sparrow]

used 3 only got it once

[gaunt spade]

brodie this is A/B internal models

[quartz pike]

oh sh!t yall talking about a/b mb

😭

i thought yall were talking about coding

mb mb

[gaunt spade]

nah i told him to test it on coding

but its very rare to get

[quartz pike]

i think it might be gemini 3 flash lite

idk tho

[gaunt spade]

u probably need 5 google acccounts to get 2 A/B testing in total

[gaunt spade]

or gemini 3 ULTRA

or gemini 3 GA

[jovial star]

Is lm down ?

[quartz pike]

ga?

[gaunt spade]

global availability

[quartz pike]

o

[gaunt spade]

they'll get updated with better performance

[jovial star]

Why I cant access lmarena ?

[quartz pike]

ngl since we are in the preview phase of gem 3 i wonder how good full gem 3 will be

[gaunt spade]

veo 4 is on the way too

[quartz pike]

yea

[gaunt spade]

my friend also showed me some genie 3 leaks

[quartz pike]

veo 3.1 was absolute buns

[gaunt spade]

he did a few prompts for me

[quartz pike]

WHAT

SEBD EN

SEND EM TO ME

[gaunt spade]

genie 3 is good but not that "good"

[gaunt spade]

i cant he told me that i cannot share it to anyone

i wish i could

[quartz pike]

damnnnnn

[gaunt spade]

genie 3 has nanobanana pro integration

u can make nanobanana pro images and it makes a world for you

in the same app

[frigid tusk]

???

[quartz pike]

lmarena is down

[frigid tusk]

dang

[jovial star]

Same 😢

Seem like the server is down

I can't live without Lmarena

[frigid tusk]

i was using it like 2 min ago

[jovial star]

Ye it's down now

[frigid tusk]

u know for how long?

[jovial star]

Do you some alternative pls

[jovial star]

No idk bro

[gaunt spade]

yupp.ai

flowith ai

[jovial star]

Oh thx 🙏

Give more ye

Thx so much

[gaunt spade]

zai.is

[zealous sparrow]

the A/B is happening on the gemini 3 pro model

[gaunt spade]

tbh gemini 3 flash testing was also on gemini 3 pro

[jovial star]

You saved me

[quartz pike]

ooooo full gem 3 pro coming soon?

[zealous sparrow]

for me it was on flash 2.5

[gaunt spade]

are u sure? cuz alot of people showed it on gemini 3 pro model

on ai studio

[jovial star]

Bunni you are the best thx for those gems tools

[gaunt spade]

np

[zealous sparrow]

im kinda confident this might be a gemini 3 pro model tho but im not sure

[gaunt spade]

gemini 3 ultra xhigh codex max thinking deep think heavy pro

i predicted it

[quartz pike]

was this before the release of gemini 2.5 flash latest?

[quartz pike]

wat tools

[gaunt spade]

i sent him alternative of lmarena

[quartz pike]

oooo

send it to me too

[gaunt spade]

alr

[thorny schooner]

https://tenor.com/view/andrealphus-helluva-boss-andrealphus-helluva-boss-gif-2611929239645866336

https://tenor.com/view/larry-david-curb-your-enthusiasm-my-interest-is-piqued-gif-17804129

[gaunt spade]

1- Yupp.ai
2- Flowith.io
3-Zai.is
4-Outlier.ai (requires some information and takes longer to register but gives you unlimited access to LLMs like Claude 4.5 Opus with no limits)```

for yall

@thorny schooner

[high aurora]

And I have it ALL THE TIME. FOR EVERY STEP. THIS IS BRAIN-F******. It's f****** annoying. And he can't f****** leave me alone, d*** it. F***. S*** f******.

[gaunt spade]

just use flowith.io if u wanna chat or create images

it doesnt have weird captcha

[frigid tusk]

Well thats bad, ill wait for lmarena

[gaunt spade]

its only in the US and some parts of Europe i think

and u cant use VPN

[frigid tusk]

ah.

[wraith sedge]

anyobody else cant submit prompts

the button is greyed out

[fiery void]

hello

[golden ocean]

https://cdn.discordapp.com/attachments/1442924145615568956/1442924156546056272/MConverter.eu_v14044g50000d485kpfog65rnjameha0.gif

[native yarrow]

can u use a temp mail with flowith or nah

[torn swallow]

Hello, can you tell me where Veo 3 has gone or how to turn it on?

[tired herald]

Yes you can lol

The only issue is that this doesnt "require some information" because it asks for literally everything, a face verification, a phone number, a linked in, a resume. They expect your life, which they then sell off

[calm vector]

Hello, I'm new here

[fickle venture]

Flowith.ai is it free?

Cuz dam that's so good

With many features

[sour spear]

Is it just me or is Nano Banana suddenly generating copyrighted stuff like Disney characters without any restrictions? I'm pretty astounded right now.

Meanwhile ChatGPT...

[next thistle]

I want object/subjet (like characters, props, background, etc.) uploading option. How can i get this??

[abstract tundra]

is there any way to recover this text?

it streamed on and on until it hit out of bounds and threw the error but the streamed text still stayed. after revisting it's gone

maybe through network requests or other devtools method?

[fickle venture]

That's just a well known issue

Either refresh website or wait abit will fix it

[abstract tundra]

refreshed multiple times, still same

i tried to reproduce it by making it write long text again in another chat so it ran into error midstream again, i refreshed and it removed text and showed error text only just the same

[fickle venture]

The heck

[urban sky]

Yo

[fickle venture]

Is there a server for it?

Tbh the models list is just weird to read

[gleaming roost]

Does this key change over time? Or is it permanent?

[fickle venture]

Permanent

Or until he wants to reset it

[gleaming roost]

oh

[urban sky]

Permanent I might make all api keys work later but for now this is it

[fickle venture]

Oh so you made it?

[urban sky]

Yes

[fickle venture]

Dam that's very nice

[fickle venture]

Btw reading the models list is kinda not organized

[urban sky]

Hm

It kinda has to be like that for it to work

[fickle venture]

Oh alr

[urban sky]

https://sharetext.io/3e66238a

Here

[fickle venture]

Tbh I kinda want to donate 20$ for you to add new models but bruh I am kinda poor right now

[urban sky]

It’s fine

Maybe someone will 😢

[fickle venture]

Hopefully 🙏

[fickle venture]

What does these models do

OpenAi and OpenAi fast

[urban sky]

It was part of the provider I use

I honestly don’t know

[fickle venture]

Dam

Okay

Is it okay to use this on cursor 🥲

[fickle venture]

The team will try to help you cuz this issue is very said here alot #1343291835845578853

[abstract tundra]

posted

[zealous sparrow]

Gemini 3 pro update in January btw

[urban sky]

Yes

[storm needle]

What is gpt45?

Gpt 4.5?

[urban sky]

Yes

[storm needle]

How do you have this model if it was removed from the api

[urban sky]

I am not directly using OpenAI API

[storm needle]

chatgpt pro plan?

[urban sky]

No

Diff api provider

[storm needle]

bro its not gpt 4.5

[topaz bay]

MY EYES

Minimax 2.1 whyy

they don’t have a dark mode

They’ve got the proprietary open source model, but their app doesn’t have dark mode

[jade basalt]

Is there an ai video generator site which has unlimited uses for premium plan for models like veo seedance etc?

[fickle venture]

@urban sky why Claude 4.5 Sonnet doesn't know itself

[urban sky]

Ask what AI model it is or what AI it is associated with

[fickle venture]

Try the model it literally doesn't know it self

[urban sky]

Idk then use a diff model

[fickle venture]

Ok

[fickle venture]

The heck is this model doing here

The first one

[urban sky]

Yeah bro

That was in the provider I used

Add as many models as I can

[fickle venture]

Like what

Do you think you can add GLM 4.7?

[native yarrow]

@urban sky this is what it says for me loading up ur site

[urban sky]

That’s the API

[fickle venture]

That's the base url

[native yarrow]

oo

[urban sky]

My other site

[urban sky]

Wrong link fixed it

https://theoldllm.vercel.app

[fickle venture]

Woah these all on web not the API dang

[urban sky]

Yeah

I’m not making a free api with all those

Will get abused and exploited

[mint dome]

I have a script and want to add voiceover which Elevenlabs does. Is there any model integrated in LMArena which can help with this ?

[fickle venture]

Oh these happens?

[urban sky]

For api 100%

[fickle venture]

Voice model?

[mint dome]

yes

[native yarrow]

so whats the api key for? @urban sky

[fickle venture]

Like send a lot of requests? Dang

[mint dome]

or any free tool that helps - if you know,

[fickle venture]

There is no voice models

[urban sky]

Or use in their site

[fickle venture]

Tbh you can find a lot of models here like voice models:
https://huggingface.co/spaces

[mint dome]

I will definitely try, hope they are not paid

[whole sundial]

<@&1349916362595635286> is this allowed?

[fickle venture]

Bro it's free 😭🙏

[urban sky]

If not I can take down

[fickle venture]

Can I have it private if you don't mind cuz my wallet is kinda homeless

[urban sky]

I haven’t created a api with all them yet

[mint dome]

i didn't get this one, once I open the link, get something weird,

[fickle venture]

Can I have if you do it cuz these models looking like

[urban sky]

Yea sure

[urban sky]

Those are model IDs

[fickle venture]

Use base url and API key to send request and that's the model list

[fickle venture]

Right now or later? 🙏

[urban sky]

It will take a while maybe I will do it

It will be very exclusive

[fickle venture]

I will be using it slow cuz dam

[green yacht]

@urban sky it shows GPT-5.2 is a model in the website, but for the shared text with the models, it doesn't have GPT-5.2 in the list. does that mean gpt-5.2 is not supported by api, or you just didn't add it

[urban sky]

It isn’t added to the api yet

[native yarrow]

lowkey how do i even use this lmao

[urban sky]

It is a open a compatible api

Mostly for coding

[fickle venture]

Use it on cursor

[native yarrow]

oo i see

[echo aurora]

This feels a bit too much like an advertisement tbh

[urban sky]

Should I delete?

[echo aurora]

That would nice, thank you.

I think if you’d like to talk about your project that’s okay.

[mint dome]

any model in LM arena that helps convert a script to video?

[fickle venture]

Text to video?

[mint dome]

yes

free tool

[green yacht]

love the tool but chat we gotta gatekeep this 🙏

[opal geode]

hi, also after some use of claude opus or any other model, it gives red text Something went wrong with this response, please try again. and its over, cant refresh or revive it. Is it suppose to be like it?

[urban sky]

new api link is https://the.oldapi.workers.dev/v1

[torn swallow]

Hello, can you tell me how to make a video with veo 3

[subtle frost]

What would be the best model if you're trying to look for specific factual scientific information?
would I be best off just going with the model with the lowest hallucination rate?

I'm using these models to help me study

[scenic hawk]

Where to find or gain access to API keys if you're willing to share?

[urban sky]

I can’t share the discord server here just dm

[viscid cloak]

looks like nbpro experiencing high traffic/error rate at the moment.

[knotty fable]

I really wish I could provide you with a good answer, but the reality is that there's none.
No current public AI model to my knowledge look for the actual sites where new papers are published, and then give priority to the most recent ones. [I have taken a look as well as some people I've talked to.]
One reason for that is that quite a number are behind a paywall, I actually got an idea on how it would be possible to create a decent model anyway, but I am not the person to create it.

[subtle frost]

I mean as long as it was said in some paper then it's mostly fine. I just don't want the AI to just make something up. Yeah some old papers may not be as up to date but generally that's not too big of an issue. At least for stuff that I'm learning.

You seem to have asked other people about this, so you must have also been using models for a similar as me

Can I ask if you are using AI to study like me and if so, what are you studying? And if you were to pick a single model, which would you pick?

[rigid copper]

sup

[rigid copper]

i actually tried that all the time, lol

[rigid copper]

ooh, that's why, but for me it look the same

[knotty fable]

Yes, that would be the criteria I would use - the source from actual revised papers and not something from lets say Wikipedia.
I actually don't use any, I had a look into the matter to see if I could find things more easily. But to no avail.
While we do study for life, I'm no longer a student as such, and a researcher in biology.
I will give you a DM on where to look for material and aources and where you will have the benefit of asking actual people instead.

[lucid zinc]

Yeah good stuff 😂

[frigid tusk]

i tried logging in but it just keep get me back here

[fading rover]

It looks like the gemini 3pro model has some issue past two hours it keeps failing with something went wrong reply

[viscid cloak]

right, about 1/6 or less success rate.

[fading rover]

For me 0/10

[viscid cloak]

tried an alternative platform, at least nbpro runs perfectly

[fading rover]

Gemini works fine must b some load on lmarena api

[golden ocean]

fr lmaoo

disable them manually by clicking on the hidden "Cookies" button in the bottom of the side panel

after clicking accept

[lucid zinc]

Why even have the option if 'manage cookies' doesnt do anything

[vernal zodiac]

can't use claude sonnet in my chat anymore

it says "something went wrong with this response, please try again"

there is too much context in the current one

so i can't create a new chat

heelp please

[obtuse smelt]

Something went wrong with this response, please try again.is really bad

[signal kayak]

Hello all gm

[fading rover]

Seems like all models have same issue ??

[obtuse smelt]

uhmm

[viscid cloak]

Highly possible. But don't know if mods/programmers would fix it in holiday season or not.

[fiery gull]

I'll be crazy, I soooooo python code I cant read, I just trusting in opus 4.5 to don't explode my pc ;-;

I even don't know how to write hello world in html lol

[wicked sage]

what in the world is gpt image 1.5

[fiery gull]

lol

gpt internal test

[wicked sage]

al

alr

[obtuse smelt]

need fix this

[hollow flicker]

GLM 4.7 fell off

[gaunt spade]

why

[hollow flicker]

every time i tell it to write code EVEN ZIPHU SAID IT WAS TRAINED FOR CODING and it cant fix my discord bot

[gaunt spade]

damn, sometimes even claude models fail that too

[hollow flicker]

sometimes

but this is ez

like it should be able to output 3 lines of code

[gaunt spade]

what model do u use for coding most of the time

[hollow flicker]

k

[hollow flicker]

mainly claude for small files sense LMArena doesnt allow you to upload files to claude opus 4.5 but i use Gemini and other ais depending on the task like for UI gemini and backend claude

[gaunt spade]

yeah that makes sense

i wonder how claude 5 opus would perform

[peak sapphire]

@echo aurora Video generation has been added to Lmarena, but there is an issue: models like Kling 2.6 Pro and others feature video editing capabilities, yet for some reason, Lmarena has only implemented image import. Please fix this oversight.

[gaunt spade]

does it work?

cuz it didnt for me

[peak sapphire]

Video generation works, but video import is missing.

[gaunt spade]

i dont see video feature

on lmarena

[peak sapphire]

It is very difficult to obtain; I spent over 40 minutes on it. I managed to achieve this using browser incognito mode + a VPN.

[gaunt spade]

wow

[peak sapphire]

I was surprised myself because video generation is very expensive, and I didn't think they would add it to Lmarena for free.

[ocean vortex]

It's expensive for sure, but generation itself not gonna cost them more than on https://discord.com/channels/1340554757349179412/1397655695150682194 I would think

[leaden meteor]

So, are we getting grok 4.2 on wednesday? I dont see any chatter yet...But 4.1 was a surprise release...so, you never know...

[ocean vortex]

They still need the same API to do this on discord. It's just video storing part that they eliminate by having this on discord

[gaunt spade]

i got the feature yesterday, made a new account to test it and after that, it bugged the site even tho it says im on the account. It says make a new account

i cant even use it

[sterile otter]

I can't log with my Google account pls fix that problem

[gaunt spade]

I have it on my browser, but it doesnt work it says "make a new account"

[queen veldt]

Yes

It doesn't work

You cant get the button

But when you log in

It doesn't say you're logged in and can't use video at all...

[gaunt spade]

yes thats my exact issue

it broke my new account

[queen veldt]

Everyone has exact issue

Video button = can't use account

[gaunt spade]

yea very weird

[obtuse smelt]

several hours always error

[sterile otter]

Who can't log to lm with his google account ?

[sterile otter]

Same problem

[torn mantle]

prompt?

[obtuse smelt]

is not prompt

in several second or hour is look error

[torn mantle]

yea but whats the prompt you used

or it happens with any prompt

could be rate limited or something

[peak sapphire]

It worked for me once, but now the video generation button just doesn't appear. I've tried changing browsers and VPN servers several times, but nothing helps. Also, the account login has stopped working.

[whole lotus]

Is there a checklist for what is considered NSFW in an image and what is not?

[ocean vortex]

If you mean censorship/flagging, that is gonna be different depending on provider unfortunately.

[uneven yew]

is there anyway to fix the Something went wrong with this response, please try again thingy

[ocean vortex]

Same is gonna be only extreme things like full nudity, excessive blood, deepfakes, cruelty etc

[viscid cloak]

For most cases it would not be from users’ side.

[uneven yew]

so its internal

[viscid cloak]

Which means there is really not much we can do

[gaunt spade]

yeah the devs probably shutted it down

[proud bobcat]

DeepSeek is back

[reef garden]

Is there any way to use this bot unlimited time

[sage hinge]

has the video generation button disappeared?

[opal geode]

So claude opus 4.5 32k thinking is impossible to use, after asking one task it does it,meanwhile halfway it stops because of this error "Something went wrong with this response, please try again." and it gets stuck again halfway another task,basically impossible to use. only on small short tasks.... is it possible to fix it?

[sterile otter]

Fix the god damn server

[sage hinge]

I believe they are officially implementing the video generator, which is why the servers are buggy

[sage raptor]

source ?

[sage hinge]

I just think 😅

And I hope so.

[echo aurora]

Morning everyone I'm back this week so will be more available. Thank you all for your patience with the team while we're off. I'll be going through #1343291835845578853 #1372230675914031105 but if anyone has questions don't hesitate to ask. I hope you all had a nice holiday!

[glass pagoda]

Hey pineapple, will something be done on the recaptchas?

[opal geode]

Hi, posted bug for claude 4.5 , thanks in advance

[echo aurora]

Sorry to hear this. This error message unfortunately is generic so the problem isn't going to be one issue I can point to. That being said, rate limit tends to be the most common reason this error is happening. You can verify it's rate limit by...

Open Developer tools in your browser, go to the Network tab, search Stream, run a prompt in LMArena that'll result in an error message, check the file named with that sessions Eval ID (random set of #s and letters in the URL), then look for a Status Code. If that code = 427 you're being rate limited.

This article here has some steps that may help along with a way to report the problem with greater detail - https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message.

[visual osprey]

how many votes did ernie have when it debuted on text leaderboard?

its imperative you reply

[echo aurora]

It's currently an experiment, so it may or may not be there.

[echo aurora]

For 5.0? I'm not sure if we disclose that specific information.

Can I ask why it's imperative to know?

[visual osprey]

theres a hostage situation with my cousins and the hostage takers say they need the answer or theyre getting beheaded

tough stuff

[echo aurora]

We are always going to be lookinginto ways to enchance our captcha system. For now, the best thing to do is report individual chat session Eval IDs in #1451574502369656842 .

[sage hinge]

oh okay, I only managed to test it once yesterday and it disappeared 😂 thank you for clarifying!

[echo aurora]

I'll check with the team if I can answer, but your cousin may be out of luck

[visual osprey]

its okay i dont really like them but id be happy to get an answer

[sage hinge]

@echo aurora since you're here, I'd like to thank you for the great work, lmarena is amazing and it's free, that helps me a lot, so... thank you!

[drifting crow]

https://cdn.discordapp.com/attachments/934024231098925078/1455215602703536394/media-G9TLFgDWoAEW094.png?ex=6953eaa4&is=69529924&hm=6496a10edec0cda11d0ff2f07714e6b4cf66f4b36ccf68aaf68c362537bdf89a&

[sour spear]

I kinda feel sorry for it. 😉

[keen beacon]

What

@echo aurora Help Me

[serene roost]

clear cookies

[echo aurora]

Oh I'm sorry I think I misunderstood you the first time, that's my fault. On the leaderboard the total amount of votes is listed - https://lmarena.ai/leaderboard/text

[opal geode]

I did exact same thing in order to check Eval ID if error appears, and now it did send me code and everything fully,so i think it was server bug before?(like 3 times)

[visual osprey]

oh thx i didnt see

[echo aurora]

Sorry to say this is an issue on our radar. If you clear your cookies/cache this should help it. Let me know if this doesn't help.

[vernal zodiac]

some ideas?

[echo aurora]

What was the status code?

[echo aurora]

@vernal zodiac this is the same issue we're discussing here. I'd recommend these steps.

[echo aurora]

Thank you, I really appreciate this ❤️

[safe token]

guys

if its the end of the world

and chatgpt was taking over the world

which ai would beat it

[opal geode]

no status code, no error message, it works now again

[echo aurora]

Well glad to hear it's working again. If you hit the something went wrong error try to repeat the steps and see what that status code is and keep me updated.

[hearty prawn]

In coding I would say it is better, I mainly use glm for coding.

[vernal zodiac]

it should be something like "e/?ip=0&_=1238 ..."?

[zealous sparrow]

water

[safe token]

https://tenor.com/view/ishowspeed-meme-trying-not-to-laugh-laugh-homeless-gif-9458928528704746136

ig its the right answer

but like

i was expecting like

an ai model

[echo aurora]

This is the Eval Id

[vernal zodiac]

ah okay

where i can find that file?

check the file named with that sessions Eval ID

[echo aurora]

It's going to be in the Developer Tools -> Network tab, in there if you search Stream it should appear.

[vernal zodiac]

ah i see

[echo aurora]

But note that you should have this open, then run a prompt on LMArena

[echo aurora]

What status code are you seeing?

[vernal zodiac]

429, but before that i tried to send prompt again several times

so even unsuccesful attempts to send a message are counted

and that's what i see now

[echo aurora]

Gotcha, yeah that's going to be rate limit.

[vernal zodiac]

So i will try again in 49 minutes🥲

[twin hamlet]

Will we have any updates on elo score soon when does the Christmas break end

[mild minnow]

Does anyone else have issues with Lmarena where it lags hard or freezes sometimes?

[balmy mist]

is there a spatial reasoning benchmark?

[echo aurora]

We'll be sure to put out an announcement when the next leaderboard update goes out

[echo aurora]

Can't say I've been hearing reports of this recently. Awhile ago there were reports of issues when chat sessions get too long, but I haven't heard this in awhile. Can you elaborate a bit more on what is happening?

[mild minnow]

So, what happens is when I talk to the AI about any topics that usually take a long time, it starts off smooth and fast, but as you keep consistency, it starts to slow down. Sometimes, it will say, "Sorry, something went wrong. Please try again," after lagging extremely hard. Not only that, but when it lags hard, the entire UI disappears, leaving a blank slate.

[vernal zodiac]

tried again

but still

in the new chat it works fine

[half mist]

https://discord.com/channels/1340554757349179412/1455254199435657382 Having issues with LMArena. I can’t just paste large html code without it failing

[whole lotus]

I was initially on LMArena Main website getting an option to create video. I entered the prompt, I was asked to login and the video option was no longer available after logging in. So is there no option on the main website for video creation?

[echo aurora]

Hmm this is interesting:

but when it lags hard, the entire UI disappears, leaving a blank slate.
I heard something similar from someone else so this is starting to look like a new bug. Let's use this channel to dive in more: #1455003207817363509

[echo aurora]

Hmm that's odd. Seeing you were once seeing 429 (rate limi) and now seeing 200 along with the error that's very strange.

[hardy lion]

video on website is currently an experiment not enabled for everyone all the time

[echo aurora]

Could happen! A good place to keep an eye on for all updates is our Leaderboard Changelog: https://news.lmarena.ai/leaderboard-changelog/

[placid lava]

just a question. why is my gpt-5.2-search searching for so long? i left my pc on for 3h and it is still searching

[echo aurora]

that sounds like a bug, may want to start a new chat.

[placid lava]

😭 but then he will forget everything

i will have to go from start

[vernal zodiac]

yep

have some ideas?

[echo aurora]

Yeah that's true. Have you tried using a different browser? Clearning your chache/cookies + a hard refresh may help here.

[placid lava]

i will try

[echo aurora]

I do not unfortunately. The overall problem of this Something went wrong where the status code = 200 is something that's on our radar.

[vernal zodiac]

oh..

maybe it's possible to download all the information from my chat?

[echo aurora]

Sorry to say we don't have this feature currently.

[hearty scroll]

We are just at the beginning of AI. Discord servers stay tuned!

discord is the future of AI

[ocean vortex]

Gemini Flash is one of the least reliable models out there

Hallucinations are through the roof

It does perform, but the price to pay is that - hallucinations and it making things up

[proven bronze]

Again with the captchas, even for retrying one lol 🙄

[fiery gull]

Nobody buy api from glm, so expensive lol, have the code plan with almost infinite tokens for 3$

[fiery gull]

But gemini 3 still in preview, will be released the original version

[echo aurora]

Be sure to share Eval IDs in #1451574502369656842

[surreal creek]

next update will likely be GLM-4.7, unless there’s a surprise Grok 4.20 release by EOY?

[ocean vortex]

Preview is just a name. You shouldn't read too much into it

They also have a history of literally renaming preview checkpoints into GA

[heady narwhal]

why do I keep getting theses errors when trying to use the website?

[ocean vortex]

what specific url is this error from?

[heady narwhal]

https://lmarena.ai/

[peak sapphire]

How to fix this error that there is a message in the middle of the dialogue the genini 3 pro it just starts giving an error and that's it

[golden ocean]

that preview image

[echo aurora]

Can you try clearing your cache/cookies and then trying again to access the site?

[heady narwhal]

yeah lemme try

[echo aurora]

Can you try the steps in this help center? https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

It's likely going to be a rate limit issue. You'd be able to verify this by: going to Developer tools in your browser, go to the Network tab, search stream, run a prompt in LMArena that'll result in an error message, check the file named with that sessions Eval ID (random set of #s and letters in the URL), then look for a Status Code. If that code = 429 you're being rate limited.

[pearl wyvern]

Hello A question

[echo aurora]

Not a fan?

[echo aurora]

Fire away

[pearl wyvern]

I see that the video says the website is free without needing to pay and that the AI models are free. Is this true or false?

??

[peak sapphire]

Its true

API keys for Lmarena provided by the developer companies themselves for the purpose of testing their AI models on users before release

[pearl wyvern]

Wow, that's great, but you don't need to pay for anything?

[gaunt spade]

@echo aurora i also get this error when i made a new account

[peak sapphire]

No, its 100% free

Absolutely free

[pearl wyvern]

Thank you very much

[heady narwhal]

is there any limits to how much videos or images you can generate?

[echo aurora]

We pay for the inference cost too btw. I don't think is too note worthy but wanted to share.

[echo aurora]

Yes, for Video Arena it's going to be 5 generation requests per 24/hour

[pearl wyvern]

Which model do you recommend for playing a good RPG game?

[heady narwhal]

ah ok thx

[pearl wyvern]

Chat-based

[echo aurora]

Also, on the site models are going to have their own rate limits

[peak sapphire]

I only found limits for the most expensive and trendy models like nano banana pro 2K; I didn't find any other limits for photos. For videos, they logically exist and it's only 3 videos every 24 hours, but that makes sense since video is the most expensive type of AI content generation.

[echo aurora]

Would point you to our Text Arena leaderboard with a Creative Writing filter applied - https://lmarena.ai/leaderboard/text/creative-writing

[echo aurora]

Yeah sorry to say this problem is something that's on our radar.

[pearl wyvern]

Thank you

[gaunt spade]

thanks

[echo aurora]

it's only 3 videos every 24 hours
This is true for the site experiment, but using Video Arena via the Discord bot it's going to be 5.

[peak sapphire]

I meant lmaren's site, I don't know about Discord

[peak sapphire]

I've tried everything, but nothing helps. This happens to me in many chats when there are a lot of messages, as if the LLM is running out of context window and can no longer respond, which is strange for such advanced models as Gemini 3 Pro.

[woeful saddle]

guys i don't know why but i had the video feature for a sec on lmarena web, but whenever i tried generating something it asked for login. so i did, but once logged back in it disappeared. was it meant to be shown right now? is there an admin i can ask about? I'd like to have the feature back haha

[peak sapphire]

Since Gemini 3 Pro has a context window of about 1 million tokens and I’ve barely used 100k, yet I’m already getting a "something went wrong" error. It’s more likely that the issue isn't the context window, but specifically the number of messages—as if there’s a limit of no more than 40 messages per chat, even if the AI model's context window allows for more.

[peak sapphire]

Yes, this is a common problem. To get this feature back, I spent 40 minutes switching browsers and changing VPN servers. It’s very hard to access right now because this video generation feature is currently in testing. You can reach out to the administrator @echo aurora

[woeful saddle]

Thank you so much. I tried too, vpns, even using an US browser. It took me like 4 days, this is the second time and for the second time i felt like being a schizo. I'll try contacting him tysm!!

[gaunt spade]

did pineapple give u the feature

[peak sapphire]

No, I noticed it myself today, but then it disappeared, and I started looking for a way to get it back. As I understand it, only a couple of percent of lmarena users have it right now to test the feature before its general release.

[woeful saddle]

Yeah, but the problem is that it just disappears whenever it asks for login. I tried generating a video but it simply doesn't work.

Even if I "gain" that access, it vanish the second I log into the account, just like another feature I don't remember the name of and don't have any screenshots.

[frank stone]

hello. Is there no way to put open source models against proprietary ones?

[sturdy prairie]

Why do I always get the following message claude-opus-4-5-20251101 Something went wrong with this response, please try again.

How can I delete my entire chat history?

[sleek crow]

@echo aurora Why every time i got captcha when using the chat im going insane

fix this

please

[whole sundial]

share the eval id (the random string of numbers and letters after the lmarena.ai/c/ in your web browser's search box) of the chat you have the problem in this thread: https://discord.com/channels/1340554757349179412/1451574502369656842

[hardy lion]

If you use side by side mode you can select any pair you want, open source or proprietary.
https://lmarena.ai/c/new?mode=side-by-side

[frank stone]

@hardy lion sure, but there are quite a bit of models lacking there.

[frank stone]

i hate captches with all my soul, i feel sorry for ya.

I wish I could clash Apriel against GPT-5 or Grok and Gemini. 😄

[native yarrow]

groks lowkey underrated SOMETIMES

[echo aurora]

Sorry to hear this! Be sure to share with us the Eval ID (random set of numbers/letters in the URL) for the chat sessions this is happening to in #1451574502369656842

[frank stone]

@sleek crow 👆🏽

[hoary egret]

👋

[woeful saddle]

Hi, I sent you a dm but I don't actually know if there's an appropriate thread section for contacting mods. What should I do to talk with you?

[echo aurora]

I'll respond soon. Thank you!

[woeful saddle]

Thank you so much sorry to bother! 🙏

[dreamy crystal]

my name is moe n2m u all

guys

please if u guys have experience, what is the best deep research model?

[drifting crow]

a grad student

[echo aurora]

It's no bother at all!

[dreamy crystal]

man if anybody knows id rlly appreciate it brothers and sisters

[jade egret]

personally Gemini but it really up to you

[dreamy crystal]

ty my friend

I had a tough time choosing between opus 4.5 and gpt 5.2 but I never tried gemini

[jade egret]

all decent ngl

[raven stone]

👋

[echo aurora]

Why

[golden ocean]

Not a fan?

What about this one:

[echo aurora]

Big fan

All your cat videos bring me joy

[half mist]

Got this new button, but it does absolutely nothing. Unless @echo aurora can tell me what beta feature this is, it remains a mystery

[echo aurora]

That's an experiment with Auto-Modality. With that enabled it'll automatically send you to the correct modality instead of manually selecting. For example if you prompt "generate a picture of a blue dog" it'll send you to Image modality.

[half mist]

Ah, that’s kind of cool

[echo aurora]

Agreed! It should save users a lot of frustration and confusion, esp those that are new to the site.

[half mist]

I think it’s pretty cool and definitely useful. I have a suggestion though that I will also put in the feedback section. I want to be able to combine the code arena, and search arena so it can code something, but search stuff up while making it

[sterile otter]

I can't live without LMarena ⚔️

[echo aurora]

Gotcha. Yeah that would be nice to have. I think where the problem exists though is what happens to models that don't have that tooling, and would evaluations be fair to those that don't?

[half mist]

That’s why I think we might need the LMArena Playground that someone else suggested since it would be a playground that doesn’t involve evaluations and what not

[echo aurora]

Makes sense.

[noble pawn]

grok not on any of the image or video leaderboards?

[echo aurora]

We don't have grok models for image or video modalities currently.

[noble pawn]

I wonder around where it would rank

[ocean vortex]

Probably not very high

They are behind on this stuff

[noble pawn]

I see

[grave bay]

would it be insider trading if the devs bet on any of the kalshi markets that use LMArena as a resource? they didnt ask for permission from what i gather and its manipulatable (albiet at the cost of their integrity) scores.

[whole sundial]

no, because xai hasn't added them to their api, although there has been a few grok image gen models on the arena in testing in the past month or so

[echo aurora]

I'm not sure if that'd technically be insider trading or not. Regardless the people working here understand that'd be very unethical and the organization wouldn't be okay with someone doing that. It's really important we operate with integrity for our leaderboards to be taken seriously.

[grave bay]

i was just looking over the single market where they specifically name the site, turns out almost everything AI based there is verified through LMarena. do the devs know that kalshi is making millions off them?

[echo aurora]

We are aware of these betting sites.

[fiery gull]

You're sure, but it's first version, you get it?

[vernal zodiac]

could it mean something? @echo aurora

[echo aurora]

Tbh I'm not sure, can I get a bit more context?

[echo aurora]

As in do we make leaderboard updates based off of what these betting sites are doing? No. (Apologies in advance if I'm misunderstanding the question here)

[whole sundial]

they just added another one named sumo to lmarena btw

[neat apex]

Gemini 2.5 flash were stable, and they updated it back to preview 😅

[neat apex]

The main gape is the ranking is made with hidden models here, and Yupp not

Barely changes a thing, most people in Yupp will not have mercy in Gemini 3 if Mixtral 7x8b do a better answer

[keen beacon]

uh

why is there recaptcha popping out

[vernal zodiac]

it is in F12-Issues

Just noticed it, norhing more

appears when the site loads

[keen beacon]

@vernal zodiac yo

you got recaptcah in your lmarena?

[vernal zodiac]

nope

[keen beacon]

the hell