general

[ocean kindle]

gpt image 1

[hard drift]

Can someone please mute this guy @echo aurora

[hard drift]

Hmm it’s good also but it doesn’t generate image sin 9:16 format

I mostly use for making app previews, so need something that can do 9:16 high res

The Web and Apps LM Arena is not working; it says: “Failed to create sandbox”

Is anyone else facing the same issue or knows a solution? I’ve been trying for a few hours.

[proper spindle]

is there a way to make images which are not 1:1 format on lmarena?

[hard drift]

Yeah, just use a blank image of the same size (white image only) and then give your prompt along resolution. It will give you same resolution of your uploaded images (but it also doesn’t work well for all the models)

[hard drift]

It works for qwen image edit flux konnect, nano banana, but not sure about others

[proper spindle]

doesn't really work with seedream unfortunately 🙁

[hard drift]

Yeah it doesn’t work with seadream and gpt image

I don’t know but I am worried about the lmarena owners 🙂

I don’t know how they gonna monetize it, hope they don’t burn their own money lol 😁

Thanks for the best AI tool, it is going to replace all other tools

[potent snow]

Will we ever be able to upload and edit pictures in side by side mode?

[echo aurora]

You can; however, the option won't appear unless both of the models you select have image edit capabilities.

[potent snow]

Oh, missed that

[echo aurora]

It's an easy thing to miss

[glossy umbra]

But also gives tenfold than the side effects

chat with anyone at km, play , work,etc

even use lmarena

😅

[neat apex]

Does the relative sucess of Gemma Vault means they will creat an 10 trillion parametters model using all youtube data ever?

[whole swallow]

So do we have any news about sonoma models? Are they from grok or who?

And which one do you guys prefeer? Sky or Dusk

I feel like dusk got a little more smarter lately

[glacial dock]

Today’s word of the day is quack, I will now use it in a sentence.

What duh quack ?

thank you

[vital lake]

Do you guys get paid for introducing a code-name LM?

[sullen sand]

So are you saying that lmarena are going to become something to pay for use it?

Or I am crazy right now

[torn bison]

Can we get gemini 3 pro before november?

[worn schooner]

hey people anyone here have already created any virtual influencer? i would like to know how to create one!

[vital lake]

I feel like Gemini 3 gonna be overhyped

Tbh

All the big jumps come from OpenAI models (o3, o1)

The other companies just trying to catch up

[eager crag]

i am sorry for using only direct mode... i kinda feel bad now.

[keen beacon]

I think that you will eventually turn the Arena into a platform where users would be able to use all features of all models they want for free, in exchange for their data used for training, and if they'd want some privacy, they'll have to pay for it themselves.

[echo aurora]

That's not what I'm saying, we have every intention to keep LMArena free and open.

[keen beacon]

Which is not a bad thing at all if you handle careless users from leaking their PII into public datasets (oof)

[sullen sand]

Oh ok I was scared

[neon idol]

We are in 2 🥀

[wicked dome]

Ai

Wsp

[proud hazel]

https://tenor.com/view/because-of-you-assistant-special-agent-in-charge-jubal-valentine-fbi-youre-the-reason-thanks-to-you-gif-25728086

[neon idol]

Hi man

[proud hazel]

👋

[keen beacon]

Because, kinda, I haven't yet had a usecase where Id need some sort of secrecy when prompting models, in fact it is even better on the Arena because it acts as proxy and with Tor, these are generic requests that can't be traced to my identity

[neon idol]

Hru?

[keen beacon]

Which I believe is true for 99% of users

[eager crag]

what do you mean?

[proud hazel]

I'm still crying because of how LMArena changed.

[wicked dome]

Hm?

[neon idol]

Do you have other sigma banana stuff 🥀🥀

[keen beacon]

how? captcha?

[neon idol]

Huh? What is bad?

I see that they keep always the site with the latest models

[keen beacon]

@echo aurora how do you feel about this idea? I don't think it's bad. Most tasks people use llms for like math or programming are not really privacy compromising lol. Even when people work at proprietary software they ask questions on stackoverflow.

[proud hazel]

Oh, you know, that annoying account requirement, since the site won’t work without it.

[neon idol]

Just do it

What's the problem

[echo aurora]

It's an interesting idea for sure.

[keen beacon]

Thanks, hope you'll keep it in mind to keep providing us the best models for free 🩵

[proud hazel]

Well, it worked for months without it. Sure, you can create a Smurf account, but it's still a bad solution.

[neon idol]

With this you can have access at all the chat in your device and also in others

For me is really good becuase I use lmarena on the computer, on mobile and on the tablet

[proud hazel]

That's all well and good, but what if you don't need it? I only use LMArena on my PC.

[neon idol]

🤷

[proud hazel]

😕

Well, since they're unlikely to change it back, I'll just have to accept it.

[neon idol]

But what is the problem in just doing an access with a Google account?

[proud hazel]

Yes, I'll probably just create an extra account for this site. I feel that my private account is a little too insecure.

[stiff glacier]

Why is he lying?

Deepscam

[proud hazel]

The whale is unaware of its existence.

[stiff glacier]

And this?

[proud hazel]

Could be true.

[stiff glacier]

Well, says "Oceanstone," which he was trained by Google.

[remote arrow]

🍌

[ocean vortex]

https://webdev-arena.vercel.app/leaderboard

this is so close for nr1 spot now 🤯

[ocean vortex]

A platform where you can use battle mode to test the models and then have 5 messages after voting for the actual testing independent of the voting with the models already revealed, or pay a fee to use all models in direct chat mode with only fair-use limits. Just my 2 cents.

I think that would actually drive engagement and help everyone long-term.

[neon idol]

Hi man

[keen beacon]

DeepSeek indeed conceals his identity sometimes. It probably was trained at the output of these LLMs so...

[ocean vortex]

I think it's more of it having no clue what it is without a system prompt lol. Original R1 was notorious for saying it's an OpenAI model. All they have to do is not to include additional fine-tuning data and no system prompt for it to be "concealed"

[keen beacon]

LOL, maybe.

[barren prairie]

Not new it was there a month ago and there is flashcards and visual things if you noticed, also some podcasts. Now chatgpt have the same feature too. But using canvas for those things are better than those quizez.

[hardy lion]

what do these options mean? 🤔

[vital lake]

It felt hollow, like it pretended to know but didn't actually know

[lone vector]

do you guys think OceanStone is gemini 3.0

[remote arrow]

Hi there..

[keen beacon]

Models always do this when responding. You just need to have a prompt difficult enough for them to expose them. Ask GPT 5 to write a story in a conlang similar to English - it will just confidently spot out English words alongside gibberish

I do not believe that they can't reason - rather I believe that they reason in their own, mysterious, non-humanlike ways

[calm lagoon]

Friends, what does everyone think about Reve? The image generator, is it better than Nano Banana?

[pulsar walrus]

hi!!!

[echo aurora]

[vital lake]

I know this

What I'm saying is the weights weren't improved, it was just trained to try and act like, for example, o1, but wasnt actually that good at answering things at it's base. This is especially bad if it doesnt have enough fine tunned data for other topics. The only reason R1 was considered "good" was because it was the second reasoning model ever and it was next to o1 in coding.

[eager crag]

hey pineapple! i want to give my thanks to you for giving free access to models! i will donate when i'll have some money!

[toxic egret]

what's the best model for working with math and LaTeX?

[safe sleet]

Yes

[keen beacon]

Ohhhh IIRC I came across this guy on Artificial Analysis Arena a couple of times and it was impressive.

[eager crag]

i'd still wait for innovation to do it's thing, before i get to use image generation or editing AI seriously.

one year ago, i asked flux inpaint to edit out a door from a photo and replace it with a simple white wall, and all it did, was some jumble of black lines in it's place.

i don't know if it's a workflow i downloaded, or it's the model, but this made me think i should wait for it to be ready.

[calm lagoon]

Thank you❤️

[pine sorrel]

please bring back image upload for Claude models

[vital lake]

Yall are on something

Everyone who said yes never used it

[burnt sinew]

gemini 3 will probably release in around 45 days

according to polymarket at least

[candid nest]

when i go to type ina prompt it wont let me type anything not even /

[ocean vortex]

they better release it sooner than gpt5.1 drops. That's not a guarantee though

With their track record you can kinda expect anything lol. They can be doing 3 new mystery models and then silence for weeks

[ocean vortex]

If you have spent $200 you are unlikely to say it's not worth it at least for the duration you are paying for it...

[jade egret]

def not worth it

but it rlly depends

ig

[remote arrow]

🍌 + 🥤

[supple vector]

lmarena

[keen beacon]

Probably from a breathing thing at night

[burnt sinew]

lmarena direct chat 😄

[mortal coyote]

nano banana cant follow instructions - just gives the same image created , rather than the prompt given to edit the image

on LMarena

works fine on Ai studio GOOGLE

[strong basalt]

@primal solstice wen

[quiet bramble]

hi all...just joined

[echo sinew]

You'll want to check #1397655624103493813 to learn how to properly use the bot.

[remote arrow]

This server really needs a "welcome" or "intro" channel.. 😏

[hard jolt]

hi ... just joined 🙂

[earnest rover]

Why have to do these. Remember we did it and they gave us a rate limit.

[abstract zephyr]

Seedream-4-high-res is unavailable since yesterday, is it a common maintenance ? Where can we check the status of models when they disappear from the models list ? 🤔

[brisk token]

#做一个视频

[stuck plaza]

New here

[vital lake]

Wym 5.1?

[deep nest]

Hi

[storm sphinx]

hello

[visual chasm]

is gemini currently down ?

[vital lake]

Just so yall know, if it doesn't use Tree-of-Thoughts that GPT 5 Pro uses, then it wont be better. https://arxiv.org/abs/2305.10601

[robust yoke]

Greetings.

[robust yoke]

True.

[robust yoke]

Welcome to the server.

[robust yoke]

Greetings.

[keen beacon]

GPT Pro or GPT High?

[vital lake]

Pro

[keen beacon]

I need unlimited free or cheap site like google ai studio

[sour saffron]

Guys
Is it possible to share a Convo from lm arena?

[robust yoke]

Well, this is the place for just that.

[robust yoke]

As of right now, there isn't any feature to. However, you could just copy-paste the conversation.

[sour saffron]

Oh alr thanks

[robust yoke]

My pleasure.

[stiff siren]

I have concerns that how can I use seedream 4 model ?

[robust yoke]

You can use it on LM Arena.

[patent hornet]

hi there, how can i know the latests models?

[robust yoke]

Hmm...

[patent hornet]

i mean, how can i use the Arena Battles App like in this screenshot?

is a kind of "alert system" or anything like that, isn't it?

[robust yoke]

I think so.

[knotty tiger]

hello

[polar kiln]

hello

[hardy lion]

the most surefire way to know the latest models is to chat with them

[random canyon]

Does Claude Opus 4.1 have a maximum context of 16,000 on the LM Arena?

[vapid hill]

🦴

[alpine elbow]

hello training to learn

[finite verge]

HELLO

[proven lava]

Hello

[proud hazel]

Heylow

[whole furnace]

helooo

[elfin epoch]

HI

[cloud marten]

what is the best AI tool for making thumbnail

[wooden gust]

⚡ “Endless Waiting in Model Comparison: Still No Results!”

“I have been trying to compare the two models for quite a while, but I still haven’t received any results. This long delay is quite frustrating. Please look into this issue seriously and ensure faster result delivery.”

[ocean vortex]

I meant whatever they are gonna call the next gpt model update, probably not gpt6...

[verbal nimbus]

That's the max number of tokens it can use for reasoning per turn I think

[verbal nimbus]

Knowing OpenAI, it's probably gonna be called GPT 4.3 /jk

[abstract zephyr]

Can anyone confirm me if you have an access to Seedream-4-high-res right now on the models listed ? 🤔

[sly lily]

Hey everyone, anyone has used AI for digital retouch on products? I'm trying to enhance a flat packshot into something close to a studio photo retouched, but results arent getting there. I've been using nanobana and fixing small errors on photoshop.

Any suggest or prompt?

[bitter lotus]

Whenever I am generating images using seedream 4 high res model of Indian woman's output always has the same face of women however I haven't uploaded any reference image how to fix this

[noble quest]

Hi, is there a way to use the video arena privately, without others being able to see the results?

[abstract zephyr]

@bitter lotus is seedream 4 high res working for you now ?

[stone oxide]

yea the LMArena personal tracker browser extension turned into a mess of a "project" and the AI can't even fix click-through issues.
On top of that, the extension managed to absolutely wreak havoc on all other extensions, for example my youtube downloader uh... yea the two extensions are having some sort of weird UI battle.

Time to laugh about it and come back in another half year when AI is at the point it doesn't do stuff like that anymore xD

[fierce leaf]

Hello

[robust yoke]

Greetings.

[robust yoke]

I see…

Maybe Qwen could fix it.

[tropic nova]

Hello

[robust yoke]

Greetings.

[bitter lotus]

Yes it's working!!

[abstract zephyr]

@bitter lotus I don't know why it's not in my list anymore since yesterday. Also, to fix your problem of having the same lady on your picture, click top left icon and click > new chat, if you restart from scratch the AI won't use the old pictures

[bitter lotus]

Yes I am creating new chat and even deleted old chats but still a problem is there

[abstract zephyr]

that's creepy, maybe try from a new account if you can switch for just a test ?

[robust yoke]

If you're not able to see Seedream 4 High Res in your model picker, then it could be either one of these two issues:

1. You don't have image generation enabled, which can be easily enabled by clicking the image generation modality button
2. You're experiencing a browser bug. In which case, all you have to do is simply use a different browser

Some issues are browser-specific. If you switch to a different browser, that should likely resolve the problem you're currently experiencing.

[bitter lotus]

Thank you for replying Yes I'll try and let you know

[ocean vortex]

or they gonna call it gpt5o lmao

[sullen quest]

gpt o5o... plus

[upper steeple]

How to use LMArena to generate videos

[echo aurora]

The inforation you're looking for can be found here - #1397655624103493813

[barren prairie]

Is there any Gemini3 or Gemma on arena now ??

Anounymous

[jovial mulch]

Sup

[hot anvil]

how many videos can i generate in a day?

[terse sleet]

https://tenor.com/view/hi-hello-there-hello-sup-swag-gif-17997537889289813507

Hello people

[echo aurora]

5

[echo aurora]

Hello

[barren prairie]

Prompts ?

[echo aurora]

Yeah

[barren prairie]

Logan now said " I am a large language model trained by Google" so something must be dropped into arena... I HOPE it is not the useless Gemma

[light stump]

🔥

[prisma girder]

hello

[dense dew]

hii guys, does anyone know how to create a prompt for generating an image like this?

[mellow frigate]

Here you go:

close-up of a model wearing a colorful galaxy mask and crystal helmet, with metallic pink and pastel glitter in the style of haute couture fashion. photographed for the cover of vogue magazine, featuring gold chain necklaces, with a futuristic vintage aesthetic, 2057.

[terse dagger]

Hello

[dense dew]

is that work bro? what kind a tools did you use?

[keen beacon]

Codex for the price is a much better tradeoff than Opus

@echo aurora do you remove stealth models off the arena sometimes?

Can't meet Raptor again, met it thrice yesterday!

[echo aurora]

Yeah, models using a codenames can be removed from Battle mode

[keen beacon]

I see, pretty much explains why Ii can't meet Raptor anymore. sigh...

[hot anvil]

how can I generate vertical images on lmarena website?

[echo aurora]

It's possible you're just not being sampled it in that generation you've made, and it's also possible it has been removed. Just because you're not seeing a model doesn't necessarily mean it has been removed.

[keen beacon]

Hmmm, interesting.

[echo aurora]

You're unable to set the ratio sizing with image generation.

[hot anvil]

so its always 1:1?

[mellow frigate]

i used midjourney's /describe command

[robust yoke]

Greetings.

[robust yoke]

Greetings.

[molten solar]

https://www.tiktok.com/@van.melodies/video/7550732410846022926?_t=ZT-8zmRYjzRXLS&_r=1

[sudden socket]

hi

[molten solar]

hii

[keen beacon]

Is multiple unrelated problems in one prompt more difficult for a LLM than one problem per prompt?

[robust yoke]

Greetings.

[robust yoke]

I feel like multiple different problems in one LLM usage is more difficult than simply splitting the problems up one-by-one.

[keen beacon]

source?

[robust yoke]

That is, unless the LLM is insanely talented and is able to handle multiple problems, like Claude, for example.

[robust yoke]

Source: my knowledge.

[keen beacon]

yes, that claude that is insanely talented at being bad at everything except coding where it exceeds gpt 5 a microstep for 100x increase in cost.

[robust yoke]

Yeah, exactly.

[prime mulch]

I would said seedream 4 highres in first and gemini in 2nd

[robust yoke]

True.

[winged mauve]

Yup, seedream 4 high res now generates great results

[prime mulch]

Is it looks like ai generated

[robust yoke]

For editing, it is on par with ChatGPT.

[surreal creek]

Just realized I’ve been reading it this whole time as Seed Ream and not See Dream lol

[robust yoke]

For actual generating, it's on par with ChatGPT.

[mellow frigate]

license plate looks weird

[prime mulch]

@winged mauve @robust yoke

[robust yoke]

Heh.

It's all good.

[prime mulch]

Rather than that what you think

[winged mauve]

not really, until closely observed

[robust yoke]

Somewhat.

[prime mulch]

That's why I like seedream highres

[mellow frigate]

Looks pretty nice. i didn't know motorcycles can have chains

[winged mauve]

Yup

[prime mulch]

Bruh don't you know about motorcycles?

Chain is must

Without chain how it runs

[robust yoke]

The motor. 🤓

[mellow frigate]

Yeah im searching it now, looks like they actually do have chains. never knew

[prime mulch]

Chain is connecting tyre to give power to rotate

[robust yoke]

Yeah, just a joke.

[prime mulch]

All hail seedream 4 highres

[neon idol]

Hello

[proud hazel]

https://tenor.com/view/salute-cat-cute-yessir-gif-17449845613192073437

[remote arrow]

Oh soda dream.. 🥤

[ocean vortex]

codeing module?

[tribal wedge]

[leaden sun]

check this out #ai-news message @keen beacon You're absolutely right!

[robust yoke]

Oh neat.

Does that mean Otaku is correct, or am I correct?

Or are we both correct?

[glossy umbra]

Lol . Its like comparing 2 different infrastructures. One is 99% and the other 100%. But the 100% uses 1000x more than the 99%. And people still use the 100%.

[astral meadow]

yo guys new here
wanted to know why is Midjourney not ranked at a good spot in the leaderboard, have heard that its one of the best text to image models out there.

[neon idol]

Not real

[astral meadow]

is it not a great model?

[novel nimbus]

hello!

[brazen coral]

Guys is there a buh

G

[rancid orbit]

what does people use to upscale and unblur messy images

[echo aurora]

What seems to be the issue?

[brazen coral]

Stay loading

[echo aurora]

That's strange. Can you go ahead and create a post in #1343291835845578853 with more details?

[brazen coral]

I do

[frosty berry]

Does anyone have a prompt to do a face swap with nano banana?

[proud hazel]

Not supported.

[sterile isle]

just upload the two images and tell it to change the face its that simple but nano can be a bit of a wuss if it thinks it is a celeb

[proud hazel]

Here's my pathetic attempt.

[willow grail]

diud i just read on reddit that gemini 3 is on lmarena?

[little narwhal]

Link pls

[willow grail]

https://www.reddit.com/r/Bard/comments/1ni5cbc/so_ig_new_model_on_lmarena_is_gemini_3_maybe_or/

[neon idol]

💀

[willow grail]

OceanStone, NightRideOn, NightRideOnV2, SkyTrail

[neon idol]

I'll dream about this tonight

[willow grail]

possible gemini 3 names on lmarena

[exotic tartan]

Is it only me or 4o is better than GPT-5? Only today I had 2 examples where 5 would just go on and on and miss the point entirely, while 4o gets the right answer, while being concise and not responding in 6 bulletins every time. I even double checked the same prompt and had the same results to make sure it's not a fluke. The f is going on?

[neon idol]

Yes you are the only 💀🥀

[willow grail]

actually gpt 3.5 is much better than 4o

i swear by my...

[exotic tartan]

No, it was really bad and 4o is an obviously better model

[willow grail]

no gpt 2 is better

[exotic tartan]

With 5 it's not that obvious in many cases

[vital lake]

LOL

Ragebait

[vital lake]

You using free version or auto router

[exotic tartan]

I'm using the normal desktop agent, Pro tier with GPT-5 on Auto

Switching between that and 4o. 4o kicked its ass multiple times today.

I have 0 reason to simp for 4o, I'm not one of those weirdos who hate the change in character or whatever. I just want good answers

I just felt like 5 is being a dumbass lately, so decided to switch back to verify since I didn't remember it being so off

[sick sedge]

hi every body

[empty stump]

Which one is the best one

[robust yoke]

Greetings.

Claude can also give good answers, I believe. It's also pretty good at natural creative writing and helping with getting accurate information.

But to each their own, I suppose.

[barren prairie]

Are they wow models or just good ?

[neon idol]

Hello!

[robust yoke]

Greetings, Greg.

[neon idol]

Hru?

[robust yoke]

Doing good, and you?

[neon idol]

Me too thx

[robust yoke]

My pleasure.

[neon idol]

:)

Waiting for some memes on nano banana or soda dream 😞

[robust yoke]

That's fair.

In that case, I could create a naner-bananer meme.

[neon idol]

Do it 😂

[robust yoke]

Bet.

[neon idol]

?

[robust yoke]

È un modo per dire “Va bene”.

[neon idol]

Ohhh

Thx

[remote arrow]

Memes are forbidden in this channel, you'll get a hot slap..

[neon idol]

Pineapple with us 😞🥀

[remote arrow]

🤣 🤣 🤣

[neon idol]

Omg these emojis remember me The Big Brother of George Orwell, idk why 😂😂

[remote arrow]

I only post on #ai-creations when I have time and idea. Only come here to read random HI and HELLO..

[neon idol]

Me too

[remote arrow]

I keep wondering why the name is GENERAL, while it's not allowed for general talks. Said it must be "AI related only," yet it's mixed with so many HI, HELLO, and I AM NEW. Or "Can you help me to make my photo smile?"

[robust yoke]

https://messages-prod.27c852f3500f38c1e7786e2c9ff9e48f.r2.cloudflarestorage.com/823ac0e8-9bd8-410d-881c-8f9873469ea2/1758059092097-af82a6a4-3af8-4afd-8ce6-42e7e80b58bf.jpeg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=c86e09ae0bc1d897b03dfaa30a8b51f3%2F20250916%2Fauto%2Fs3%2Faws4_request&X-Amz-Date=20250916T214452Z&X-Amz-Expires=3600&X-Amz-Signature=6f271331c62da664ffa46f8d99d005bf8c5bf2e3f7b1f1236b53d6a73a1a89ea&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject

[remote arrow]

Hey, that's not AI related. Because banana is natural-related. Hide it!

[robust yoke]

Gee willikers! I think I'm in t-t-t-t-trouble!!!1!1!!1!1!

[neon idol]

Always pineapple for revenge

[neon idol]

Ahaha

[remote arrow]

Hi. I am not new.. 👋

[neon idol]

Who cares?

[robust yoke]

Hi, I like to mew.

In fact, I'm mewing right now.

[remote arrow]

Hurt... 🤧

[robust yoke]

🤫🧏‍♂️

[remote arrow]

Just exploring. And I am wondering why there are so many people posting prompts. In almost every channel. Even the leaderboard channel..

[neon idol]

Answer at this question, level impossible final boss 🥀💀

[remote arrow]

You'll need a cheat engine

[neon idol]

For real

[remote arrow]

... ... ... ... but ... why?

[thick moss]

Hello world! I'm new to the internet, what's there to do around here?

[neon idol]

...

........

@remote arrow no sooner said than done

[neon idol]

New on internet?

[remote arrow]

...My biggest accomplishment today: I can show my family that I can type on a keyboard...

[neon idol]

😂😂

[robust yoke]

https://messages-prod.27c852f3500f38c1e7786e2c9ff9e48f.r2.cloudflarestorage.com/823ac0e8-9bd8-410d-881c-8f9873469ea2/1758059820208-7edab72c-af48-4a3b-b9f1-48179b836134.jpeg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=c86e09ae0bc1d897b03dfaa30a8b51f3%2F20250916%2Fauto%2Fs3%2Faws4_request&X-Amz-Date=20250916T215701Z&X-Amz-Expires=3600&X-Amz-Signature=3c1c41513d213cede1d8c666b605d822339c7102034a48c5e2890305a5a7edbb&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject

https://messages-prod.27c852f3500f38c1e7786e2c9ff9e48f.r2.cloudflarestorage.com/823ac0e8-9bd8-410d-881c-8f9873469ea2/1758059989924-014a5702-6fb5-487e-a9b0-9edb86ceb778.jpeg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Content-Sha256=UNSIGNED-PAYLOAD&X-Amz-Credential=c86e09ae0bc1d897b03dfaa30a8b51f3%2F20250916%2Fauto%2Fs3%2Faws4_request&X-Amz-Date=20250916T215950Z&X-Amz-Expires=3600&X-Amz-Signature=c91da5ec68e2cfae04952b61e00ab8265507ca6817b7d36f7f9cfd515e3577b9&X-Amz-SignedHeaders=host&x-amz-checksum-mode=ENABLED&x-id=GetObject

[robust yoke]

Heh.

[toxic egret]

whats different between gpt-5-high-new-system-prompts and gpt-5-high? or more precicely what means new system prompts?

[remote arrow]

There: your banani and dreamy..

[remote arrow]

Main difference: one is new system, the other is too old..

[robust yoke]

GPT-5-High-New-System-Prompt is basically just GPT-5-High but with the personality of GPT-4o. The regular one is known to be “cold” and “professional” (or “corporate” for a lack of better terms). Unlike GPT-4o, which is more sycophantic but is more supportive.

And “sycophantic” means to be overly supportive.

[robust yoke]

https://tenor.com/view/gülmek-komik-gif-1506407869824367367

[toxic egret]

ty

[robust yoke]

My pleasure.

[hollow stag]

hey

[polar niche]

@echo aurora What happened to my bugs thread?

[echo aurora]

I closed it since the bug was no longer a bug, right?

[polar niche]

No, because they are still not generating

I thought it was a generation issue.

[echo aurora]

Oh my apologies, I got ahead of myself there, I've reopened and will followup there.

[still spire]

hello

[echo aurora]

[mortal star]

😎

[rocky stream]

Hi

I am here to try out video creation

[echo aurora]

Welcome welcome, be sure to check out #1397655624103493813 for more information on how to use the bot properly. Let me know if you have any questions.

[elfin sundial]

hey good ppl first here

[alpine star]

Hello mate

[stuck drum]

Hello. Looking forward to trying new AI platforms.

[hot pelican]

Hey guys, why has the leaderboard stalled, especially since August? updates every 2 weeks?

is lmarena intentionally slowing down

[dusty needle]

Hi, here to try different LLMs.

[surreal creek]

pretty sure updates have been staggered to sync with new model releases, but yeah it is a bit hard to get a grasp on the current state of text rankings with the gap in updates

[polar marlin]

Image creation seems slower than before.

[hot pelican]

@surreal creek Yeah, I have almost lost habit of visiting the leaderboard, as it is always the same.

I hope there is a way to remind the team the bad effects of this

@tiny palm are updates planned to be slower and slower?

[surreal creek]

I don’t think it’s a net-negative thing, batch estimation of Bradley-Terry scores with millions of votes is pretty computationally intensive, so it’s not really viable to have a real-time leaderboard - the point is to have it up to-date with the latest models

[hot pelican]

Yeah, I was hopping if it was feasible to update at least every week

also makes news and publicity for the leaderboard when models overtaking the top get announced by users and shared around. Instead of missing such events and dumping a long time update including with new models.

I am sure a change was there with this week's update. But if the update is going to be combined with Gemini 3, it's another top position lockup by google

[robust yoke]

Greetings.

[robust yoke]

Greetings. Hope you have fun in doing so.

[robust yoke]

I'm sure you'll like it here.

[robust yoke]

Very interesting.

[hard quiver]

SeaDream4 still has the restriction where output always produces square images, which makes image editing—especially tasks where I want it to fill in a blank part of the image—turn out quite poorly. Are they going to change that restriction at some point?

[robust yoke]

Well actually, it seems like the aspect ratio is built into the actual model itself, not the site. So in order to change the aspect ratio, the people who made Seedream 4 would have to manually change what kind of aspect ratio it can output.

At least I think that's how that works anyway. I think you can specify what kind of output ratio it sets by creating a very detailed prompt as to what kind of aspect ratio you want coming out of it. I remember I made a prompt that was very detailed and specified a pretty wide aspect ratio, and it gave me a pretty wide aspect ratio upon generating the image.

[hard quiver]

I’m about to tell that to ChatGPT

[robust yoke]

Alrighty, good luck.

[remote arrow]

Platform based. The image generator itself is adjustable, just like the other.

[robust yoke]

I see.

Perhaps it simply might just be that the version they used has a default ratio of 1:1.

[remote arrow]

LMArena might adjusted for lighter load. And it is just alright, to set the way it is now, because the main purpose is to test the quality.

[robust yoke]

Yeah, exactly. The model doesn't have to output a perfect thing; it's solely just for testing.

[remote arrow]

It's just like a bakery promoting a new snack and distributing free tester pack.

People that are asking full pack for free might be marked with big tattoo on their forehead.

[robust yoke]

True.

A big tattoo that says 'Freeloader' on their forehead.

[remote arrow]

Banana-shaped question mark

[robust yoke]

https://tenor.com/view/the-rock-dwayne-johnson-raised-eyebrow-gif-6506694932635492760

[remote arrow]

https://www.youtube.com/watch?v=wCkerYMffMo

[robust yoke]

https://tenor.com/view/eyes-chew-chewing-big-eyes-wide-eyes-gif-8128904

[unborn lantern]

that is seedream 4 fal?

[old mountain]

https://tenor.com/view/kitty-cat-hello-chat-gif-5801500268894668499

[robust yoke]

The thing I sent?

[robust yoke]

Greetings.

[remote arrow]

🤔 Who? Where? What?

[robust yoke]

Who, where, when, what, why, was?

[remote arrow]

How?

[robust yoke]

How, who, when, what, where, why?

[remote arrow]

And this makes me want to send another AI creation or a new meme.. 🤣

[robust yoke]

Heh.

[spiral zephyr]

Hello

[remote arrow]

The restrictions are acceptable so far. Rate limit, aspect ratio, etc. That are not the true annoying agent for me. There's one other significant thing that is the most annoying.

[robust yoke]

Greetings.

[robust yoke]

And what would that be exactly?

[robust yoke]

True.

[chilly granite]

hello

[robust yoke]

Greetings.

I hate to break it to you, Safeer, but I don't think this is the place for that.

#video-arena-1 <- This is, though.

[obsidian shell]

is the site down ?

[robust yoke]

Lemme check...

[empty stump]

it loads fine

[robust yoke]

Nah.

[fiery leaf]

Hi

[remote arrow]

I'm down. The site's up.

[robust yoke]

I'm up, the site's down.

[remote arrow]

ALERT!

[dawn mortar]

hi

[jovial nova]

hey friends i m new here kindly accept my greetings

[dreamy lily]

hello

[tribal wedge]

[echo aurora]

[robust yoke]

😈

[remote arrow]

What's SodaDream 4 fal? Something modded? LORA embedded?

[jagged ferry]

He

[slow beacon]

Hlo

[strange yacht]

Hi everyone

[astral meadow]

yo guys new here
wanted to know why is Midjourney not ranked at a good spot in the leaderboard, have heard that its one of the best text to image models out there.

[arctic gazelle]

Hi everyone, where is gemini3?

[crisp brook]

hi

[tacit wave]

Hello. I want to create video for my educational youtube channel

[echo aurora]

- we don't have Midjourney on the LMArena platform. There isn't an available API for the model

[dense fox]

how i usE IT ? TWO PLATFORMS LIKE veo3 and seedream at a time ?

[leaden sun]

you->plural in this context 😉

[valid flume]

hi i want image to video

[robust yoke]

Gotcha.

Neat.

[robust yoke]

You can visit the #video-arena-1 channel for that.

[robust yoke]

Greetings.

Greetings, newcomers.

[tulip grove]

Some time ago, before logging in, all chats disappeared from my history. A few days ago, one chat “froze” during generation. I reported the bug, but it wasn’t fixed. While browsing the options, I accidentally deleted that thread. Is it possible to recover them? At least the one associated with my account?

[vernal meadow]

Why can you choose between Seedream 4 and Seedream-4-fal ? Are they the same?

[fresh flint]

Guys send me some potrait prompts...

[dapper nexus]

Greetings from Kenya

[fiery lichen]

hi

[brisk path]

hello

[honest ore]

hello

[worn folio]

#1413497734756827186

[prime mulch]

Seedream is already surpassed mid journey

[worn folio]

Promt nội dung

[heady kernel]

I have my logged in chats on my PC. On my phone I have different chats and not logged in. If I now log in into the same account from my phone will the chats merge into the same account and then I will be able to access them from both my phone and my pc?

[limber wind]

hi i am new here loving this platfrom amazing work thanks team for making this tool

[fresh basin]

hello everybody it's a fantastic tool here to create magic video

[limber wind]

yeeh

[quasi sleet]

Hey there its good to be here lads!!!

[bitter lotus]

What is the difference between seedream-4 & seedream-4-fal ?

[patent sandal]

Hello !!! I'm here to research video generators .... hope you're fine and happy

[coarse geode]

hello, came in to explore the great tools available here

[exotic tartan]

check out #video-arena-1

[lavish stratus]

Hello guys Its a pleasure!!!

[night fern]

Hi! Hoping to find a suitable LLM we can host to replace our CAT tool

[teal lion]

hello everyone... i'm here to explore the best tools in the world

[barren prairie]

ليس هنا

[hollow imp]

I didn't understand

[chrome swallow]

hello guys

[sturdy mica]

hello guys

[lucid flint]

I vahe just heard of LMarena! i should have knowned it a long time before

[sturdy mica]

error

[acoustic remnant]

你好

[keen beacon]

Hi, does anybody know what seedream-4-fal version is compared to the high res and normal version?

[naive ravine]

Hello Gs!

[opal stream]

hey guys new to the community here to explore

[obsidian shell]

its their image edit

[keen beacon]

Ok, thx

[rustic phoenix]

heloo

[verbal nimbus]

Nightride-on is very likely a Gemini 2.5 Pro checkpoint. Their responses are almost verbatim (top is 2.5 pro).

I think it uses the Internet though, isn't that kind of cheating?

I think nightride-on = nightride but online.

[atomic stream]

Lmarena died?

I'm getting always errors. Just one time worked i don't know how.

[nimble vale]

10-second vertical animation. A large bag of Kuriyama 54 Stars Rice Crackers opens in the center, releasing floating star-shaped crackers and mini packs. Koshio (a cheerful young boy in flight attendant-style outfit) happily makes a star gesture, while Mumu (a cute girl mascot with big round eyes) hugs a rice cracker and jumps with joy. Background with dreamy starry sky and soft pastel glow, family-friendly, heartwarming, shareable mood. No text。

[solid beacon]

#command

[sterile sequoia]

/videos

[tame kiln]

the person in the image as a reporter in a report about a tsunami

[exotic gust]

hi

[prime isle]

hello, i come here to create advertise video .

[willow grail]

lets not forget google is the one who has money. not anthropic

@glossy umbra

[glossy umbra]

anthropic is work harder not smarter

because their infrastruture is good yes but it cost too much to maintain

[willow grail]

which then speaks for google, not anthropic.

[willow grail]

what the hell does that sentence even mean.

[remote idol]

Well, I choose creativity and design Opus, but the Gemini 3 is good as well.

[willow grail]

gemini 3 will be build on something completely new

thats why u cant just look gem 2 to 2.5

[remote idol]

Is there a possibility that it might be better at design?

[willow grail]

yes.

[fallen hemlock]

Helloooo

[remote idol]

Do you have any information about when Gemini 3 will be released?

[willow grail]

anthropic mentioned big event this week. gpt codex is released.
so most likely this month for gemini 3

[stuck storm]

hmm does gpt-5-high or gpt-5-high-new-system-prompt error out for some more intense questions (that it must spend more time to think over)

take this one for example, a random slightly less solved project euler question:

A triplet of positive integers ( ( a, b, c ) ) is called a Cardano Triplet if it satisfies the condition:
[
\sqrt[3]{a+b \sqrt{c}}+\sqrt[3]{a-b \sqrt{c}}=1
]

For example, ( ( 2,1,5 ) ) is a Cardano Triplet.
There exist 149 Cardano Triplets for which ( a+b+c \leq 1000 ).

Find how many Cardano Triplets exist such that ( a+b+c \leq 110000000 ). Your solution must be in Python.

[remote idol]

I just checked it's not this month. Derek Nee tweeted saying it's not this month but an advantage is that Gemini 3.0 Flash is > 2.5 Pro which means it's better.

[stuck storm]

gpt-5-high ALWAYS errors out on this after quite some time thinking with Something went wrong

[remote idol]

If only they made an operator that has an agent that could work on a VM and execute instructions like ChatGPT Operator then that would be amazing

Could Gemini 3 have been capable of that?

If so, then Gemini 3 is the best AI in the history of the world, even better than local models that run up to 500B parameters.

[stark hearth]

9:16 format. Animated infographic video style. A cartoon electricity meter spinning rapidly, exaggerated motion. Smooth glowing energy lines flow outward from the meter into the air, symbolizing wasted electricity. Continuous dynamic motion with gentle pulsing glow effects. Eco-friendly colors: greens, blues, soft yellows. Minimal clean background, modern and uncluttered. Duration: 6 seconds, seamless loop, smooth camera zoom-in toward the spinning meter.

[remote idol]

This is not the channel for video generation.

[quasi atlas]

Please, #1397655624103493813 to learn how to prompt and properly use the bot. You can use the Video Arena channels for your creations: #video-arena-1 #video-arena-2 #video-arena-3

[quasi atlas]

Please, #1397655624103493813 to learn how to prompt and properly use the bot. You can use the Video Arena channels for your creations: #video-arena-1 #video-arena-2 #video-arena-3

[remote arrow]

🤣

[remote idol]

What is so funny? lol

[remote arrow]

It's always hilarious for me, anytime I check this server and found all the channel flooded with prompts..

[remote idol]

Alright.

[ocean vortex]

They probably figured they gonna push 2.5ultra forward into 3.0 version stage and release them all at once 🧐

[remote idol]

Yes, since the tweet I sent confirmed that the release is not this month, so...

[ocean vortex]

I suppose it makes sense for naming. Would have been weird to have 2.5ultra with no 2.0ultra ever being there

[remote idol]

Although I am still happy that 3.0 flash will be better than the 2.5 pro.

[dusky pier]

What's the point of adding a new model

[remote idol]

Windows 12?

[remote arrow]

No point if you can't even max out the existing model. Like buying a newest hypercar when you don't even have a driving license..

[tight spire]

Hello, I am here to compare ai tools

[leaden sun]

I guess Anthropic has completely lost the race now, havent they? time for them to consider a holistic restructuring if they want to survive...

[steel pecan]

Hello, I am here to compare ai tools

[leaden sun]

they work hard and smart actually, just wrong ideology unfortunately and probably lots of mismanagement and miscalculated risk assessment too

[neon idol]

What is the difference between the normal seedream 4 and seedream 4 fal?

[loud junco]

Hello. Here to compare ai tools

[ocean vortex]

It's not gonna be better. At best it will compete with it in most benchmarks

gpt5-mini is not really better than o3 etc

[barren prairie]

But Gemini 3 will be on arena logan said something yesterday

[pine knoll]

hi

[barren prairie]

No google used to make the next flash better than the previous pro version like gemini pro2. 0 and gemini flash 2.5

[leaden sun]

when?

[ocean vortex]

flash is like o4-mini. It's good for mechanical dumb work but falls apart quick on something like SimpleQA or logic fundamentals

[ocean vortex]

Hence why I said it may compete on most benchmarks, but can't realistically replace one gen older bigger model

[weary cedar]

Hi

[robust yoke]

Greetings.

[echo aurora]

Good timezones

[wintry tinsel]

2.5 pro isn’t close to opus for many of my uses I need 3 pro to beat opus

[wintry tinsel]

Still by far the best at creative writing and natural conversations 😅

[sinful ember]

Hi all

[tall summit]

Hello, I am here to compare ai tools

[polar niche]

Just like all of us lmao

[floral comet]

Gpt 5 high new system prompt? Is it considered to be better than the original gpt 5 high?

[remote idol]

It is and I am sure of it, because why would they lie? This does not mean that the context window is higher, though.

[ocean vortex]

No one is lying about anything. There are just certain things it can't possibly match the bigger model at

[viral notch]

gemini 3 is out? is it on lmarena now?

[remote arrow]

I asked the very same question earlier, even to the angels at my left and right side. No answer.

[remote idol]

Well, I am certain that it is going to be better than O3 and GPT-5 combined.

[ocean vortex]

The Flash?

🤣

[leaden sun]

not anymore, the reminder injection Anthropic secretly has attached to every user prompt has ruined it, they did it to avoid... many things but mostly to save compute resources

[maiden dune]

hi

[elfin oak]

may be oceanstone and oceanroof

[ocean kindle]

obviously its phantom ai

[elder burrow]

?

[ocean kindle]

?

[elder burrow]

what phantom ai

[ocean kindle]

phantom ai

[elder burrow]

what is it

??

[ocean kindle]

it is phantom ai

[elder burrow]

so you're telling me its https://phantom.ai/ ??? 🤣

[elfin oak]

Maybe we should consider abloud any-925

it seems from Apple

[elder burrow]

hi again

[echo aurora]

oh yeah

[slow hare]

Hello all, fairly new to AI. In a night class and looking to work with midjourney, discord, Higgsfield and LMarena to create. Looking forward to all the knowledge here.

[echo aurora]

Welcome welcome would note we don't have Midjourney on LMArena

[wicked sage]

hello!! i found lmarena.ai because one of the members from a server called "websim" said something about it!

safe to say, its amazing

[solid brook]

Btw logan said this is fake

[elder burrow]

3.0>2.5 flash?????

ok well

https://cdn.discordapp.com/attachments/1382160535675469985/1382913521368563782/image.gif

lol 😭

DUDE!!!

found a way to differentiate between flash and pro, I guess 🤣

WHAAAAAHAAHAHAHAHAHWHAHWHA

mb for choosing google, I'll press tie for my next tests

[prime mulch]

Why people asking for mid journey we have seedream 4 highres that already destroyed mid journey soon more models will come that destroys the mid journey to absolutely nothing

[elder burrow]

yeah i spread it in websim lol

[elder burrow]

👀

[elder burrow]

it's insane how bad 2.5 flash is

[vital lake]

Buddy you said WORD

A

nvm

LOL

I'M STUPID

😭 🙏

[elder burrow]

😭

safe to say, oceanreef is gemini 3 pro - responds with "Google" consistently, flash models never respond properly.. lol

actually.. both oceanreef and oceanstone

[wicked sage]

ty brody

anyways whats going on rn

[elder burrow]

findin out which model on lmarena is gemini 3

[wicked sage]

evil...

wait question

[remote arrow]

Because that name always pop-up in all type of advertisement. Even when you chat with ChatGeppeto and asking about image generating, it will soonly mentioning Stable Confusion and Mad Journey..

[wicked sage]

are there actually like released unreleased models that arent listed

[elder burrow]

so, lmarena gets exclusive access to "stealth" models from ai companies so they get feedback before releasing it

[wicked sage]

i see ts

wtf is oceanstone

[elder burrow]

gemini 3

[kind lava]

Hello

[wicked sage]

so did you.. get it?

[elder burrow]

yeah, you can get it in the battle mode only

[wicked sage]

thats sad

[elder burrow]

its ok

[wicked sage]

but also funny at the same time

[prime mulch]

But truth is mid journey is already destroyed by nanobanana and seedream people don't understand that I was so disappointed

[wicked sage]

whats the best ai for coding web games?

i tried out making a tycoon game with chatgpt 5 high and it turned out great

i liked the ui and it included dark mode only

and the game was interesting

oh thats why

#video-arena-1

text wall

[surreal creek]

Gemini 3 doesn’t appear to be releasing until around Halloween time, however

[remote arrow]

Most people don't know. Probably it's only us who intensely involve in AI know the new names. The other (commoners), only know what they seen on media. Around me in real life, AI for them is only ChatGeppeto, nothing else. Image generating: only Madjourney.

[wicked sage]

another question

whats the most up to date bot

i mostly used chatgpt 5 chat and i recently found out that its from 2024, im kinda stupid how i found out this late ngl

[prime mulch]

The answer is all the bot you see in lm arena

[wicked sage]

oh so all of the bots have their info in 2024 or smth?

[prime mulch]

I hate this corporate manipulation

[remote arrow]

What makes you stupid by using this? As long as it can works and helping you to gain your goal, it's just fine.

[wicked sage]

it makes me stupid because i found out the bots info was from 2024 too late

i honestly couldnt care really

icl

[remote arrow]

Recently happened in Nani Banani mass-money-burning. They pay almost all available influencer to glorify it. The next day: million of people searching the web for a peeled banana..

[prime mulch]

It works differently all bots are connected with cloud bots are not news channel the bot that only focus on how they analyze your information and how they replay you

[wicked sage]

ok thankies

👍

[remote arrow]

As long as it can access the web. There's nothing outdated.

[tranquil anvil]

guys so i just created an image to video prompt but it doesnt have the sound, whyis that?

[wicked sage]

fair enough

[viral notch]

but does oceanstone code well?

[prime mulch]

People don't know they will lost their data to that nani banana and all free things

[wicked sage]

we dont know yet since its basically battle mode only

[viral notch]

gemini 2.5 pro is decent but 4.1 opus and gpt5 are better

[prime mulch]

Focus on action of the subject and camera movements for video

[wicked sage]

on what stuff exactly?

[remote arrow]

Not all the video generator able to produce sound..

[tranquil anvil]

so it cant make sound?

[tranquil anvil]

oh ok

is there any way to force it tho or you cant?

[prime mulch]

Some models can make sound and some model not

[viral notch]

from my experience? programming web pages that use libraries like threeJS, and adding functionalities on top when requested

[wicked sage]

is there a way i can make a .lrc creator

[prime mulch]

You can't

[tranquil anvil]

so its basically just luck

right?

[viral notch]

like gpt5 and 4.1 opus nailed the tasks

[remote arrow]

No. It depends on the generator. Veo3 can make sound, but other like RealMotion can't. The best workaround: manual dub.

[static jolt]

/video গ্রামের এক প্রান্তে ছোট্ট একটি ছেলে থাকত, তার নাম আরাভ। দাদু-দিদার সাথে থাকত সে। খুব চঞ্চল আর কৌতূহলী ছিল আরাভ। প্রতিদিনই সে মাঠে, বাগানে খেলতে খেলতে নতুন কিছু খুঁজে বের করত।

একদিন দুপুরে আরাভ আমগাছের নিচে বসে ছিল। হঠাৎই কোথা থেকে একটি ছোট্ট বাঁদর এসে হাজির হলো। প্রথমে আরাভ ভয় পেয়ে গেল, মনে করল হয়তো বাঁদরটা তাকে আক্রমণ করবে। কিন্তু আশ্চর্য! বাঁদরটা মাটিতে পড়ে থাকা একটি আম কুড়িয়ে এনে তার দিকে বাড়িয়ে দিল।

আরাভ অবাক হয়ে হেসে ফেলল। তারপর নিজের পকেট থেকে একটা বিস্কুট বের করে বাঁদরটাকে দিল। সেদিন থেকেই শুরু হলো তাদের অদ্ভুত বন্ধুত্ব।

আরাভ বাঁদরের নাম দিল “মোন্টি”। দুজন প্রতিদিন দেখা করত। মোন্টি তাকে গাছে ওঠা শেখাত, কখনো নারকেল পেড়ে খাওয়াত। আরাভও তাকে ফল আর রুটি এনে দিত। এই অদ্ভুত বন্ধুত্ব দেখে পুরো গ্রামে হইচই পড়ে গেল। সবাই অবাক হত— “ছোট ছেলে আর বুনো বাঁদর কিভাবে এত ভালো বন্ধু হলো?”

একদিন গ্রামে মেলা বসেছিল। আরাভ মেলায় গিয়ে ভিড়ের মধ্যে হারিয়ে গেল। ভয়ে কেঁদে ফেলল সে। ঠিক তখনই গাছের ডাল থেকে লাফিয়ে মোন্টি এসে হাজির হলো। সে আরাভের হাত ধরে তাকে দাদু-দিদার কাছে নিরাপদে পৌঁছে দিল।

সেদিন গ্রামবাসীরা বুঝল, এই বন্ধুত্ব শুধুই খেলা নয়, বরং আসল ভরসা আর ভালোবাসার উপর দাঁড়িয়ে আছে।

আরাভ আর মোন্টির এই অনন্য বন্ধুত্ব সবাইকে শেখাল—
ভালোবাসা আর বোঝাপড়া ভাষা বা প্রজাতির সীমা মানে না। মানুষ আর প্রাণীও একে অপরের সত্যিকারের বন্ধু হতে পারে।

[prime mulch]

You can do another thing if you got any student offer from Google you can change year from your old id and claim it and use veo3 for free

[wicked sage]

DUDE

[prime mulch]

Wrong place use this in video section

[wicked sage]

stop it with the textwalls please

[tranquil anvil]

thanks so much for that

[wicked sage]

also why does #video-arena-1 have to be the weirdest channel ngl

[tranquil anvil]

also thanks @remote arrow

[remote arrow]

🤣 isn't it funny. Even in EVERY channel in this server..

[wicked sage]

again, couldnt care less

or more

idrk

ill brb

[remote arrow]

Try rm dot digen dot ai, where you can combine your muted video with the sound you created for free.

[elder burrow]

??? how

[tranquil anvil]

Thanks alot

[elder burrow]

I didn't give it today's date btw

[wicked sage]

please speed i need this

what is nightride v2

[elder burrow]

possible gemini 3 flash lite

it responded almost instantly

[wicked sage]

how are you guys finding this

[wraith cliff]

guys is anyone good at prompts 4 veo 3

[wicked sage]

thats MY question

[echo aurora]

Hey checking in - are others seeing model response seem very slow right now just in Battle? Eventually they'll respond, but taking noticeably longer.

[elder burrow]

i unno, just now nightride responded instantly

its inconsistent

[wicked sage]

gpt 5 high still didnt respond which i think is the funny part

while claude opus 4 finished

[elder burrow]

i believe battle mode now has a system where it makes both models respond at the same time to prevent judging by speed

[elder burrow]

yeah, gpt 5 non-chat is incredibly slow on lmarena

[elder burrow]

you should use 4.1 16k thinking

[wicked sage]

i am

check image

..im not

wait am i stupid

i swear i was

no i did not

https://tenor.com/view/ohio-confusion-i-show-speed-ishowspeed-speed-ohio-gif-18271165267203003105

[elder burrow]

its 4-1

[remote arrow]

Only Geppeto-1-Image slow as usual. It have constant brain damage. The others are just fine..

[wicked sage]

oh you said 4.1

sorry im blind

im legally blind*

i swear

[spare rune]

Ni hao chat

Anyonghaseyo chat

Hello chat

[elder burrow]

idk why they decided to make the model 4-1 instead of 4.1

anthropic stinky as usual

[wicked sage]

retrying ...

[elder burrow]

holy locked down

[elder burrow]

gpt 5 high is the slowest model on lmarena

[wicked sage]

wtf is sorting-hat

[wicked sage]

not surprised

[elder burrow]

another stealth model xddddddd

[wicked sage]

this guy might expose lmarena

this guy might be the owner i cant prove it

[versed nova]

what is seedream-4-fal?

[spare rune]

Idk

[wicked sage]

@elder burrow from what i see claude opus 4.1 thinking is cooking

[spare rune]

It’s cooler

[wicked sage]

ty

[spare rune]

If you use it you become cool

Not cool as me tho

[wicked sage]

gpt 5 high already started

great

[verbal nimbus]

I hope it isn't, because it's not as good as GPT-5

It has factual inaccuracies

[neon idol]

😞

[wicked sage]

okay

this is cool

[candid storm]

@echo aurora Its been a while since the leaderboard got an update?

[leaden laurel]

i might try to guess

i think i did guess stealth model to be seedream 4

before

but its easier on images

[verbal nimbus]

GLM 4.5 > Longcat

[verbal nimbus]

I can test it

[ornate kayak]

Hi there, I'd like to try the video generation bot

[verbal nimbus]

By checking if Pro makes the same factual error as reef. If not, then oceanreef is Flash

[elder burrow]

yo... i figured out which one is truly gemini 3 flash

[gleaming kestrel]

Hi! As a new user, just wanted to say that I'm working with AI since quite some time and I discovered the benefits of LMArena in being able to early test and compare side-by-side models which is very helpful, and this is what brought me here.

[verbal nimbus]

Wow, the results are very surprising

[wicked sage]

<-- gpt 5 high claude 4.1 opus thinking 16k -->

[verbal nimbus]

Pro actually got it wrong like oceanreef

Flash actually got it right

[wicked sage]

lmk which looks better

personally i think gpt 5 high looks better

@leaden laurel jumpscare

[elder burrow]

look at what i just sent

oceanstone is gemini 3 pro

[wicked sage]

hey omnificat

[verbal nimbus]

Why?

[wicked sage]

yay!

[verbal nimbus]

If it is, then I'd be quite disappointed

[elder burrow]

responds with the same name

[elder burrow]

nonono, oceanstone, not oceanreef

[verbal nimbus]

Even GLM 4.5 beats it

[wicked sage]

how do i feed claude opus 4.1, gpt 5 high .css

[elder burrow]

are you sure

btw look, they behave very differently now and i like it

[verbal nimbus]

Oceanreef though, not oceanstone

[elder burrow]

yeah but I believe stone is pro while reef is flash

[verbal nimbus]

Could be Flash

[wicked sage]

wait did you make it so its automatically sending oceanstone reponses

or smth

[elder burrow]

nah

[verbal nimbus]

Flash 2.5 got it right, Pro 2.5 had a factual error

Which is... Odd

[echo aurora]

It's on the way! We're collecting votes.

[verbal nimbus]

Not just a factual error, but one said yes, the other said no

So quite a big difference

[wicked sage]

hey pineapple i got a question

[leaden laurel]

asked some mystery ai to make bytebeat song and it made this: (t>>10^((t>>14|t>>10)&(t>>15&31)))(t>>7&1?127:64) + ((t5&t>>11|t3&t>>13|t7&t>>17)^t*((t>>14&3)+(t>>12&1)))>>2

run it

[elder burrow]

OCEANREEF IS WILDLY DIFFERENT

[leaden laurel]

(its not good)
oops bad copy paste

[wicked sage]

t=0 error: ((t >> 10) ^ (((t >> 14) | (t >> 10)) & ((t >> 15) & 31))) is not a function

[leaden laurel]

how do yall get it so often

[leaden laurel]

atleast i got phoenix

onc

e

[wicked sage]

yeah true

WHAT IS PHOENIX

[leaden laurel]

idk

[verbal nimbus]

How often are you getting it?

[elder burrow]

labubu phoenikx

[wicked sage]

i am the owner of the 24 karat golden... labubu

[elder burrow]

almost every prompt for some reason

lmfao

[verbal nimbus]

Oceanreef sucks tbh

[elder burrow]

it does

[verbal nimbus]

Like even GLM beats it

[leaden laurel]

i mean phantom

[verbal nimbus]

Wish I got oceanstone

[elder burrow]

oceanreef is consistently like this, too

[wicked sage]

you guys impress me

how are you finding ts

[verbal nimbus]

I haven't seen phantom either

Wasn't that an anonymous model way back

It's so easy to distinguish Qwen 3

“It likes to write like this” ✅

In quotes

[verbal nimbus]

@echo aurora

It's not really a blind battle if I can identify Qwen 3 every time, lol

[wicked sage]

overall, i think opus claude 4.1 thinking 16k should be top 1 in webdev

[elder burrow]

oh god

[wicked sage]

even tho the css is... meh its still fast and gets the job done

[elder burrow]

i just got the whole system prompt of oceanstone

[leaden laurel]

i think sonnet 4.1 is soon

i told that in other servers but

[verbal nimbus]

Whaaat

[wicked sage]

SHES SONNET 4.1? ZAYUM

[leaden laurel]

i think mystery "carrot" model in anycoder is sonnet 4.1

[elder burrow]

i'll send in websim 😅

[verbal nimbus]

How are you guys encountering it so often 😂

[leaden laurel]

what is this

[wicked sage]

hey omnificat

what prompt do you use for finding stuff like this

its kinda cool ngl

[wintry tinsel]

When was this implemented and has it been tacked on sonnet 3.5 and 3.7 too?is this on API too?

[wicked sage]

@elder burrow pls...??

[wintry tinsel]

I did notice the quality didn’t feel quite right but I wasn’t sure if it was my system prompt or what

[verbal nimbus]

Maybe Baidu?

[echo aurora]

I wasn't able to replicate this, do you have an example prompt you'd be able to share? Ultimately, we do care if there are ways users are able to skew votes in a certain way that'd create inaccurate leaderboards. We're constantly on the hunt for preventing votes that aren't authentic human preference.

[verbal nimbus]

It's pretty easy to replicate, let me try to cook one up

Usually programming questions

[strong basalt]

baidu ernie x1 - probably an identical version but faster and/or quantized

[leaden laurel]

ok

[pallid raptor]

flux-1-kontext-dev to noSomething went wrong with this response, please try again.

[wicked sage]

https://tenor.com/view/my-honest-reaction-yahiamice-gif-15470270493139603428

[leaden laurel]

the multiple mosue

mouse

[verbal nimbus]

Here's a dumb one I cooked up (dumb since it doesn't really make sense):

What does the receiver do when a duplicate TCP packet is received? Analyze which approach is best regarded in terms of performance and congestion control.

Got Qwen 3 Max to write in quotes again. The more complex the question, the more likely it is for quotes appear, I think.

It should be easy to find in the dataset. Filter/grep for > “

(Edit) See this to this message for a better example + screenshot: #general message

[gleaming kettle]

i still believe nano is better than seedream 4

[wicked sage]

i love lmarn ea

[glossy umbra]

give nano a man with a hat. tell nano to make the hat scaled bigger and change nothing else . it will either not change anything or completely change everything that you didnt ask it to

it is not good with retaining objects while editing their attributes

sure it can take an object and add it into an image or remove one, but it cant work with them that well

[obsidian current]

how to make video

[strong basalt]

does someone know what is the "lmarena-internal-test-only" ?

[verbal nimbus]

@echo aurora Here's an even better one:

What is the best approach when it comes to exchanging private keys remotely and securely?

😃 Summary

“As seen, Qwen 3 models are very recognizable because they write like this.” ✅

[teal mantle]

Is it just me or gpt 5 instant feels like they backtrained with 4o synthetic data?

[strong basalt]

gpt-5-instant/chat are trash

they're on gpt-5-nano lvl

[wicked sage]

android or iphone

im deadass r

rn

[glossy umbra]

Dont even use it. claude 3.5 sonnet is better than gpt 5 chat.

You'll end up arguing with a wall if you use 5-chat.

[wicked sage]

oh it is?

thank you for telling me this dude

youre a life saver

[glossy umbra]

I know you're sarcastic but it's really weird to see . Even gpt 5 mini is better than chat.

[verbal nimbus]

I still haven't gotten oceanstone 🤪

[wicked sage]

no i swear im not

i mostly use gpt 5 chat, i didnt know it was this bad

it was either gpt 5 high or chat

again, ty

[glossy umbra]

No worries man
GPT-5-HIGH is very good , but slow. If you need quick answers but not on the joke level of 5-chat, I suggest using Sonnet 3.5 .

[wicked sage]

yeah that gpt 5 high statement is true as hell

ty!

[verbal nimbus]

Oh just got it. Oceanstone > oceanreef

But about on par with Qwen 3 🤷

I feel like it's a Flash checkpoint

Doesn't feel like Gemini 3 to me

[verbal nimbus]

If it is, then would be disappointing

[burnt sinew]

What are we in 2024

[wicked sage]

Humans

✌️

[burnt sinew]

???

[wicked sage]

idk what to respond with so thats what i said

also

claude sonnet 4 thinking 32k
or
claude opus 4.1 thinking 16k

[glossy umbra]

No. I'm really trying to point out that even Sonnet 3.5 is better than gpt-5-chat.

[wicked sage]

ts might be better

[limber marlin]

Hello

[wicked sage]

hi

[burnt sinew]

Let's check

[burnt sinew]

Wrong, by so much

1368 vs 1430

[wicked sage]

so sonnet 4 thinking 32k is better?

atp i might short it to sonnet 4 32k

[burnt sinew]

Sonnet 4 non thinking is better

if you need speed

[empty stump]

is gpt 5 chat the instant model n chatgpt?

[wicked sage]

whats the diffirence to

oh

so thinking is just more time but better responses
non-thinking is less time but.. not sure if they are bad responses

if someone can help me, which is better opus or sonnet?

[sullen quest]

Opus

[wicked sage]

idk which is better

[wicked sage]

ty

[verbal nimbus]

Also, Longcat talks like this, but it overdoes it (multiple consecutive quotes).

It just have used Qwen data

[wicked sage]

what does this string of numbers mean

[empty stump]

8/5/2025

?

[wicked sage]

oh

im stupid

[burnt sinew]

Opus 4.1 thinking

[wicked sage]

ty

[wicked sage]

ty

[verbal nimbus]

E.g.

why is it putting everything in quotes, lol

[wicked sage]

l o n g c a t

my question is where you found this

[verbal nimbus]

Random battle

[wicked sage]

ok

ty

[verbal nimbus]

Should be on direct mode too ig, since it's not anonymous

But the other Chinese models are better

[wicked sage]

w model name

[verbal nimbus]

Lol

Logcat

[wicked sage]

first try non-direct chat

nice

[wicked sage]

interesting

nvm i see why opus is good

back to back

[elder burrow]

yeah

its pretty bad though

[wicked sage]

yep

[verbal nimbus]

Haha I'm not encountering it now

[wicked sage]

its funny seeing random ass models i have never seen

[elder burrow]

they only have a flash model and they trained on other models

i hoped longcat was a good model

cz u know