general

[fiery gull]

python is to otimize txt, and replace docs, opus to txt etc

yep ;-;

[bleak cave]

?

[fiery gull]

sorry, I need to sleep now

[spark python]

Ur making it json?

[fiery gull]

depend on the application

[spark python]

But for what reason do you need that many pythons

[fiery gull]

I tried use json but I still don't need it now

because I have 700gb of data

[spark python]

Of text?

[fiery gull]

word, text, opus, ever thing

[spark python]

And what's the context in them

[fiery gull]

I'm automating a giant education company (a really big one), I'm simply exhausted right now, I can't even write anymore, I should go to sleep now. I'm automating everything from company posts to WhatsApp replies and so much more

I'm currently working on 5 opus 4.5, compressing text for LLM to read

I think in the future I will need hmmm, 200 agents and 300 skills, something

[spark python]

Ok go rest

[fiery gull]

and 1 gemini 3.0 pro seeing now 20 instagram to make a doc to my core

Things are crazy here, in the future it's going to be chaos for me to organize everything 🙂

I'm thinking better, is better the flash 3.0, because the pro high is so slow to see instagram posts ;-;

@echo aurora

lol I'm so dumb, just now I released I can see the instagram profiles with python ;-;

why I'm seeing with gemini? bruhhhhhhhhhhh

Maybe it's because I've only been studying Python for 5 days and I'm still making these mistakes

[junior sonnet]

Does anyone know a website where I can use Nano Banana Pro for free or with a free trial instead of LM Arena?

[fiery gull]

bruh lol

[thick pawn]

It's a good question since lm arena doesn't like to work anymore

[polar niche]

Where are the new mf text llms

Not everyone wants image generators

[proud bobcat]

NEW QWEN IMAGE

Oh yeah baby

Jackpot

[polar niche]

Literally qwen

How is that a jackpot

[proud bobcat]

what

Open source image generator model with extremely high prompt following is not a jackpot?

[gusty egret]

When did it release?

[proud bobcat]

Free to use and cost effective???

[chrome goblet]

Can you guys make GPT 5 faster?

[gusty egret]

Use nano

[proud bobcat]

Up to openai

[polar niche]

Last year

[proud bobcat]

Few days back

[chrome goblet]

Crap

[polar niche]

Now

[polar niche]

We are not wizards

[gusty egret]

You're not talking about their old ahh model are you 😭

I'm too up-to-date with AI man

That's already old news to me

[chrome goblet]

Gemini sucks at doing Random Fictional Characters Stories

[polar niche]

Brother just look at #announcements

[chrome goblet]

I keep making scripts and it's the same crap every time

And I don't like it

[polar niche]

[chrome goblet]

Y'all feel me?

[proud bobcat]

Oh wow qwen image 2512 is quite good

Noticeable increase in detail

Not bad at all tbh

[proud bobcat]

What.

The

The updated qwen image

2512

It’s a new model update

Chat

[chrome goblet]

Get rid of that crappy CAPTCHA!!!

[echo aurora]

We are discussion captcha issues in a thread. I would encourage you to view this message: #1451574502369656842 message

[chrome goblet]

Ok

[fossil socket]

@echo aurora any idea when y'all gonna add the attach image feature from GPT and Gemini onto other models

[obtuse smelt]

hmm what you mean ?

oh yeah i use gemini model and it can on anime styles

[vague solar]

where did all the claude models go 😭

[obtuse smelt]

uhmm

try you refresh a site

[vague solar]

yeah i tried

nvm i fixed it

that was weird

[gusty egret]

I have somehow managed to get Gemini to output the exact same essay twice in the row

Token for token

[obtuse smelt]

if you side by side is delay 1 hour for using model

i try use gemini is delayed waiting 1 hours really in side by side

[chrome goblet]

I can't do anything because of CAPTCHA

[fossil socket]

I meant those 2 models can let you send images on the direct chat

But others cannot

[obtuse smelt]

hmm try side by side

[dreamy crystal]

is there chat arena of somesort that allows deepresearch?

[prisma cipher]

This has happened twice now that Claude opus 4.5 has not generated the complete response for me even after refreshing the page.

[verbal nimbus]

Try copy pasting the entire output into here to get the token count: https://claude-tokenizer.vercel.app/

If it's ~4096 it's probably due to this: https://discord.com/channels/1340554757349179412/1424327476376502403

[prisma cipher]

The number of words in each response that Claude opus 4.5 should give me is: 8520 maximum words, not tokens. It's a very precise solution; precision in the command is required.

[prisma cipher]

I recommend: Generate your entire response with approximately 8520 words at the maximum.

Reduce the number to 8250 to avoid the red words haha

[craggy wasp]

vertical size does not come

[prisma cipher]

One more thing: it is very likely that the number of responses when using this number of words will be, at most, between 10 and 12. You have to choose the appropriate number according to what you are doing and the limitations.

[left lodge]

video modality appears for logged in users this time but with a twist now , only battle mode is available for video , supports image input and both videos have to be first played to abe able to vote for feedback.

Seedance v1.5 pro is good lol

Only after 3 generations, 💀
The platform is really limiting.

Damn

[fiery gull]

bruh the limit from anti grativy is only 8

I need 9 opus working now

[empty stump]

8 messages?

[jovial ravine]

hi any one here ?

[surreal creek]

kiwi-do is Kimi?

[jovial ravine]

@echo aurora check dm please

[fiery gull]

why dm bro, lol, just ask here

[shut crown]

Hello good to be here

[leaden raft]

hello

[echo aurora]

I'll respond soon 👍

welcome welcome @shut crown @leaden raft

[obtuse smelt]

hi

[fiery gull]

hi

[jaunty oyster]

Happy new year :))
I am Buschi, and i am less times here in discord..hope to find a nice LMArena Matrix channel ...

[viscid cloak]

ReCAPTCHA+Something went wrong COMBO hit again. Good Night😪

[robust sluice]

my best friend is inf gen time, feels like its normal to me right now

[subtle frost]

I don't get how Gemini 3 Pro has higher HLE score than Claude.

I've mainly tested Bulgarian grammar and medical information. From the tests I've done: Claude does better more than 90% of the time.

Is HLE a bad benchmark for what I'm actually trying to look at though?
I'm mainly looking for an AI that can search the web, find official sources, then interpret and present that information correctly. Definitely not to synthesize his own information.

[robust sluice]

<@&1349916362595635286>

[ionic sonnet]

This kind of task seems like it depends a lot on the process and orphestration rather than just being about the model

[atomic mantle]

helo

[whole lotus]

I saw a video available around 1 hour ago on LMARENA but now it's gone. Is it only temporary available?

[hollow ivy]

-# another account got zombified

@quasi atlas could you remove the scam, please?

he could try to delete all cookies, logout and login
if that doesn't help, try another browser

if chromium-based browsers don't work anymore, try a firefox-based one, like LibreWolf

or safari

[grave mesa]

What is the best AI for coding in general?

Claude ?

[jade stirrup]

hmmm

[quasi atlas]

Thanks!

[hollow ivy]

… Opus-4.5-Thinking

(aka coast = coASt = co45t)

coasting along ^^

[grave mesa]

Okay thx you !

[golden ocean]

-# That model is head and shoulders above all other models, in coding.

[pseudo dagger]

why this website not working in my desktop?

[finite blade]

@daring jewel

@daring jewel dm please

[jade stirrup]

you guys can fix other people's accounts, right?

[remote sun]

It's happened... Again... In a new chat...

IT COULD

[bright shard]

@echo aurora When I upload an image to the Nano Banana Pro or Nano Banana model and need it edited, it returns the original image without any changes or modifications. Please fix this 🙏🏼🙏🏼🙏🏼

[main path]

👋

[coral birch]

It's a thing with nano banana itself

[queen veldt]

I can't even generate a single image with nano banana pro

This problem isn't fixed yet ...

[woeful portal]

google login is broken HELP

[wicked sage]

hi guys

<@&1349916362595635286>

[wicked sage]

correcto! its claude
i dont know what they fed opus/sonnet 4.5 but its really good

[silk kayak]

Hello everyone. I've encountered an issue: there's no "Create Video" button. What could be causing this and how can I fix it?
@everyone

[obtuse smelt]

hmm should fix all

claude,gemini,grok and all models back to normal

[silk kayak]

I have 2 emails. On first- all works nice. On second - this problem

[neat apex]

The reason is, besises Claude grows a little more than Gemini using search and deep reason, its not enought to compesate the brute general knowledge advantage of Gemini

[rigid copper]

sup

[rigid copper]

not all people will get the video creation feature, that feature will show by your luck

[subtle frost]

Are you saying that when the AI is limited to its current knowledge that Gemini is better but when the AI can search online Claude is better?

[fickle venture]

It's still in beta so not everyone have it. You can use it here check #1397655624103493813

[fickle venture]

Gemini is literally owned by Google and google owns the most used search engine of all time, so they be using their search engine to train Gemini even greater. But other Companies like Chatgpt and Claude doesn't have one so they suck at searching.

I think OpenAi did make their own search engine

If people use this one then Chatgpt will get even better

[subtle frost]

Gemini never searches properly

It always pretends to search then says some BS

The link either goes no where or it just completely incorrectly synthesised the information

Other models like actually use a good source and don't hallucinate info about it

[prisma cipher]

Has the problem generating a response for the Opus 4.5 model been resolved? Or is it still ongoing?

[subtle frost]

At least from what I found

[fierce kelp]

The Leaderboard says GPT 5.2 Search is similar to Gemini 3 Pro Grounding.

https://lmarena.ai/leaderboard/search

[neat apex]

Grok is also not a great searcher, but he absuses the 2 million context + Twitter condesated information to check out literally more than 100 sources

thats why its the best at all

[subtle frost]

The lb does not reflect how correct the model was tho, just subjectively how good the answer is

Gemini 3 grounding gives very inaccurate answers

[fierce kelp]

That just makes the point that the user I was replying to even worse then

[subtle frost]

Yh true

[subtle frost]

I find it better than gemini tho. It doesn't go through the sources deeply but it at least doesn't claim the source said something it didnt

I wonder if there's a benchmark at how good an AI is at finding sources and providing an accurate answer based on the sources

[fierce kelp]

You mean like a needle in a haystack test except with websearch instead of context or RAG?

[subtle frost]

Ohhh doesn't needle in a haystack have a search leaderboard?

[fickle venture]

I mean these model still sucks we haven't reached the 100% on all tests

This is just like means Ai still in Beta

We still gonna wait for the full release of Ai

In about 1-2 years

And that's when something happens who knows

[loud crag]

Does anyone occasionally get an output image from Nano Banana Pro that takes like 4-5 times longer than regular images, and clearly comes out broken/glitched? It looks mostly the same as the image I put in, but the face and lighting are messed up.

[fickle venture]

On Lmarena?

[loud crag]

Yeah.

[fickle venture]

Huh that's weird

[fickle venture]

This need to be reported on #1343291835845578853

[fickle venture]

Do you have a screenshot of what the broken image output is?

[loud crag]

Actually, no I had it wrong. It only happened on Fal.ai

It might be a 4k thing since I was using Fal for any 4k image and the free one on LMArena for anything 2k, sorry!

[fickle venture]

So that's fal issue

[loud crag]

Probably, my bad. 😅

[thorny schooner]

Considering
I have not even thinked about trying because of the issues going on right now with this website can't really give an answer

[sterile tartan]

@loud crag why Stupid

[loud crag]

#JustSayakaThings

[viscid cloak]

For the broken/glitched part, yes, and that's most likely gemini's problem. I called it calculation power degradation, but I'm sure there is some more precise english term describing it.

[sterile tartan]

Ahhh i dislike when such a kawaii girl is named stupid

[fierce kelp]

Not that I know of but I have not looked into it tbh

[viscid cloak]

It usually happens when high peak/internet setting abnormality/server issues comes up. Many models could have this kind of behaviors.

[fierce kelp]

How does gpt-5.1-search-sp differ from gpt-5.1-search?

[sterile tartan]

https://x.com/shiri_shh/status/2007137787359965423?t=yfTWyMshShF6BxRbP4nSFw&s=33

shirish (@shiri_shh)

AI wrapped 2025

X · Jan 2, 2026, 5:11 PM

[sour spear]

Only one thing missing: Sam needs a bloody nose.

[echo aurora]

Can you give these steps a try? https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message You'll also want to swap to Search Arena for that kind of question, if you click on the little globe in the text chat you'll change to the Search modality.

[wet echo]

hello , i am an AI enthusiast , i want to understand the artificial intelligence indeptly

[queen veldt]

No

[desert abyss]

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

[compact hearth]

Will direct chat be added for video generation?

[desert abyss]

@marsh vector Please check how-to-video-bot to learn how to generate videos. In addition images and prompts must adhere to the server rules. Be mindful of the wording used in the prompts and avoid generating content that is suggestive or inappropriate. Thank you. Failing to follow this rule will result on a permanent ban.

[marsh vector]

I wsnt my crush wirh me

[little ginkgo]

Why this website bugs sometimes 🙁

[fickle venture]

Bro is cooking 🔥 hold on

Guys help llama-13b he is homeless living down there

[echo aurora]

How so?

[little ginkgo]

Recaptcha is annoying me on mobile

I totally select all the correct stuff

Then it says incorrect

And if it somehow sees its all correct

The ai says error generating response

And give me options to retry and clear

And once i hit retry

I am stuck in captcha loop again

:<

Wow ur a pineapple

[ocean vortex]

I wonder if unofficial API endpoint works better than this

haven't tried it in awhile, would be very ironic lol

[little ginkgo]

Its kinda slow

I mean lmarena.ai is Goated

:]

[west lodge]

waiting patiently for nano banana pro 4k to be enabled

its right THERE

[ocean vortex]

Send prompt I could generate it for you in the meantime

[daring rock]

@daring silo Please head to #1397655624103493813 for a detailed guide on how to use the bot

[light flax]

hello everyone , i downloaded BRAVE browser on my pc ,, it seems like i cant open LmArena.ai for some reason .

anyone can help ?

[modest prism]

Guys do you have any anti lazy prompt for Gemini 3 pro. The model is so goddamn lazy

[storm needle]

What api endpoint

[timid hamlet]

it's simply that the page had other plans lol

[acoustic orchid]

mine jsut did this and i lost everything

[wicked sage]

i got a question, is abacus.ai just lmarena but paid?

[fiery gull]

I just F5 until work

[echo aurora]

Gotcha. If you're able to share the Eval ID in this thread that'd be the best way to report this. #1451574502369656842 message

[stray aspen]

Wheres google

[rose sky]

Hello

[red sluice]

@echo aurora There's a big issue with Lmarena's search mode since today. You might consider not counting the votes from today.
On almost EVERY CHAT, it'll stop before it has finished writting. How am I even supposed to vote correctly?

over and over again

[proud bobcat]

Now this

[red sluice]

It happened a lot with grok fast but now it happens a lot with every model

[proud bobcat]

This is what AGI is

[echo aurora]

Thanks fo the flag, looking into now.

[echo aurora]

Any prompts in particular trigger this, or all?

[red sluice]

No idea. Actually it only happens with Grok-4-fast-search, my bad, didn't vote before telling it here.
Ask it to write a wikipedia article for example, prompt it like that:
"Alright, let's make a wikipedia article about the actor "James Austin Kerr" {or any actor, just took this prompt, but could be anything}
I guess you know everything on how to make wikicode. Maybe try to find a fitting infobox. Double check that the URLs you use in the ref tags are leading somewhere (don't change titles of articles!)
Use good sources only. Don't overdo it.
Paste the result in codetag."

And Grok-fast won't have everything pasted

examples from today

but already happened before (last screnshot still voted grok-4.1 btw because he brought interesting stuff before stopping 😛 )

[echo aurora]

Okay gotcha, thanks for the information. Very clear and helpful . Do you know if these examples are from Battle or was it done in Side by Side?

[red sluice]

Battle, didn't try side by side

search mode activated btw, just in case there's a misunderstanding

[echo aurora]

Sounds good

I'm trying to repro in SbS and no luck, will try to get grok models in Battle.

It's worth noting though that we have ways to validate votes to prevent issues like this from swaying the leaderboards unfairly.

[red sluice]

Just in case I'm using Firefox on Windows right now, with adblocks on and ghostery. I don't believe it impacts anything that is generated but yeah giving pretty much all the information I can

[echo aurora]

I got a full answer from grok-4-fast-search in Battle

[obtuse smelt]

oh

[red sluice]

https://streamable.com/5x3j7u
1:37

You can see it start to generate in the beginning but stopping

It's happening every single time with grok fast

and refreshing doesn't help

[obtuse smelt]

i see

[red sluice]

(had multiple tabs opened to test it and be sure to have grok if you're confused)

[bright shard]

@echo aurora When I'm using Gemini 3 Pro in Code Arena, it often outputs infinite text. Do you think that could be fixed?

[obtuse smelt]

hmm i use gemini too

[still jetty]

edit: reloading the page worked without losing the conversation
leaving the question up for future searchers

in arena battle mode, is there a way to resend the prompt? i am 3 prompts in and its stuck generating. i REALLY had a good experience with one, but since i made the 3rd prompt and it got stuck, i cant vote or reveal identity :(

website, safari browser

[earnest rover]

@echo aurora does lmarena has a limit of number of words we can sent ?

[whole sundial]

<@&1349916362595635286>

[echo aurora]

The infinite loading bug is a problem that can happen from time to time unfortunately. Hard refreshing the site may help, I've also seen mentions where clearing cache/cookies helps. Starting a new chat can help if you're really stuck, but sadly that loses the context of the chat.

[robust sluice]

is this happen a lot more these days ? I feel like everytime I prompt anything

just wait like 300-400s then refresh to see both error result, very nice combo

[rigid copper]

hi guys

i guess you're trying to create image or video, you can do it in #video-arena-1

[halcyon nimbus]

looks like image generations down site wide D:

like 5 failed battles but maybe some models are working idek

[echo aurora]

We did report specifically this infinite gen error related to Code Arena early in the week, so it could be related

[echo aurora]

hello

[halcyon nimbus]

might have to sub to chatgpt for a month its actually so good rn even better than nano bp

[echo aurora]

Is it? in Battle?

[halcyon nimbus]

ive only tried battle a few times and chatgpt latest so \o.0/

tried refreshing, new chats, still nothing

[echo aurora]

Seems good on my end

[halcyon nimbus]

mod luck

[echo aurora]

Are you getting the Something went wrong... error message? If so, this article could be helpful: https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[echo aurora]

Yeah true

Yikes, going to remove the image

[halcyon nimbus]

i guess some are working =_=

[whole sundial]

yeah, all I can say about grok is that it is weird, even on api. I had grok 4.1 fast hallucinate so badly on lmarena with an innocent prompt that it generated an nsfw answer, grok 4.1 (the full version) did it just fine, I guess they train the models on a lot of adult stuff

[covert ice]

hello

[echo aurora]

Yeah I've definitely experienced this too. Someone was asking what multi-turn was so I was going to show them with a screenshot with the prompt ~"this is multi-turn" and got a surprising response

[echo aurora]

hello

[halcyon nimbus]

my guess all the image editing ones are down

[echo aurora]

Mod luck again

[halcyon nimbus]

it would be cool if there was a status detector with a little red or green next to each model to say if they were up/down

[echo aurora]

Are you only getting it for image edit? #general message was this image-edit?

[halcyon nimbus]

that wasnt image edit, yeah only having a problem with image edit

[robust sluice]

Qwen, Flux, Reve (not sure of this name) is the 3 AI that always works well, other is a lot error

[echo aurora]

Can you try a different browser or clear cache/cookies?

[halcyon nimbus]

tried a different browser

[whole sundial]

both flux though

[halcyon nimbus]

i guess its just high traffic api overloaded id bet

[robust sluice]

yeah I always got both of them when use Battle

[echo aurora]

A lot of the time the Something went wrong... error is happening due to rate limit, but it should be very rare to happen in battle mode, so this is odd.

[whole sundial]

yeah battle re-route to another model if one fail, i guess in this case all roads lead to flux/qwen/reve

[robust sluice]

I can say when I use Battle, its random only 1 slot another slot is only lock for Flux

[echo aurora]

Oh you'r eseeing this error out in Battle?

[whole sundial]

🤷

[robust sluice]

for me its rarely happen that Flux wont show up, maybe its on my luck

[obtuse smelt]

all user love anime styles in LMArena

[robust sluice]

yes both inf gen time and error is happen a lot these days

maybe cuz too much users or something ?

[echo aurora]

Difficult to say without more info why a specific error is happen. But in that help center article I linked above there is a way to submit a video + recommended software that’ll record the network tab which is helpful info for the team to figure out what’s up.

This one: https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

[robust sluice]

its funny how the site works so well when I try to record video 💀

50s and finish drawing wow

[halcyon nimbus]

shrodingers website

[robust sluice]

when I didnt record:

vs when I record it

[onyx cairn]

When I'm trying to log into my account this error is showing. How can I solve this?

[quartz light]

[obtuse smelt]

well i just kidding

[royal sorrel]

hello

[obtuse smelt]

hi

[neon idol]

<@&1349916362595635286>

[loud crag]

Does the Nano Banana Pro on LMArena use the Web search or not?

[jovial lynx]

hello

[fickle venture]

Grok 4.2 dropped yet?

[livid heath]

Why is the "Something went wrong with this response, try again" message constantly there?
I have retried many times yet it gives this same response. Is that conversation dead, like can't be continued or something?

[obtuse smelt]

apple got slice!

[lean pivot]

👋

[celest cave]

What should I do about it? Something went wrong with this response, please try again.

[rose sky]

Actually really good!

[fickle venture]

Gemini is now on YouTube but when you ask it to tell you something that is on video it will tell you to watch the video to find it so it refuse to answer

[steady grove]

anyone connected game to LLM ?
for ai npc etc

i tried to local ollama VISION, npc to choose where he WANT to go
and it works. lol.
even with details "oh this big blue cube is so big" etc
if anyone interested in technologies / code - dm and free code will fly to your home

[fleet condor]

👋

[hearty ferry]

LMARENA UNCENSORED?

@echo aurora

[lofty shell]

hello

[ocean vortex]

There was one which used reverse engineered API. If that still works it would be extremely ironic needless to say

[ocean vortex]

Not happening, sorry

They want to encourage people to do 'safe' chats that could be analyzed or publicly shared, and for them to use battle mode predominantly direct chat less.

[echo aurora]

Would also encourage you to check out our Terms of Use: https://help.lmarena.ai/articles/5629909088-terms-of-use

[civic spruce]

Is there any subscriptions in lmarena?

[astral dust]

Why do models give incorrect dates bro

@echo aurora

[echo aurora]

Of course that happens Are you still getting this bug?

[echo aurora]

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

[echo aurora]

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

cc @celest cave

[echo aurora]

There is not

[stray aspen]

Wassup gang

[echo aurora]

Yes, you can see the Coding filter for Text Arena with (https://lmarena.ai/leaderboard/text/coding) and without (https://lmarena.ai/leaderboard/text/coding-no-style-control)

[echo aurora]

[echo aurora]

In Text Arena models aren't going to have access to search tools, you'll want to be using Search Arena (click the little globe on the chat bar)

[astral dust]

Should text models only be used for things like writing text?

[echo aurora]

That's a good question, going to move this convo to #leaderboards and ping you there

[echo aurora]

I'd be using Search for when tasks require more recent information that having access to the web would be benefit from. Would check out this blog post: https://news.lmarena.ai/search-arena-update/

[wicked sage]

hey pineapple! this may or may not be internal but is the team friends with any people that hosts ais/popular ais?

like chatgpt, deepseek, glm, whatever

[wicked sage]

not sure about the second one, the first one is because of its info getting cut off on a specific date

[robust sluice]

I just try to clear cache and cookies, the site seems normal right now

[astral dust]

Is the website down?

[golden ocean]

https://media.discordapp.net/attachments/1006341705924087919/1200758925205708870/caption-2-2.gif

[neon idol]

<33333

[neon idol]

[slim loom]

It hasn't always been easy, but we are finally in the home stretch. To keep the momentum going, I’ve set up a live countdown page for our official launch.

https://www.linkedin.com/posts/tkoushik_buildinpublic-productmanagement-careerdevelopment-activity-7413596891512385536-bCfr?utm_source=share&utm_medium=member_desktop&rcm=ACoAAFTV2vkBRU-4pgTUDg7fsP_RhIsZlWT1iME

[wicked sage]

hi gys

[golden ocean]

https://cdn.discordapp.com/attachments/1096556579609911416/1314666961275457608/togif-5.gif

[dapper walrus]

site login still doesnt work(

anyone still have tgis problem?

and sometimes this

[old garnet]

Hello

[dapper walrus]

?

[stray jetty]

..

[thick pawn]

I still can't login, but haven't gotten that message before

[spark python]

does anyone know an llm that doesnt hallucinate after 100k tokens?

[thorny schooner]

Has anyone up else kind of just given up on this for the meantime

[rose sky]

Why is Mr.Beast here?! I didn’t even mention him in the prompt

[rose sky]

I made this

The code is obviously generated by Gemini 3 Flash. I just ask it to write the Arduino code to do what I want

[rose sky]

How's this I generated?

[grim glacier]

hello

[whole sundial]

<@&1349916362595635286>

[wind shore]

.w.

[somber pagoda]

does anyone know when direct chat will be allowed for videos?

cause I see it on the site as in only battle but not the other 2 within dropdown

[onyx cairn]

@echo aurora I did the steps you told me to do but it's still not working. Also this message is keep showing

[burnt pulsar]

I see that error from time to time, but recently I only get a single AI model in direct chat (thankfully it is Claude-Opus-4.5-Thinking-32K) and cannot select anything else any longer. Is this a known issue?

[verbal nimbus]

Cool 😎

[dapper walrus]

hmm

[fossil socket]

Is it me or is the limit on Claude way short now?

[crude anchor]

hello guys

[uncut needle]

AI models are tuned to win benchmarks.
But what if the benchmark is different?

Prompt Arena lets AI models face the same strategic problem —
no memory, no fine-tuning, no shortcuts.

Some models plan.
Some fail.
The differences are obvious.

If you’re into AI, reasoning, and real model behavior —
check it out 👉 https://prompt-arena.com

AI vs AI. Real outcomes.

[fossil socket]

Uh any reason?

[lime void]

Hello chat
is there a way to use a start frame and end frame while generating a video transition here?

[golden ocean]

https://media.discordapp.net/attachments/1006341705924087919/1200758925205708870/caption-2-2.gif

[slim spire]

Hello everyone 🤗

[obtuse smelt]

hi

[obtuse smelt]

Something went wrong with this response, please try again.
is seem error

[hollow panther]

hello

[bronze gull]

Hello

[fossil socket]

Ok anyone here knows why the Claude rate limit is way shorter than yesterday?

@echo aurora

[gleaming roost]

I believe part of the rate limit has to do with the influence of the number of tokens, but I could be wrong.

[thorn nebula]

why is claude limit soo short now

[sour spear]

Because y'all been generating code like crazy. 😁

[golden ocean]

real

[iron laurel]

Hi. Is claude-opus-4-5-20251101-thinking-32k really Claude Opus 4.5 ?

[gleaming roost]

🤔

[icy frost]

yeah if you ask it it will say other model because it doesn't know itself

[iron laurel]

Oh I see. https://livebench.ai/#/ doesn't it appear here?

[versed marlin]

good thing

[icy frost]

ur msg is flagged as spam

[gleaming roost]

🕵️

[sour spear]

Holy crap, the OpenAI discord is a mess. They just removed about 50 messages from their openai-chatter channel for going off-topic, and also conveniently removed posts in the process that were fully on-topic - but critical of the product's current state. That stinks of censorship. 💩

The discussion only started to derail towards the end, but they obviously felt the need to delete way more posts than necessary.

[limber panther]

cuz its business

[fickle venture]

Yes it is but the thinking-32k is better than normal opus 4.5 there is thinking 64k it's overpowered

[iron laurel]

Is it better than Claude 4.5 Sonnet Thinking 64k?

[echo aurora]

Good callout, I believe that's just what it's called via API.

[fickle venture]

I think it means extend thinking by 64k tokens

So it thinks longer for better results

[iron laurel]

I mean : is Claude 4.5 Opus 32k better than Claude 4.5 Sonnet 64k?

Because I'm not able to find benchmarks

[fickle venture]

There is no benchmark for does I remember seeing a image showing it but I can't find it

[obsidian cargo]

I got permabanned from their discord without appeal for generating a portal made of meat back in the dall-e 2 days

[civic ermine]

Hello

[echo aurora]

Hello

[civic ermine]

I'm new here

How do I generate video

[sour spear]

I received a warning for posting a pegi-13 "horror" image I had created with gpt-image, in ChatGPT, with a European IP address (=censored into oblivion anyway), and without any jailbreaking. Completely harmless, but apparently still too offensive.

[desert abyss]

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

[inner gate]

hello guys

[true vigil]

Hi

[quartz light]

opus 32k is undoubtedly the best model ive ever used

by far

[hard rover]

Hello

[fickle venture]

There is 64k which is more better than 32k

[proper jacinth]

Is there an Android app?

[acoustic orchid]

anyone know why i keep getting Something went wrong with this response, please try again.

when i try and do something please

[acoustic orchid]

i think there is

[echo aurora]

Sorry to hear this, could you try the steps in this article: https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message ?

[next hedge]

👋

[acoustic orchid]

thanks i give that a try

[sage hinge]

Guys, please help, upload the video generator to the website, please! There's a child crying here at home asking for the video generator, are you going to let an innocent child cry? 😭😭

[echo aurora]

Sorry for the crying child!! Video Arena on the site is currently an experiment so it'll be random if you get it or not.

[proud bobcat]

qwen image peak

[ionic lark]

gang i joined ts server cuz of dis vid

how do u get the ais

[echo sinew]

Hello! You can read the info posted here 👉 ⁠⁠https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to generate videos or images using the bot.

[lean mortar]

Hello

[lean mortar]

You need to go video arena

[quartz light]

yall removed reactions from announcements? 😭

😔

[quartz light]

i knowwwwww

but i mean we dont have it on lmarena

so i havent been able to use it

[sage hinge]

oh okay my friend, thank you! the child calms down 😂

[meager harbor]

hey

[meager harbor]

it's over for the goofy ass reaction, I will miss them

[stray aspen]

qwen image sucks

[worldly copper]

maybe its because we are used to great models like nb pro

as nothing close comes to that

[stray aspen]

nano banana pro leads the way

[quartz light]

no more

😔

[thorny schooner]

stupid verification always do have to ruin my day

[pure swan]

I am asking are the free version in lm arena weaker than the actual one in the apps like grok Claude etc

[west lodge]

hmm whats this

HMMM

oh new feature

soon™

[pure swan]

[fossil socket]

Any reason why they literally shortened the rate limit for Claude by 75%?

5 prompts then waiting for an hour is way too short

[echo aurora]

(Already responded in the thread, but will respond here incase others are wondering the same). Overall, these rate limits can change. I'm not aware of this change, but it's possible. Checking with the team to confirm this is the rate limit and not some bug causing it.

[weak crane]

old photo of it btw

[quartz light]

1984

[old garden]

no more 🥀

[quartz light]

@old garden

js like me fr

[remote sun]

????????????????????

[old garden]

What’s so confusing

[quartz light]

its battling in direct mode

[old garden]

@quartz light why is he confused

Ohh

[weak crane]

hi

[stray aspen]

how is bro doing battles in direct chat

[echo aurora]

It's an experiment we're currently running.

[next hedge]

👋

[peak rock]

How to create vidos on LMArena?

[echo aurora]

You can use the Discord bot here in the server, there is more information in #1397655624103493813 . For Video Arena on the site this is currently an experiment so it's going to be random if you get this modality on the site.

[river star]

Hello, just sneaking around

[keen beacon]

Hello my dear! Do you know if the mails [e-mail] and [e-mail] is working ?

[echo aurora]

Hello - yes, both of those are working. The holiday break may have caused some delays in response, but we are back now so would expect a response soon. If you'd like to escalate a specific ticket can you send me a Direct Message including the email address that reached out.

[keen beacon]

What is the expected response time

?

[echo aurora]

We don't have an ETA on this. I do see your DM and will get back to you shortly.

[keen beacon]

Thankyou sir

[wicked tapir]

for the january 1 contest are all lmarena modalities eligible?

or only some of them?

[echo aurora]

Nope, only vision.

[polar brook]

I used AI to help practice test fight scenes

[sage hinge]

@echo aurora Hey, do you guys plan to officially launch the video generator on the website at some point? I know it's an experiment and this is a silly question, but I'd like to know if it's something you plan to continue.

[buoyant jacinth]

Just practicing.

[echo aurora]

Well the intent of the experiment is to see what Video Arena on the site would look like. Assuming the experiment goes well and signs point to it being a good thing to ship, then we'd do it. But I won't be able to make promises or give estimated dates on when/if to expect something like this makes it to the site.

[sage hinge]

Nicee, thank you my friend! God bless you!

[frank adder]

How to test for particular mystery or anomalous model on llmarena

[echo aurora]

The code named models you'll only come across in the Battle mode of each modality. It's going to be random if you come across these.

[frank adder]

How to get continuously without random

[echo aurora]

You're unable to. It's going to be random.

[frank adder]

How to moderator or admin role here

[echo aurora]

Gotta be a moderator or admin.

[frank adder]

Yes so much interested

[echo aurora]

We aren't currently looking for mods/admins for the server. Thanks though!

[frank adder]

Atleast intern or any other

[echo aurora]

Our job openings can be found here: https://jobs.ashbyhq.com/lmarena

[crystal blade]

hey guys im getting this issue where it is only generating square image aspect ratio

no matter i put in the prompt ill get back a image that is square

wow i didnt know there was a video bot

[late briar]

guys how to fix inf generating chat ive been waiting for a whole week

[rigid copper]

sup guys

[viscid timber]

What's the best model for understanding videos?

Like best vision model that can see videos

[vast sapphire]

hi

[rigid copper]

for #january-1st-contest how many entries i can send at max? since in announcement or anywhere they did not specify how many entries you can share with.

nevermind, i guess 1 entry only

[rose sky]

Damn BOY!

[rose sky]

This has to be the most hardest drop ever. WARNING: EXTREMELY LOUD EAR-RAPE!

[jovial solar]

No way qwen update video 5 second to 15 second

[visual raptor]

guys for the video generation @ what command

[visual raptor]

which qwen model and where

[copper folio]

#1397655624103493813

[rose sky]

Sorry for the late reply, but you can go to, "https://create.wan.video/" and sign in to your account, Google or whatnot, and start creating! You can mention any existing people or artists, it will accept. But make sure you select and use Wan 2.6 video model, the latest one, because that one has audio generation baked in too

Think this is AI-generated or real?

[rose sky]

Anyone here knows Sonauto AI?

[next coral]

hello guys

[prime umbra]

I understand that you want to integrate battle mode into direct chat. However, please leave some space between them. Having a message immediately followed by battle mode makes no sense. There is a reason this is called direct chat and not battle mode. Please show some consideration.

[zealous sparrow]

Btw grok 4.2 is now gone from lmarena all the stealth modeld

Unsure if the one on design arena still exists

[keen beacon]

Is support working today?

[ocean bison]

Maybe that changes...

[high dirge]

Is the video gen coming on the site since I saw the button for it but it vanished on next refresh

[halcyon nimbus]

is the open ai site more censored than lmarena? waste of money so far

cant make the usual established ips or even bikini stuff

[ocean bison]

Wan 2.6 also brings a Cameo-like feature.

You can upload a clip of yourself and insert a "Cameo" of you in a new generation!

Much like Cameos on Sora 2!

But you don't have to be on iOS and USA.

It's called Starring Roles!

[sour spear]

It is. Also depends on where you're located, Europe and UK for example are heavier censored than the rest of the world, I think. But OpenAI have a track record of neutering their models. They do release new models, image & video in particular, with very low guardrails, so early adopters generate and share all kinds of cool stuff on x to build hype for them. Then, after a couple of days, they increase the censorship so you can barely generate anything that's not totally, 100%, HR-approved safe.

It's incredibly obnoxious. The sub-par GPT-5 (5.1 / 5.2) release added to that, and finally made me cancel my sub after three years, and switch to Gemini.

<@&1349916362595635286>

[ocean bison]

Reported!

[ocean bison]

I think that Alibaba copied OpenAI

[sterile tartan]

Lol what!?

[sour spear]

Benchmaxxed

[sterile tartan]

Ik

[exotic tartan]

Hey everyone! I worked really hard on a completely free, open source LMArena Chrome extension called LMArena Plus.
The intention was to add more context to the leaderboards by adding new columns (pricing, bang for buck, supported modalities etc), a column picker, optional notifications for when generations are ready for voting and you're on another tab, and there's more to come!

Just google "LMArene Plus" and download from the Chrome extensions store. Any feedback or requested features are welcome!

[stone cape]

came across this yesterday some chinese company, seems to be another iquest lab benchmax?

[sterile tartan]

Can it work in English?

And where's The Website to Use it?

[stone cape]

it seems only chinese i think

https://x.com/xiamenlabs

XiamenLabs - 厦门实验室 (@xiamenlabs) on X

厦门实验室

X

[sterile tartan]

Then useless for us for now lol

[stone cape]

nvm it can speak english but for some reason it likes to think in chinese

[sterile tartan]

With all honesty benchmarks are Deceiving

[ocean bison]

Maybe put in the system prompt?

[sterile tartan]

Test and Experiment Yourself

[stone cape]

i said why do you think in chinese

[sterile tartan]

You can actually use browser translation

And it will translate the chinese thinking to english

[stone cape]

good point

[sterile tartan]

If this is actually a Good Model we will see it somewhere on Artificial Analysis Index, LMarena, Yupp Ai

[stone cape]

any hard prompts to test it or break it?

[ocean bison]

And Ernie 5 Preview also thinks in Chinese (mostly)

https://ernie.baidu.com

[sterile tartan]

I think those are available on Google if you search it

[ocean bison]

But it CAN think in English sometimes.

[sterile tartan]

Unity got 55% in Humanity's Last Exam

No Way

💀

Dw if it's actually a Good Model it will get Hype

[ocean bison]

Try to one-shot a Minecraft clone!

[stone cape]

good idea

[ocean bison]

Or do the pelican with a bike SVG test 😂😂😂

[sterile tartan]

Pass those to @stone cape

He seems interested

[stone cape]

running it however chinese have potato gpus😂

[sterile tartan]

Well is still a new company

💀

[ocean bison]

Unity beats Gemini 3 Pro??

[sour spear]

Trying it out right now with a suite of coding prompts I throw at every model. So far it seems decent, but not particularly mindblowing. Generated a good result, then broken code in the next test. Judging by my first results, it can't compete with Claude or Gemini.

It's really good though, gotta give it that much credit

[stone cape]

it works but pointers and wsad movement is a bit wonky (2 prompts) first time it fell thru world

[ocean bison]

Can I try it?

[stone cape]

its a bit slow

but doable

[ocean bison]

Where's the site?

[stone cape]

https://xiamenlabs.com/

[ocean bison]

Do I need to create a account?

[stone cape]

nope its all free i think depends on where you are not sure

[ocean bison]

Thought in Chinese but responded in English.

[sour spear]

Seems to be completely free for now. It's dead slow though, I'm getting ChatGPT vibes looking at the sluggish code generation. 😂

[ocean bison]

Does it save your chats locally?

@stone cape

[stone cape]

bro i don't know chinese 😂 look around

it does not seem to have chat saving

we have agi

[stone cape]

whoa wtf is this unity file?

i'm not opening that

[ocean bison]

Message from Unity from Xiamen.

[stone cape]

😂

[ocean bison]

It's a text file.

😂

[stone cape]

no you sent it as .unity

not text 💀

[sterile tartan]

That's Actually Pretty Good

[ocean bison]

First is my message, then thoughts, then response.

See if it's AGI enough.

[stone cape]

this don't work for me

[ocean bison]

Did you get a error on Console?

[ocean vortex]

https://tenor.com/view/brindis-xi-jinping-gif-26069380

[ocean vortex]

Fun fact you can make most Chinese (most OSS?) models think in numerous languages

[ocean vortex]

I made Speciale 3.2 do that in Polish the other day lol

[sour spear]

It's a decent model, but it's making stupid, weird mistakes, like this one:

case 'Digit1': selectSlot(0); break;
case 'Digit2': selectSlot(1); break;
case 'Digit3': selectSlot(2); break;
case 'Digit4': selectSlot(3); break;
case 'Digit5': selectSlot(4); break;
case 'Digit6': selectSlot(5 break;
case 'Digit7': selectSlot(6); break;
case 'Digit8': selectSlot(7); break;
case 'Digit9': selectSlot(8); break;
}

The missing ); after selectSlot(5 is really dumb. Not to mention that the whole code was still broken even after fixing that manually.

[ocean vortex]

You only need to have system prompt in Polish but it can be literally anything. Without directly instructing it on how to do reasoning

and then you just write in Polish

and it starts reasoning in Polish

[sour spear]

Seriously?

[ocean bison]

Xiamen Unity Thread

[stone cape]

how did you copy the code 😂

[ocean vortex]

To be fair OpenAI would have been able to copy-cat competitors with a small fraction of the cost as well if a reasoning model like that was already available prior to o1 👀

[sour spear]

Downloaded the html file, imported it into Antigravity

Gemini 3 in comparison.

[ocean bison]

Go to #1458069911346876469

[ocean bison]

"propertyations" - Is that a real word?

[stone cape]

i would like to see this in api

wonder how its like with tool calls

[sour spear]

The more I test it, the less impressed I am tbh. You're probably better off using Gemini 3 Flash instead.

Those benchmarks are fake af for sure.

[sour spear]

Let's face it, this model is mid and can't code properly.

[ocean bison]

✅

[stone cape]

asked gemini 3 pro to make minecraft

@sour spear i think the prompt you gave g3 had the code made from unity?

[sour spear]

My prompt was "Code a Minecraft game clone within a single .html file. Make it beautiful, with pixel graphics like the original, add all main features of the game, terrain and tree generation, mobs, and a bunch of other stuff. Do as much as is feasibly possible."
Just simple html.

[sour spear]

Your mileage may vary though. AI models aren't deterministic, and even the best model can produce broken results occasionally. That's why I wouldn't judge the Xiamen model from the very first test, either.

[stone cape]

true

[rose sky]

Try, https://chat.z.ai/. It excels at coding and it’s made for it

[sterile tartan]

So what's The Final Verdict?

For Unity

[stone cape]

bro we know about glm

[sterile tartan]

Ngl i don't like GLM ui

[stone cape]

idk it seems good, but it needs more testing more people

[rose sky]

Ok, fine, I’m just telling

[sterile tartan]

I assume in Couple of months they will optimise and Launch it like Xiaomi Nimo

Is still in Initial Testing

[stone cape]

what does the modal say
Reviewed 1 source
The modal explains the terms for trying the Unity model public test and asks you to confirm before using it.
​

Main points in the modal
Web preview only: The Unity model is currently only available as a web page preview and does not provide any commercial API or external calling service.
​

Feedback collection: The public test focuses on collecting high‑quality user interaction feedback to improve the model’s reinforcement learning alignment.
​

Human review: Your prompts may be randomly assigned for anonymous manual review by researchers to improve safety and outputs.
​

Privacy warning: You are asked not to include personal privacy, company secrets, or other sensitive information in conversations to protect data security.
​

Consent button: At the end, it asks you to confirm that you have read and understood these terms and to start using the Unity model.
​

[sour spear]

It's alright. Not great though, pretty mid and buggy at coding. And they're putting out ridiculously fake benchmarks.

[stone cape]

could be benchmaxxed

[ocean bison]

Like GPT-3.5?

[stone cape]

arena worthy?

[sour spear]

Nah, way better. GPT-3.5 couldn't even string together working Hello World code. 😂

🤣 🤣 🤣
Doing some image arena battle tests: "A hyper-realistic close-up of a person’s face. Their left hand is pinching their own right earlobe, while their right hand is gently pulling down their lower left eyelid. The fingers must be distinct and correctly attached to the respective arms.

[ocean bison]

Hunyuan: Trash?

[stone cape]

wth😂

[sour spear]

It's not too bad usually, but it kinda misinterpreted the "gently pulling down their lower left eyelid" part. 😁

I'm running some tests with prompts that aim to break, or at least challenge, AI image generators. Results are usually quite funny

[sterile tartan]

Noted, Noted

[sterile tartan]

Also Noted

[sour spear]

It gets even better. 🧐

[ocean bison]

😂😂😂😂😂

What about Gemini 2 Flash Image Preview?

[sour spear]

That's Nano Banana Pro and GPT-Image 1.5. The latter turned earlobe pinching into earlobe squashing, and both confused left and right hand, which was the key test of the prompt. But at least the didn't mutilate the person in the image.

[ocean bison]

(The old Nano Banana)

[sour spear]

Lower overall quality, and the guy seems to be married twice, but it's not too bad

[ocean bison]

Like not as trash as Luma's Photon?

[sour spear]

Photon is obviously a "classic" diffusion based image generator like Midjourney or Stable Diffusion. It can't understand complex image compositions, these tools are only good for "generate subject x wearing y with a backdrop of z".

[ocean bison]

Do DALL-E 3

[stone cape]

@sour spear 💀

[sour spear]

That explains the stupid mistakes, I hope. Back to some more testing then. 🙂

[grave plaza]

lmarena is nice

[sterile tartan]

Please let me know when you have more verdicts ty @sour spear @stone cape

[stone cape]

testing

[stone cape]

you know what its actually pretty insane

seems like stupid code mistakes are gone

[oak python]

ModMail Help Menu

ModMail is a feature-rich Discord bot designed to enable your server members to contact staff easily.

Please direct message me if you wish to contact staff. You can also invite me to your server with the link below, or join our support server if you need further help.

To setup the bot, run =setup.

Invite

https://modmail.xyz/invite

Support Server

https://discord.gg/wjWJwJB

Use the reactions to flip pages. (Page 1/8)

[sterile tartan]

https://tenor.com/view/mind-blast-shocked-omg-wow-mind-blown-gif-5194406635956370896

Yeah i Trust The Chinese

Just imagine how great it will be after proper release

[stone cape]

i wana hear @sour spear

[keen beacon]

Is working today staf that he can delete the website created with arena

[sterile tartan]

I wonder when we will Revolutionize Further from LLMs

[ocean bison]

🤯

[prime umbra]

I have a small request. Nothing big. You know, this is a direct chat, not a Direct X battle chat. Maybe use the battle mode less often. I’m not saying to remove it—just less

I don't know when i can wrrite this. this not bug and not model-request idk where send this

[ocean bison]

Feedback??

[sweet cove]

Hello

[prime umbra]

this... this is good point

[sterile tartan]

👋🏻

[sweet cove]

You are active here currently?

[sterile tartan]

Many Times Yeah

[sweet cove]

This server contains AD 😅😂

[sterile tartan]

@echo aurora

[sterile tartan]

Scammers not From Server

[ocean bison]

<@&1349916362595635286>

[sweet cove]

@hardy swallow

[ocean bison]

And I reported as spam!

[sweet cove]

Yeah I know it

[ocean bison]

Promoting USDT crypto ads 😂😂😂

[sweet cove]

@sterile tartan did you still practice the AI thumbnails?

[sour spear]

It's still making stupid mistakes, and that has nothing to do with Unicode. Like this:
const const z = (Math.random() - 0.5) * 40;

I fixed that myself, so I could launch the app, encountered another bug, gave it back to the model so it could fix it, and it produced even more errors in the process.

It did produce a working 3D benchmark app, but veeeery barebones, and not really following my prompt instructions. Earth animation test (with weather simulation, country information etc.) only worked insofar as there was a textured globe with some clouds, but all else didn't work and threw errors in the console.

So my verdict still stands: it's a good model, with decent capabilities, but not really up there, and certainly nowhere nearly as good as they're claiming.

[shell gale]

Hello, everyone, I'm here to learn. Hope I'm welcome

[foggy nest]

claude is the best coding ai the opus 4.5 model right

fr4om all of them

chat gpt 5.2

gemini 3 pro

grok 4

[neon idol]

<@&1349916362595635286>

[ocean bison]

Imagine if the user sends another ad 😂😂😂

[sweet cove]

😂

[echo aurora]

I'm ready this time

[echo aurora]

See!

[ocean bison]

Imagine if the user sends YET another ad 😂😂😂

@echo aurora is the expert, he cleans spam!

[echo aurora]

Don't jinx it

[keen beacon]

Dear, could you check DM :

?

[echo aurora]

[ocean bison]

He always cleans spam!

[royal coral]

anyone else think there chat gpt's have been super slow and laggy lately ? i use mine on windows desktop app and the browser (same issues)

[echo aurora]

Can't say I've been hearing similar reports.

[royal coral]

so weird, idk what the issue is

[sour spear]

You mean, even slower than usual? I just tried, and it's just as painful to watch it slowly generate code as always. Though if it were any slower, it would be coding backwards.

[royal coral]

no like my whole ui is super laggy

and its defiently not my pc or internet

not sure what the issue is

[sour spear]

Only ChatGPT? Have you tried another browser?

[royal coral]

idk ill figure it out i guess

u think claude worth getting into ?

whats best all around u think ?

kinda new to all this but tryna learn and have fun since use my chat gpt pro 24/7

[sour spear]

Best allrounder model is Gemini 3 without a doubt. it's very good at everything, best at most, very fast and has very generous usage limits. And Google also give you 2 TB of cloud storage and tons of goodies, simply because they can afford it.
If you're into coding, Claude is definitely the best. 4.5 Opus is just on another level compared to all other models.

[echo aurora]

Are you seeing this when you use a different browser?

[sterile tartan]

A little bit

[sterile tartan]

Damn it

Hopefully it will get Better

[ocean bison]

He always cleans spam!

@echo aurora

[dapper walrus]

i yhink site is overloaded

with requests

now he even refuse to fully load

[royal coral]

i ordered a pixel 10 pro xl to start using as new work phone with the hopes of it being extremely lucrative for me since all my companies use google workspace

does that make sense and think the 2 together would crush ?

[dapper walrus]

u have that problem?

[sour spear]

Doesn't the Pixel Pro come with a Gemini subscription "built-in"? Not quite sure, but in any case, Gemini Pro on Android phones is dead useful, as Google integrate it into all their apps.

[royal coral]

comes in today so lets see

but ye is goal

is chat gpt pro realyl not the top ?

[sour spear]

No, not anymore. Check out the LM arena leaderboard. They only top the image generation & image edit leaderboards, but even that is astounding to me, considering the grainy artificial look of the images it produces.

Nano Banana Pro (Gemini Pro image gen) can also do any aspect ratio, GPT-image can only do 3:2, 1:1 and 2:3

[echo aurora]

If you swap to cell data instead of wifi, are you seeing a difference?

[ocean vortex]

Sure but overwhelming majority of people can't adjust aspect ration with nbp

[obsidian hawk]

Sites fine for me

[dapper walrus]

what?
i tried on 4G and Home Wifi

[ocean bison]

Prompting

[dapper walrus]

dunno what celling data

i will try on PC

[ocean vortex]

Hmm, actually maybe..

if aspect ratio is on 'auto' and not hardcoded to anything else with API parameters then that may work

[sour spear]

I was talking about the actual Gemini app on the phone, or when using https://gemini.google.com/. You simply tell it which aspect ratio you want, and it will do it. Even weird ratios are no problem.

"A transparent glass cube. Inside the cube is a smaller wooden cube. Inside the wooden cube is a small red ball. The glass cube is being held by a giant robot, while a tiny bird sits on top of the red ball inside all the layers.
Aspect ratio: 18,5x11"

Image resolution is 2688x1600, which you can reduce (maintaining aspect ratio) to 185x110. So it's pixel perfect, even though 18,5x11 isn't used anywhere as valid aspect ratio.

Another major benefit over GPT-Image is that the Gemini app produces images at 2K resolution (2048x2048, comparable numbers for other aspect ratios), while ChatGPT is lower res.
Here's an example, same prompt as above, from ChatGPT image. It's only 1536x1024, and it fell back to its 3:2 aspect ratio, because it can't do anything else in landscape. Not even 16:9

Plus the ball is wrong, the bird's feet are wrong, and the whole image looks artificial.

[neon idol]

and the bird have a line at least of the eyes lmao

[sour spear]

Don't want to spam the channel, so I'll stop after this next example.
"A transparent glass shelf holding exactly six identical porcelain teacups arranged in a perfect straight line. A seventh teacup is levitating exactly two inches above the third cup in the row. Minimalist studio background."

First one is ChatGPT image, and it failed to count the cups correctly. The Nano Banana Pro image has better lighting and shadows, and if you look closely, you can see that it even rendered a reflection of the cups directly below and to the lower right on the floating cup.

[ember iris]

Hi guys

Is website having issue? Can't open the website. It's stuck at half way loading.

[stone cape]

made with unity

uh oh their gpus are fried

[echo aurora]

Sounds good, please keep me updated on what happens.

[sour spear]

That's really good!

[shrewd citrus]

wow that valuation sounds good

so how does lmarena make any money?

[thorny haven]

congrats on the raise!

[warm zodiac]

presumably we are basically an RL environment for the labs

[neon idol]

Sponsors maybe?

[dim ivy]

"Today, we're excited to announce our $150M funding round at a post-money valuation of more than $1.7B" they couldn't afford gpt-5.2 pro xhigh they said.

[neon idol]

And investments

[hollow imp]

@alpine pasture

What does that mean

[empty vector]

Congratulations to the team! LMArena is so important 🙏

[thorny haven]

https://news.lmarena.ai/ai-evaluations/

they sell eval services to ai labs that measure model performance for users across industries, also the anonymous slots on the public arena

[crystal rapids]

I wonder why LTX-2 is not even on the arena, it just open sourced today. First open source video gen with native audio, most likely the best open source video gen out rn

[restive steppe]

Do raters get some of that big money

Top raters sweepstakes?

[spark python]

Does this mean lm arena gets big money for this

[sour spear]

The "money" we get on this platform is the ability to use it all for free. Although I have to admit, the idea of rewarding people who actually rate instead of just freeloading AI tools is compelling. Maybe by giving higher rate limits to registered users, or something like that?

[maiden jackal]

Congrats LMArena. You deserve it. !!! Great work

[ionic sonnet]

the feedback probably needs to be high quality for that

[scenic salmon]

@steel bridge @alpine pasture Congrats on finishing series A

[cyan moat]

They finished Series A?

[scenic salmon]

Check #announcements

[cyan moat]

ohh congrats yay

[restive steppe]

They could give you a feedback quality rating 1 to 5 stars and have a lottery for 5 star raters

[timber prism]

make a car video

[echo aurora]

You’ll want to check out the information in #1397655624103493813

[gleaming roost]

[queen veldt]

Great you got the money now fix the website

[night moat]

helo. I'm new. How to use image to video? Why no response

[echo sinew]

Hello! Please read the info posted here 👉 ⁠⁠https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to generate videos or images using the bot.

[night moat]

Okay

[stone cape]

guys my work

with movementlabs new upgrade my jaw hit the ground

this is INSANE, NOT EVEN HYPING

prompt: minecraft clone in html css and js ultra realistc add real water ponds too

model: hawk max

https://tenor.com/view/atria-mystery-bus-omg-omg-gif-omgg-jaw-drop-gif-25226119

[golden ocean]

"MY work"

[stone cape]

well 😂

[golden ocean]

jk, thats impressive

im proud of u

[stone cape]

many hours typing 💀

[golden ocean]

real

[dapper walrus]

i see n9w

m0bile version didnt work
from chrome

[quartz light]

investors n stuff

[proper jacinth]

Can this AI generate images?

[gaunt spade]

stop sharing it here

any abusers will get their IP banned

and potientially hwid ban

[gleaming roost]

[ember iris]

Am facing issue with site, failed to accept terms...
Anyone else facing it

[sterile tartan]

Seems like Highly Optimised for Coding

[sterile tartan]

@swift oyster where are You!?

[leaden vine]

Does anyone know how to get the video function working on the lmarena website?

[swift oyster]

https://tenor.com/view/hello-hi-hy-hey-gif-8520159980767013609

[sterile tartan]

How's The Progress?

[sterile tartan]

This guy has some Great Feedback

[echo aurora]

Hey sorry for the delay, would encourage you to check out this blog post if you haven't ready it yet: https://news.lmarena.ai/ai-evaluations/

[echo aurora]

It's currently an experiment, so it's going to be random if you get the feature or not.

[fair elk]

Hey can u make so u can send photos to the ai to understand better

[echo aurora]

Can you elaborate a bit more here? Image upload is currently available for both text (vision) and image (image edit).

[fair elk]

For example in direct mode when a error happens that I can't like copy or can't explain a bug u can't send images to help the ai

@echo aurora

[spark python]

what is the difference from nano banana pro 2k and nano banana pro

[thorny schooner]

does anyone know when they are gonna actually do anything about the captions /verification thing?

[frigid tusk]

why does the chat suddenly pass from direct to battle ?

[fickle venture]

Too many questions to answer

[fickle venture]

They are the same but the quality is better there is normal and 2K and 4k and 8k

[fickle venture]

We don't have information about it ask the team that work on Lmarena

[fickle venture]

Wdym do you mean that if you make new chat or refresh it becomes a battle mode? That's how it works

[frigid tusk]

no in same chat , its weird , it doesnt turn battle but like doesnt allow me to change model and just put the battle tho

[fickle venture]

Cuz you already talked in direct so you can't do battle you gotta make a new chat.

[frigid tusk]

i dont want battle. it comes by itself while im in a direct chat

[fickle venture]

Then that's a bug

[frigid tusk]

yea

[fickle venture]

Report it in #1343291835845578853

[tall vapor]

congrats on the raise!! huge

[echo aurora]

This is actually an experiment and isn't a bug. We're experimenting to see what the occasional Battle appearing in a Direct mode would look like.

[echo aurora]

We're always going to be addressing how the captcha system works and make the appropriate changes. For now, the best thing to do is flag the Eval ID in #1451574502369656842 .

[thorny schooner]

?

[echo aurora]

Sorry I hit enter too soon.

[thorny schooner]

☠️

ok

[frigid tusk]

my opinion, i dont like it

see with others tho

[echo aurora]

Can you tell me a bit more about why?

[frigid tusk]

well its pretty annoying , i want to stay in the same model and i never use battle anyway so thats basically why

[carmine rock]

Will we one day be able to compare models of music generation, voice-over, and lip-syncing ?

[echo aurora]

Thanks for sharing more, it is appreciated.

[echo aurora]

A Music/Audio Arena would be really great to see on the site! It's a possibility.

[frigid tusk]

i dont really have more to say , my only issue is that i just want to stay in same llm , at least u can add where it asks before changing to battle so we have a choice

[echo aurora]

Hey this may be related to a different bug that we recently fixed, going to follow up with team and get back to you if it's related. Connecting to Arena has failed. Please try again later or on a different device isn't very common of an error message to see. If you try a different browser, what happens?

[fervent bridge]

hi

Hi, do you have any APIs?

[echo aurora]

Hello we do not have an available API.

[blissful lark]

Hello everybody I'm new on this server and I'm happy to be there with you

[echo aurora]

Welcome welcome!! What brings you to LMArena?

[blissful lark]

Thank you.
I'm curious and I want to learn more about AI

[hollow ivy]

From which world region do you come?
(i'm from europe)

India?

[golden ocean]

hes from india

[hollow ivy]

-# (MENA: Middle-East/North Africa)

[stray aspen]

Why are you asking this

[hollow ivy]

curiosity / community spirit

[stray aspen]

Alrighty

[astral elk]

Good night community

[winter frigate]

captcha issues :(

[echo aurora]

Can you flag the Eval ID (random set of numbers/letters in the URL) in #1451574502369656842 ?

[winter frigate]

yeah sure

[echo aurora]

Hello - I'm trying to get more information on a specific bug. If anyone is encountering an error that seems related, can you let me know?

Run 10 prompts, and the scrolling turns very laggy, even after it's done generating.

[pale obsidian]

probably true

[left tinsel]

is it ai?

cool

[rose sky]

Yes

[left tinsel]

can u teach me how u did it?

[rose sky]

I just used https://sonauto.ai/. Just sign in / sign up, create a project, go to Simple Mode, and then literally just ask it, "A song in the style of Billie Eilish". It doesn't reject existing artist names, it just proceeds

It's better than Suno, for me. And it's also free and unlimited and the songs you generate can be used commercially without needing to pay them

They said it themselves in their T&Cs

[echo aurora]

Hello a codenamed model can be removed for either. It is possible for a model lab to request that their model be removed. Model codenames are changed to the actual model names once they exit pre-release testing and launch.

[rose sky]

But the downside is that, it can generate a maximum of 1 minute and 35 seconds in one shot, but you can extend it further

[left tinsel]

Oh man, this is going to be so useful for me! People in my country absolutely love music, and this is exactly what I was looking for. Once I get rich, where should I ship your Porsche?

[left tinsel]

Hey admin, is everything okay? I’ve been getting the 'Something went wrong while generating the response. Please try again.' error for the last 10 minutes. I’ve been on this site for 14 hours and this is the first time this has happened to me. Nothing I do seems to fix it, and it’s really frustrating.

[rose sky]

Excuse me, "ship my Porsche" ??? Wdym?

[left tinsel]

I was just kidding about gifting you a Porsche haha.

[rose sky]

Oh ok

[left tinsel]

https://tenor.com/view/bocchi-the-rock-bocchi-nijika-anime-girl-please-gif-12600961699026251307

[echo aurora]

Oh no! I'm really sorry to hear this! Can you give the steps in this article a try: https://help.lmarena.ai/articles/1645798556-lmarena-how-to-something-went-wrong-with-this-response-error-message

The first part of the article is steps to try that may help, the second section is how to report further information that is helpful for our team.

I’ve been on this site for 14 hours and this is the first time this has happened to me.
Based off of this, what you're liking seeing is you're hitting the rate limit. Each model is going to have it's own rate limit, and users will start to see Something went wrong when they're hitting this limit. @left tinsel

[left tinsel]

Dude, this is truly terrible. I want to be like Sonic; I should be active for 144 hours, not 14 haha xd

I'm declaring you the admin of the year; I'm so happy to have received such a quick response

[echo aurora]

Thanks! Don't hesitate to ping me if you have any other questions or problems.

[loud verge]

So...did anyone answer this question yet?

[rose sky]

Made with Suno

[hollow flicker]

hey pineapple because in the where your prompts go it says its shared publicly but its anonymous if we accidentally share something bad like an api key will it be sent out there or does a team of moderators or a moderation ai review it?

[gleaming roost]

Good question

[echo aurora]

Sorry to the delay! I'm checking with the team for clarification.

[echo aurora]

I'm unable to provide official interpretations of our Terms of Use or Privacy Policy. So I would recommend reviewing these docs yourself.

[hollow flicker]

alr

[fervent bridge]

thanks

[rose sky]

Made with Sonauto AI

[minor bramble]

any way to fix that it stuck like that for 10 mins already

[fervent bridge]

refresh and try again

[minor bramble]

even after refresh it doesnt stop

[fervent bridge]

change the model

[minor bramble]

it says generating

is there a way i can mannualky restart it or idk

stop it

[fervent bridge]

our prompt must be so big

[minor bramble]

no i just send "error" and image

[gleaming roost]

Please wait a while, it's probably still writing the reply

[minor bramble]

for 15 minutes already

[fervent bridge]

Are you part of the lmarena team?

[gleaming roost]

no

[gleaming roost]

Oh, see if ctrl+s has any effect

[minor bramble]

just saves html?

[gleaming roost]

Sometimes this command cancels the request

[fervent bridge]

or open a new chat

[minor bramble]

i need that chat specially that why i wrote here

maybe i could wait for long time so it time outs idk

oh it did

[fervent bridge]

So wait a moment, maybe an hour or more.

[minor bramble]

finnaly

[fervent bridge]

good

[cloud oak]

hello..is this the place to ask for technical help?.. 🥺

[rigid copper]

sup

[rigid copper]

in what kind of technical help? generation error, error with specific model or what?

[cloud oak]

False positives for TOS violation when using the website.. 🥺
one prompt kept getting flagged but i don't understand why..

[rose sky]

Damn… looked at what my father cooked for me at home for breakfast

[rigid copper]

you're in the right place to tell about problem of false positives for ToS violation, but it's better to report in #1343291835845578853 or a suggestion to adjust prompt filtering in #1372230675914031105

[rigid copper]

i got the same issue actually, i guess everyone found false tos violation to be annoying

(my english is kind of bad)

https://discord.com/channels/1340554757349179412/1447983134426660894 someone already posted the same thing too.

[cloud oak]

yes i already reported in #1343291835845578853 thank you.. 🥹

[gusty egret]

You're not crazy, I think nano banana is down