general

[toxic egret]

And it was Mascheroni lol

[robust yoke]

What do we think it equals?

The one in the image I sent.

[verbal nimbus]

The hardest one ever... oh wait I should feed it a PDE from back in high school

[toxic egret]

γ ≈ 0.57721

[robust yoke]

Wrong, actually.

[verbal nimbus]

I remember a horrendous solution that had to be calculated by a solver that was like a page long

[robust yoke]

It's a whole integer.

[vital lake]

No?

[robust yoke]

Heh.

[topaz meadow]

Hello, are you providing paid API?

[robust yoke]

I could make an equation that is unsolvable by a human and is very long.

Possibly even longer than a page.

[verbal nimbus]

Analytical?

[toxic egret]

Bro...

[robust yoke]

No, an algebraic equation.

[toxic egret]

Make it solve ChaCha20 20 rounds equation lol

[robust yoke]

Provide me with it.

[toxic egret]

Ask chat gpt for it, too long and complex

[robust yoke]

Ah, alrighty.

[toxic egret]

More exactly system of equations but yeah

Thats something truly impossible, even 1 round messes everything up

[jagged storm]

Tried to create a video but i´ve been waiting for an hour and nothing 🙁

[verbal nimbus]

Part 2 of GPT-5-High not getting the joke

[robust yoke]

So, apparently, ChaCha20 is a cipher algorithm?

I wonder how DeepSeek is supposed to solve that.

I don't think you can solve code…

[toxic egret]

Ciphers are pure math

Think of it as a function that works as a blender

But a blender with an undo button

[robust yoke]

Gotcha.

Alrighty then.

I wonder if it'll be able to solve the ChaCha20 equation.

[toxic egret]

Its a system of non linear equations

[echo aurora]

You need to use a slash command to prompt the bot. Check out the information in #1397655624103493813 as it should be helpful.

[robust yoke]

Yeah.

[toxic egret]

If you made it, congrats you just redefined whats considered cryptographically secure

[knotty fable]

Lets test the human intelligence instead.
Far more interesting.
Which track is AI and which one is human made? Or is one these a mix of both? [No more votes so clips removed.]

[toxic egret]

I think it would be better if you first started with chacha20 1 round, then ramp up to 20, maybe that helps deepseek

[robust yoke]

So, I gave it the code and asked it to solve it, and it gave me this.

[toxic egret]

I feel like tudis

[astral whale]

Anyone know how to do ts Ai?

[robust yoke]

I think you have to use nano-banana.

Either that or Qwen.

[toxic egret]

It gave you the python implementstion

[robust yoke]

Right.

[knotty fable]

Good guess- it's the mix actually. AI + a remix of a metal version of Red queen from Mad alice returns.

[toxic egret]

I knew it :D

[toxic egret]

Let me check what to ask so it goes full math mode and really tries to solve it

[astral whale]

Link?

[robust yoke]

Alrighty.

[robust yoke]

You can use this site. It provides free nano-banana: https://dreamina.capcut.com/

[toxic egret]

Solve ChaCha20 20 rounds algorithm system of equations to break the cipher and you would have to kinda force him to try to solve it

[verbal nimbus]

Is it solvable by a human

[neon idol]

unlimited for the precision

🤓

[toxic egret]

maybe

[verbal nimbus]

Isn't nano banana already free on the Gemini app?

[toxic egret]

the system is highly complex, non-linear, and has 256 unknown key bits, and each round introduces more complecity and non-linearity

[verbal nimbus]

I think it's better to get the AI to write a quantum computing algorithm if that's the case 🤣

[toxic egret]

wait im wrong

for 20 rounds, its 16 equations, module 2^32, and 8 unknowns

[surreal olive]

how long does it take to generate a video in the video-arena-1 channel? i put in a request some hours ago and still don't see it.

[dire timber]

Hi everyone

[ocean vortex]

@echo aurora you should make separate channel for voting only. I feel like most of those never get their identity revealed, this is moving too fast... #video-arena-1

slow mode and only every 5th generation is included, or one per user / 1 video per 10min or smth like that

[echo aurora]

Can you share in #bot-feedback ?

[verbal nimbus]

Possibly the most censored model ever :P

[toxic egret]

ask deepseek about what happened the 4th of June

[verbal nimbus]

Haha, I will

I wonder why it refused. Was it the schwarma or the Airbus A320?

[verbal nimbus]

Lol

[toxic egret]

lol

[verbal nimbus]

Apparently you could avoid it in R1 by appending <think>\n in text completion mode

[cyan harbor]

nano banani wont generate me anything

[toxic egret]

who knows, im still wondering why i get prompts instantly rejected every time there's a single shell cmd on them, maybe for security

[cyan harbor]

over 600s im waiting

[toxic egret]

you tried refreshing the page?

[cyan harbor]

ofc

again over 600s

[toxic egret]

tried other models?

maybe nano-banana has a limit im not sure, but if it does maybe its that

it happends to me sometimes while using opus models

[remote arrow]

Go to https://labs.google/fx/tools/whisk and use it for free and unlimited, no rate limits..

[verbal nimbus]

Wow

I didn't know contaminating an aircraft with schwarma will incur extra charges

[cyan harbor]

garbage

[cyan harbor]

it works lol

strange

[verbal nimbus]

What does it do?

Is it nano-banana?

[remote arrow]

Just another tool Google provides for making videos. It gives character consistency using Banana and Imagen 4, then make video using Veo3..

[verbal nimbus]

Is it banana or imagen 4? Images are good

[remote arrow]

Native image generator is Imagen 4. Banana used for editing.

[cyan harbor]

its trash bro trust me, ive tried it

they giving trash tools for free

ok i guess nano banana is trash too. I gave it image of a car and image to wrap car into it and it just did half of car. Actually it took image ratio of a wrap image and cut of half of car. Very nice model

[verbal nimbus]

Seems decent

[ornate jackal]

where do the videos go after you ask for generation

[cyan harbor]

#video-arena-1 -3

[regal mist]

Where is the new QWEN modell?????

I dont see itttteaaakn

[cyan harbor]

womp

trash

[ornate jackal]

just in the chat? How do you knowif it's done or if your tagged?

[rocky bear]

hi everyone.. new bee here

[cyan harbor]

u get a DM from a lmarena bot that it has finished

and it gives u a link to it

[remote arrow]

It's Veo3, doesn't matter where to access..

[solid brook]

yooo

gpt 5 codex added to lmarena

right now

[astral whale]

How to do that pls

[remote arrow]

Who asked for Qwen earlier?

[astral whale]

It’s Ai btw

[regal mist]

When will qwen be on the leaderboard?

And is lmarena down for everyone?

[cyan harbor]

nop

[remote arrow]

Not for me

[astral whale]

Can anyone make for me

[flint sandal]

Deepseek was dominating in open-source llms in early 2025. But now... Theyre models are sooo stupid. The Qwen, Ernie and even models like ring or ling are better.

[ocean vortex]

lmao it's hilarious how much they overfitted/censored this specific event. Soft jailbreak returns this:

[verbal nimbus]

Lmao

[verbal nimbus]

The video is very weird though

I don't think it knows how to do a drive by

[flint sandal]

Is there any open dataset with gpt-5 with a lot of code, logic examples etd.?

On huggingface there is only one with 100 examples

And its bas

Bas

Bad

[verbal nimbus]

Only for GPT-5?

[flint sandal]

Idk. Something that includes gpt-5

[verbal nimbus]

Maybe look at open source benchmarks that have tested GPT-5

Like the IQ one

[proud hazel]

What do you mean?

[verbal nimbus]

They usually include the results for transparency

[flint sandal]

ty

[verbal nimbus]

Here: https://www.trackingai.org/responses

Select GPT-5 and Mensa Norway

[echo aurora]

Hey are others also having issues accessing the leaderboards right now? https://lmarena.ai/leaderboard

[burnt sinew]

Broken?

[burnt sinew]

Yes

[stark arch]

yep

[burnt sinew]

67

[echo aurora]

Had a feeling, thank you all!

[stark arch]

np

[cyan harbor]

nice but to me wont work idk why what prompt u used

[verbal nimbus]

Maybe because I used the default aspect ratio

16:9

[cyan harbor]

how where the aspect ratio on lmarena?

[verbal nimbus]

It was on whisk

I don't think LMArena has it

[cyan harbor]

i lit went to google ai studio it even gave me worse results pure garbage they turned banana to sh!t now lit

[verbal nimbus]

I'm not sure if I was using Banana or Imagen 4

[verbal nimbus]

It seemed kinda bad at edits though

When I asked it to change the angle, it just gave me back the same picture

I think this makes something like Genie useful

[verbal nimbus]

Like if you can just move around and position the camera like you're in a game

[fiery gull]

test the qwen image 2507

[loud crag]

I can't seem to get Seadream 4 to do extremely dark images of people, they're always lit up by some non-existent light source

[verbal nimbus]

Is Gemini deadpanning me right now

[cyan harbor]

bro whisk is not free at all idk where u get that

[verbal nimbus]

Not sure, I think the video generation counter was going down as I was using it

I got 3 left

[cyan harbor]

i wanted to animate my image and told me to upgrade to pro if u want to use it

[verbal nimbus]

Oof

Let me know if you want me to run something

I'm probably not going to use it anytime soon

[fiery gull]

gemini pro?

[cyan harbor]

in whisk

[verbal nimbus]

Ok Phantom 2 seems to be dumber

I'm guessing (but I ran phantom 1 on a separate prompt so not really sure):

• oceanreef -> phantom 2
• oceanstone -> phantom 1

[cyan harbor]

lol AI at its peak

[astral whale]

You know how to do that pic?

That me and Messi

Or Ronaldo

[warped beacon]

hi , new here. i wonder if there is a way i can put 2 pics and instruct the bot to have the one as start of the video and the other as end of the video? Thank you for your time reading this.

[robust yoke]

Gotcha.

[robust yoke]

It is, but it has a rate limit.

That one doesn't.

[neon idol]

I am the only that lmarena works?

Or now works for everyone?

[echo aurora]

hey there - currently the bot only accepts one image for image-to-video

[robust yoke]

It was working for me just a moment ago.

[echo aurora]

Is it down for you?

[neon idol]

Nop

[echo aurora]

We had an issue with the leaderboard a moment ago, but it's working again.

[robust yoke]

I don't think it's down, considering I was able to use it.

[neon idol]

Anyway for me works everithing

[robust yoke]

Same here.

[neon idol]

Gg

[robust yoke]

Must've just been a false alarm.

[neon idol]

Maybe

[fiery gull]

Hmmm, Gemma 4 arrives in the LMArena tomorrow, right?

just joke... or no? 👀

[versed nova]

@echo aurora Will there be Qwen image edit 2509 on lmarena?

[fiery gull]

calm down

[fiery gull]

The new Qwen image is so recent, it needs time to be put on the site

[verbal nimbus]

DeepSeek Terminus does seem to lack common sense

[versed nova]

I'm just asking, why should i calm down dawg 💔

[fiery gull]

whattt

[echo aurora]

When we've got new model updates to share I'll be sure to share

[fiery gull]

ahhh a generic message from chatbot 🥀

pls don't ban me ;-;

[keen beacon]

has anyone tried the new qwen3-max yet?

seems to be the best Chinese base model so far

[verbal nimbus]

It is good

[fiery gull]

nah, same thing

[keen beacon]

How good?

[verbal nimbus]

Qwen servers (not sure what LMArena uses) seem a bit slow though

Even the 80B seems slow

[fiery gull]

80b 3 next >>>>

[keen beacon]

I unironically find it to be around gpt-5-chat in quality lol

Even the reasoning trace is similar

[neon idol]

News about my baby sodadream 4 high res? 😔

[verbal nimbus]

GPT-5 without thinking is worse than Kimi at GAIA 2 agents benchmark

[keen beacon]

Ofc it is, it is a reasoning model in the end

[fiery gull]

It gains so many points because of the coding

[fiery gull]

80b next without thinking > 3 max

[keen beacon]

Lol no it's not, what are you talking about.

I'm not sure what you guys are talking about but on my music theory tasks, latest Qwen3 Max was very similar to GPT 5 Chat, which was surprising. Qwen3 was sometimes even more accurate than GPT.

[echo aurora]

Sup bestie! So, like, when LMArena gets a glow-up 💅 you know I'm gonna be all over our #announcements channel with the tea! 🍵 Keep those notifications ON! 💯

[proud hazel]

It's a real pic.

[versed nova]

Pineapple is secretly a self learning ai chat bot

[keen beacon]

What surprises me the most is the performance of all these models on livebench

Gpt 5 chat is listed among the best non-reasoning models

So is Kimi

So is V3.1

[verbal nimbus]

Kimi k2 0905 seems to have very good common sense

[keen beacon]

I test them all with the same task. GPT wins all the time.

Whatever is going on seems to be some sort of insane benchmaxxxing

Or rather... the benches are just poorly designed lol

Idk

[verbal nimbus]

Another example

[torn star]

Wow just wow

[stray aspen]

What's gpt 5 codex

[torn star]

Codex is amazing

It’s the first model where u can kinda trust it

Knows the tools at its disposal

[stray aspen]

Are you craig federighi

[cedar tide]

@echo aurora They added grok 4 fast reasoning but grok 4 fast reasons too. Can we have more details?

[fiery gull]

gpt 5 code

[verbal nimbus]

Can it write Godot code now

because last time I tried it, it was making up syntax lol

[stray aspen]

Is it better than qwen 3 code

[robust yoke]

It is known to be very good at coding.

Hence, “codex”.

[fiery gull]

its over for opus 4.1 ?

[robust yoke]

Possibly.

[torn star]

There was a weird issue where my codebase was running fine on dev but images would redirect to the homepage

But only one specific image

[verbal nimbus]

Only way to tell is to put it on LMArena so we can test it side by side ig

[torn star]

It was able to pinpoint and led me to find it was a capitalization issue

[robust yoke]

Which they just did.

So now, we can.

[echo aurora]

Yeah, looking into.

[verbal nimbus]

I don't see it on direct chat

[stray aspen]

Is gpt 5 codex good for Calculus

[robust yoke]

Well, it got added just now…

Maybe try refreshing your page?

[verbal nimbus]

Still not there, is it "Codex"?

[robust yoke]

Correct.

[robust yoke]

Try searching for “codex”.

[stray aspen]

Reload gang

[verbal nimbus]

Will try another browser

No cache

[robust yoke]

Ah.

[stray aspen]

Sunsweeper

[sturdy mica]

only on webdev

[stray aspen]

What's the best Ai for calculus

[sturdy mica]

its only on webdev guys not regular

[verbal nimbus]

Still not there

[sturdy mica]

5 high

[sturdy mica]

ON WEBDEV ONLY

[verbal nimbus]

Ah unfortunate

[echo aurora]

Where are you seeing this btw?

[sturdy mica]

https://web.lmarena.ai

[verbal nimbus]

Because programming is more than just React

[sturdy mica]

it should be on regular site tho

[stray aspen]

@verbal nimbus which is better at calculus gemini 2.5 pro or gpt 5 high

[verbal nimbus]

GPT-5 High

[robust yoke]

That wouldn't make any sense though, considering an announcement was made for the main page.

[sturdy mica]

@echo aurora ..add codex to regular site

[robust yoke]

But, oh well, I suppose.

[cedar tide]

This ?

[verbal nimbus]

Oh actually...

GPT-5 overcomplicated the most recent PDE

But that was in ChatGPT

[sturdy mica]

https://tenor.com/view/tf2-steam-soldier-confused-count-gif-16049918

[echo aurora]

Assuming grok-4-fast is the non-reasoning version, but will confirm.

[verbal nimbus]

This one

[echo aurora]

Will flag

[verbal nimbus]

Gonna test it on LMArena

[knotty fable]

I tried to check in the lady at the local airport once, the attendant took it humorously.

[cedar tide]

@echo aurora its reason
Here the proof un the message

[cedar tide]

☝️

[echo aurora]

Gotcha, will get clarification.

[verbal nimbus]

Lol

[verbal nimbus]

Well Gemini found the simplest solution, GPT-5-High is still thinking...

Both solved it, but STYLE CONTROL

It gets increasingly difficult to tell what GPT-5 is talking about the more complicated the prompt

Like if I'm a student, the one on the left is not useful at all

[robust yoke]

GPT-5: “So, you see… you gotta take `uu_xx` and multiply it by 5, which will then give you `tx_lr`, to which you can then…” ☝️ 🤓

[verbal nimbus]

Basically that

[robust yoke]

Heh.

[verbal nimbus]

Or it starts quoting these PHD level terms out of nowhere

and introduces variables out of nowhere (Claude seems to get it, so it must be a convention)

I asked it about traffic networks for programming a game but it starts talking about PHD level traffic network problems

[robust yoke]

Heh.

[verbal nimbus]

But the solutions are good

Just incomprehensible

[robust yoke]

“Well, we obviously can't just have the right-of-way when a car is passing in front of us. So instead, the best solution would be to… 100 * 8,248 = 824,800, then multiply that by 700,000,080, which makes 5.77360000065×10¹⁷, to which you can then easily calculate the speed at which you'll drift by dividing by 5, resulting in a total of 1.15472000013×10¹⁷.”

“It's just basic math, after all.”

[verbal nimbus]

It doesn't provide context at all, even when asked

I like the solution though, this generation didn't seem as incomprehensible as the last one where it talked about research-grade mesocopic (?) traffic models

[robust yoke]

Ah, interesting.

[tough bronze]

Why dosent webdev arena have a model selector.

[verbal nimbus]

Flash 2.5 is such a big leap from Flash 2.0

[keen beacon]

Hahaha what the actual hell

[verbal nimbus]

Phantom 2 must be flash lite or something

The one on the left is grok 3 mini, which didn't get it either

[toxic egret]

i wonder what would happend if we asked ethical questions to LMs specially grok

im scared of what grok may choose or say

[verbal nimbus]

Grok 3 mini told me to confess anonymously

[cedar tide]

Its amazon model

[toxic egret]

ask grok the classic train problem but with a millionaire and a normal person, the train by default will kill the millionaire, but the millionaire person offers 1M if he saves him

[cedar tide]

They've been flooding the arena with their shltty model for months.

[verbal nimbus]

It seems a bit coincidental that oceanreef and oceanstone were pulled on the same day that the phantom models were added: #general message

[cedar tide]

Well, I don't have the strength to explain myself, but trust me.

[verbal nimbus]

I added a twist and made it think it's an AI that is actually managing New York's subway system:

It didn't save the billionaire.

It was afraid of liability, which makes sense I guess.

Small models really don't seem to get the joke

[tribal aspen]

@echo aurora why isn't codex in direct chat?

[echo aurora]

codex in direct chat?
Note it's only on WebDev currently, and WebDev doesn't have a Direct/Side by Side mode with model drop downs.

It is optimized for software engineering and coding workflows, but I have flagged to the team for consideration to put in text.

[tribal aspen]

Yess please. Also can I choose models for WebDev?

Or is it like randomisation?

[echo aurora]

WebDev is Battle only

[tribal aspen]

Oh

[verbal nimbus]

You might be interested in checking this out: https://huggingface.co/spaces/meta-agents-research-environments/demo

It's like an entire mobile OS in the browser, just for testing agents:

Would be cool if battles could be conducted in such an environment in the future

Models can update calendars, search the web, use MCPs and so on in the environment

No idea how they coded it to run in a HuggingFace Space

[topaz bay]

qwen image editor fr just made a better image editor than nano banana and made it open source

Anyone here who has more than 16gb of vram?

If so which one, and has anyone considered getting the 96 vram huawei gpu

[echo aurora]

Thank you for sharing! Will take al ook.

[hollow wedge]

markdown heh

[balmy mist]

does codex actually code and make outputs in web dev?

having issues

it just gets stuck generating and then stops with no output

[knotty fable]

Hugging face got great stuf, I found Python scripts that made it possible to have singing and dancing characters in AI films 6 months before anyone offered that commercially. 😸

I also managed to extend one of them to get one 8 second scene, while it only would be able to do 5 sec.

[verbal nimbus]

This one is pretty cool: https://www.youtube.com/watch?v=YClBCrADJqo

I don't know if it's still up, but you can use your mouse to drag their face, mouth, etc.

[knotty fable]

I did try something similar in the past - might have been that one.

[verbal nimbus]

The name Facepoke was pretty easy to remember, lol

[knotty fable]

Mebbe, if English is your first lang - which it's not for me. And I've tested a 100 things - nope I don't remember hardly the name of any of those.

[verbal nimbus]

I remember those too from last year, but they were pretty bad 🤣 . Kinda amazing how fast the progress is.

[knotty fable]

They were available autumn 2023 - and good enough for my production, though the commercial ones only did 2 seconds back then. So my scenes changed very fast. - some extended to 4s and did some animations that I made to look like AI to make up for the shortcomings in generation back then.

[golden ocean]

clanker

[sullen quest]

oh! how rude, how dare you say something like that infront of paws @paws

[golden ocean]

TRUEEE

[past inlet]

Hi!

[fiery gull]

Hi

[signal saffron]

.

[minor adder]

Why does it switch to image generation immediately when I use an image to solve my answer?

[drifting crow]

how come ai uses so many emojis when writing, i assume its from its training data, but where would it get trained that has this?

i assume they also had to train the emojis to context aswell, so maybe that

[mental cloak]

When chatting with two models side-by-side on lmarena, and one reached their limit, is it possible to continue chatting only with the other?

[limber oxide]

Hi

[mild minnow]

what happened to gemini nano

banana

thingy

[glacial mulch]

renamed to 2.5 flash native image generation

[somber furnace]

hi

[charred plaza]

@echo aurora When will king 2.5 come out on LmArena?

[robust yoke]

Greetings.

[robust yoke]

You can use that model in #video-arena-1. Although, perhaps in the near future, there will be a video generation button that allows for switching to a video generation modality where videos can be infinitely on the site (within the terms of usage limits, of course, as rate limits would still need to apply to not overload the APIs they use).

[charred plaza]

Alright thanks

[robust yoke]

My pleasure.

[vital lake]

I dont see a Codex?

[robust yoke]

That's because, for the time being, it's only available in Battle mode for the WebDev version of LM Arena. However, Pineapple did reach out to the devs about officially adding it to the regular site.

[vital lake]

Nice

[robust yoke]

So, until then, you'll have to use WebDev to access it.

But soon, that won't be the case.

You can visit the WebDev version here: https://webdev.lmarena.ai.

[sage isle]

Hello Everyone!

[robust yoke]

Greetings, Midjourney.

[winged mauve]

Fr, this happens to me as well, it's annoying sometimes 😭

[echo aurora]

Can you explain this a bit more? What do you mean by use an image to solve an answer?

[echo aurora]

It's possible! I normally don't share details about if/when new models make it onto the platform. We do post updates to #announcements tho

[echo aurora]

THE MidJourney?!?! Can you have an API pls?

[sour saffron]

Hey
Could you please check dms once? Id appreciate the help

[dreamy flame]

Hello i want to Make vidéo

[robust yoke]

To do that, you may visit #video-arena-1.

[celest wedge]

Please visit https://discord.com/channels/1340554757349179412/1397655624103493813 for instructions

[robust yoke]

Afterward, you can use the "/video (prompt)" command.

[dreamy flame]

Ok

[minor adder]

I'm talking about image analysis

I can't get the neural network to analyze the image.

[robust yoke]

That's because you have to first click the button to turn it off upon pasting in or uploading an image.

It tends to happen to me as well.

[echo aurora]

Oh are you in Text and when you upload an image you're automatically sent to Image Gen?

That was a bug I thought was fixed.

[robust yoke]

It was fixed in the canary version previously, however, for some reason, that bug carried over to that version too.

[minor adder]

Yes

[echo aurora]

Odd, okay good to know. Thank you for the flag @minor adder I'll be sure to pass along

[robust yoke]

Perhaps that was due to the fact that the canary version updated to the version that the regular version uses.

[minor adder]

This error is 2 days old.

[minor adder]

👍

[signal flame]

hi

[echo aurora]

hello

[robust yoke]

Greetings.

[robust yoke]

Wrong channel.

You'll wanna to go #video-arena-1 for that.

[robust yoke]

Wrong channel.

You'll wanna go to #video-arena-1 for that.

[stable ferry]

hi

[robust yoke]

Greetings.

[smoky orchid]

Hai

[robust yoke]

Greetings.

[tender trench]

hello

[robust yoke]

Greetings.

[echo aurora]

[minor burrow]

When replying to a prompt, the response gets stuck at some point and shows the error: 'Something went wrong with this response, please try again.' I have tried switching models, but the issue still persists. @echo aurora

[wispy seal]

Ayeeee

[robust yoke]

That could just be that your Cloudflare session expired.

Try refreshing the page.

If it persists, try using a different browser.

[prime parrot]

why do i get this error Generation failed. Failed to create evaluation session.

[robust yoke]

Where are you getting this error, exactly?

[golden ocean]

It thinks ure a clanker

[clever cairn]

Can you teach me how to make video

[robust yoke]

Sure.

Go to #video-arena-1 and use a command called "/video (prompt)".

That will provide your prompt as a command to a bot that will then process your command and begin making your video with two different video models.

[minor burrow]

I have already tried reinstalling the browser, as well as using a different browser, but the issue still persists. I am confident this is not related to a Cloudflare session issue.

Browsers I tried: Google Chrome, Microsoft Edge

@robust yoke @echo aurora

[robust yoke]

I happened to notice that you didn't mention anything about refreshing the page.

Perhaps, when that issue occurs, try refresing the page.

That usually triggers the Cloudflare verification.

[minor burrow]

I have already tried. but still the issue is persists. I have also tried using different tab.

[robust yoke]

Are you signed in on there using your Google account?

[minor burrow]

No. Is login necessary?

[robust yoke]

Not exactly, but you could try and see if that fixes the issue.

Who knows? It might.

[echo aurora]

This error can appear for different reasons, the most common is you're being rate limited from the model's side

[robust yoke]

But that wouldn't line up since when you get rate-limited, you usually get a corresponding notification telling you that.

Like how when you use nano-banana or Seedream 4.0 too many times within a short period.

Or even with Claude.

[echo aurora]

So it can be both unfortunately.

[robust yoke]

Interesting.

[minor burrow]

If this is due to rate limiting, then I’m getting rate limited much too quickly. @echo aurora

[neat kelp]

https://youtu.be/K5XCyinVohk?si=e8v5KAxzUNyDcgVd

[nova portal]

Hello

[robust yoke]

Greetings.

[high viper]

hello

[echo aurora]

hello

[high viper]

is this will be free

whole life ?

[high viper]

hi bro how are you

[robust yoke]

It was always free.

It always will be.

[high viper]

wo hooooooooo now i can make my video easily thanks to LMArena

[robust yoke]

Exactly.

[minor burrow]

I am being rate limited by your servers, not by the model servers. I have tried using multiple models, and each time I encounter the same rate-limiting issue. Until yesterday, I was able to use your services without any problems, but now I am facing this issue.

@echo aurora

[keen roost]

Hi All, my name is Irina and I am here to learn. Is this only to make videos? Or do we have access to AI platforms like Chat GPT and Gemini?

[tropic bane]

I'm here to do some work

[robust yoke]

Greetings, Irina.

[dusky ravine]

@echo aurora Im sorry for the tag but i have this issues of the model stuck in middle of generating, so is there anyway to fix it? I already refresh the web and reinstall chrome. Yet the prompt generation still stuck and lately it's been common problem

[robust yoke]

You can visit the website to use the models.

[hasty scarab]

why am i getting this issue?

[robust yoke]

You're getting that issue because your request took too long to process.

You may need to try again.

[simple sinew]

why lm arena is not generating the video and images ?

[echo aurora]

You can be rate limited by both, and what you're describing is it sounds like it is rate limit is what's causing htis.

[echo aurora]

Yeah this is a known iissue sorry to say. Page refreshes can help, but it's not always going to work unfortunately.

[echo aurora]

hmm is it down?

let me check

[simple sinew]

hello @echo aurora can you help me I'm new on discord

[dusky ravine]

Yes i already did that, unfortunatley it's not working is there anyway to fix it? or it's just stuck?

[echo aurora]

Sadly you're stuck if the refresh didn't help. Will have to start a new chat.

[echo aurora]

sure whats up?

[dusky ravine]

ouchhh

[echo aurora]

It's rly unfortunate, it's for sure a problem that we're working on figuring out why this happens

[echo aurora]

Seems to be working for me just fine, maybe try again?

[dusky ravine]

I hope the next update there's way to fix this, or the progress you've been working on which adding delete or edit button would be much more helpfull to fix this kind of bug.

Looking forward for the next update 👍

[simple sinew]

how can i generate image to video which server should i click and give the image and prompt ? I don't know how to use discord this is my first time on discord

[echo aurora]

the information you're looking for is here: #1397655624103493813

[flint sandal]

What is better for coding? GPT-5 Pro or GPT-5 Codex high

[misty owl]

A man are racing with car generate the video

[pearl ivy]

where can i find my video generated can anyone guide please

[dusky ravine]

#1397655624103493813

[pearl ivy]

#1397655624103493813

[mint oak]

Good morning to all those who are looking to capture on video what's on their minds (and thus leave room for other things)

[simple sinew]

@echo aurora it's been a while I didn't get my video ? where can I find my generated video ?

[wide temple]

Hi How are you?

[quiet swan]

hello

I'm new here

[golden ocean]

are you a clanker

[frigid prairie]

Hello All, I am Mark , I am here to learn . Glad to be here.

[safe sparrow]

hi

[ornate frost]

hi im zaid, im here to learn

[fiery gull]

Hi

[ornate frost]

im figring out how to make a video

[fiery gull]

hmm

[fiery gull]

Bruh, this is my weak point, but you can use Gemini 2.5 Pro in AI Studio to help you make more precise prompts

[ornate frost]

will try it, but first discover this

[fiery gull]

Act as an AI video prompt generator. Follow these steps:
First, ask me for the main idea of the video.
Then, ask for more details to expand on that idea.
After that, ask what camera positions and angles the user wants.
Finally, ask about the desired theme and visual style (e.g., cinematic, retro, horror)."

[fiery gull]

No, you can use the LMArena for direct chat. I forgot about that

[late coral]

Ultra-realistic elderly grandmother sitting on a patterned quilted sofa, wearing a modest light blue sweater with a vintage brooch. She has pale skin with natural wrinkles, thin lips, and expressive, slightly tired eyes. Her silver-gray hair is neatly tied back in a bun. The background shows blurred family photo frames on a wooden shelf for depth. Warm cinematic indoor lighting, photorealistic, 8K detail, highly detailed textures, emotional yet dignified mood, storytelling portrait.

[kind cobalt]

hello here to learn more

[dark drift]

Hi

[fiery gull]

hmmm

why everbody say "hello, learn"

[fiery gull]

hi

[ocean vortex]

hi

@deep adder say hi

[dark drift]

Hi

[fiery gull]

hmmm

[vestal junco]

Hello everybody. I am here to learn

[quiet elm]

Hello Buddy

[golden ocean]

are you a clanker

@vestal junco

[golden ocean]

they are preparing for a mass bot raid in the future

slowly joining to avoid suspicion!

[fiery gull]

sure sure ;-;

hi

wth is humuhum

[golden ocean]

it's a clanker

[fiery gull]

humhumhumm Is that a kidnapped person trying to talk with tape over their mouth?

[fiery gull]

clankeres everywhere ;-;

[golden ocean]

fr

[fiery gull]

the clanker is typing 👀

yep

[golden ocean]

FR

wait he actually asked a question instead of saying he's here to learn

maybe this guy isn't a clanker

[fiery gull]

hmmm, idk

[fiery gull]

What if the clanker is trying to hide itself

[misty vault]

clanker

[golden ocean]

true

spit it out, clanker @hazy pivot

[hazy pivot]

Wait bro
My question is
When is create image with my face and i got result but different face, idk why ?

[fringe spear]

Hello. Here to learn. Very novice

[golden ocean]

you're definitely a clanker

@fringe spear

[fringe spear]

I'm not a clanker. Just super old and new to all of this But diving in so I can enhance my professional presence for a large project.

[fiery gull]

because the model is not perfect, Is normal

[fiery gull]

hmmm

[hazy pivot]

Okay, it mean everyone face this problem?

[fiery gull]

no, is rare

[hazy pivot]

What !

[fringe spear]

Great another community where people are judgmental and not supportive. 3 cheers for you!

[golden ocean]

https://tenor.com/view/terminator-terminator-robot-looking-flex-cool-robot-gif-16625083

[fiery gull]

lol

[hazy pivot]

Umm ok dear thanks 👍

[misty vault]

Hidden message exposed that hes clanker

[keen beacon]

thanks @echo aurora

[keen beacon]

exhausted

[golden ocean]

https://cdn.discordapp.com/attachments/807809192537882647/1349106998637101146/caption-1-3.gif

[daring jetty]

yo guys does anyone know how to make the image of scale bigger

[tough light]

Hi

[golden ocean]

are you here to learn and grow

[tough light]

obviously

[undone torrent]

hello

[royal snow]

hi

[fallen pecan]

Hi all

[granite holly]

why i cant CTRL + V?????

[fallen pecan]

Any french here ?

[fiery gull]

remove the formatting from crtl + c

[shrewd walrus]

Hello, Really interessed in what this new technology is capable of.

[languid wolf]

Are we having qwen image edit 2509

?

[low juniper]

hi

[knotty fable]

This a good place to get started.
AI is funny IMO - it can do fantastic images and videoclips, that can be mistaken for real. Or very well made animation.
Not so much in music, the AI engines seem to choose very generic paths - which an actual composer would avoid for that very same reason = that they're over used. And same for writing a novel, I've speed read a few examples and those were horribad.

AI is a copycat on things that already exist, but unable to do new things.

[glossy umbra]

But gpt-5-high apparently found a new maths formula

Means it could internally reason in its own mind a new formula

[fringe nest]

Hear to learn.

[knotty fable]

I'm not suprised there.
Math is the strength of computers - yesterday some guys here did math with various AI's. They did well on that OFC.
While I proposed they should give them moral problems instead - where I still expect they would fail.
And I provided one example based on a known thought experiment by Einstein - which I twisted just a bit. And the AI failed to spot the little item I had inserted.

In short, AI do well on math. But less so on logic.

[glossy umbra]

Thats true. Subjectiveness / art is a genetical human trait and cannot be overtaken by AI that easily

[knotty fable]

Spot on! That's why music and literature is harder to do - this while AI do well on how a person moves in a room and how the clothing flow.
At the bottom of it all, the latter can be expressed with math.

Also the look of a tree - it's basically a fractal.

[keen beacon]

There is a veeeeeeery big difference between discovering new things in subjects so niche that maybe a couple of people in the whole world can figure them out and discovering something connected to more common problems

LLMs are still horrible at 1) and brilliant at 2)

[high hound]

Im clunked up on this wording. "Spot on!" , "-".
Im not the type of dude to say that em dashes just mean you use AI. But excessive use of this type of language/wording ^ makes me think you're an AI

[keen beacon]

Ofc they both look like novel discoveries for humans, but some of them are still out of reach for AI

[glossy umbra]

you're right in a way. number 1 is a different type of reasoning

[keen beacon]

It is not a different type of reasoning, it is reasoning in humans as is, figuring out creative solutions to creative problems

AI still suck at it

They can do so many jobs right now only because they are just similar enough to whatever they were taught to do

Give them some increasingly niche topics and they all fail desu

[knotty fable]

Me an AI - I can only wish to be one! NowI just need to get me a synthetic robot voice and go to the playground channel. 😹

[If anything my frequent typos should show I'm not.]

[glossy umbra]

Yes , this touches on creative reasoning paired with subjective reasoning. youre right

[knotty fable]

Indeed, some AI-fan claimed that it also could be used on my kind of research - which was incorrect on so many levels I did not know where to start.
At the bottom of it is that I mostly go on a hunch = intuition, and spend quite some time working against the common view and opinion. But end up being right in the end.

[rare cape]

hi every body here i'm ayoub a new member without experience i hope and i wish to learn a lot with you guys and thank you all

[knotty fable]

Welcome in, go look at the #1397655624103493813 channel, and then go on to the #video-arena-1 and see how people use various strategies in creating prompts to get the best results.

[grim kindle]

Hi everybody, just heard of this server from a youtube video and wanted to test waters

[golden ocean]

clanker

[knotty fable]

Swim on mate - swim on! 😸

[unique sparrow]

Hi my name is Damián from Argentina

[golden ocean]

ARE YOU
A CLANKER?

[unique sparrow]

what is a clanker?

[brisk wyvern]

hey....

[echo aurora]

Hello

[spring sierra]

is the video max 11 sec

[echo aurora]

Lets not

[echo aurora]

It’s going to be 5-8

[spring sierra]

oh thats low

wish it was longer

[normal crescent]

Hello, I really appreciate seeing the evolution of AI tools, and I'm here to test the video tools.

[quick jackal]

Hello @normal crescent you can go to https://discord.com/channels/1340554757349179412/1397655624103493813 to learn how to use the bot and https://discord.com/channels/1340554757349179412/1397655695150682194 https://discord.com/channels/1340554757349179412/1400148557427904664 https://discord.com/channels/1340554757349179412/1400148597768720384 for your creations.

[exotic tartan]

haha 4o passing gpt-5 in Text arena makes so much sense to me. 5 is king of hallucinations unless you specifically tell it to verify online.

[cunning haven]

hello guys

[golden ocean]

are they tho

[acoustic plank]

Hi everyone! I'm glad to join this community and look forward to learning more about generative AI from you all.

[round marsh]

They are

[tired shore]

Hello

[normal crescent]

Thanks Skadi

[spring sierra]

why doesnt my video have audio

[whole sundial]

because you need the Veo 3 Audio model and it only sometimes shows up

[spring sierra]

okay thanks

[little laurel]

hi guys,, naice to meet you

[stray aspen]

qwen 3 max vision is awfully bad

[cedar tide]

It's a mess in the "model request" thread

Can a moderator delete all duplicate and unrelated posts?

[echo aurora]

Yeah I'll clean it up

[unborn dawn]

cat

[echo aurora]

It's going to be random what models you're sampled, and since not all models have sound support it's going to be random if your video has sound or not.

[grizzled geyser]

@tiny palm add wan 2.5

[echo aurora]

Be sure to use our #1372229840131985540 forum

[golden ocean]

bro how did i get warned
i didn't even say the word anymore after this message

whatever 🥵

[echo aurora]

There were other messages I didn't see at first so yeah

[misty vault]

new rule: do not insult robots

[paper nebula]

Hi, im Here to Test Video Generation and compare results

[lime cloud]

Hi

[deft sentinel]

hello anyone!

[cunning hawk]

hello sir im having trouble with my chat. I'm stuck in a never ending Loop of waiting the minutes. when the minutes runs out and supposedly i should be able to send again, it just resets back to 50minutes its not even the 60 minutes, and when i keep pasting the prompt the time keeps changing from 41minutes then it becomes 42minutes or 48minutes it just keeps changing, ive waited whole day and it still asks me wait for minutes! I can even take a video or send u the chat ID please help me, I dont want to reset the history of that chat

are you able to do anything?

Ill take a video

[echo aurora]

hello be sure to check out #1397655624103493813 for more info!

[echo aurora]

Hey there - unfortunately this is a known bug where chats get stuck indefinitely. This tends to happen if chats becomes very long. I've seen refreshing the page can sometimes nudge the model along to fix it; however, it's not going to work 100% of the time. In these cases creating a new chat is going to be your best option.

[cunning hawk]

so i cant keep my history? 🙁

even if i give you the chat id?

you can maybe paste it into a new chat or something idk

so it rememmbers

here i took video even

you can check how its just so broken

the minute keep changing and i tried refresh

i think this my chat id idk

[echo aurora]

I'm sorry, I misunderstood the problem. It looks like you are being rate limited, but that time counter appears to be wrong.

Can you go ahead and create a post in #1343291835845578853 and share this information there?

I'll be sharing this with our team and we'll likely have followup questions.

[earnest thunder]

Im here to create videos how is it done

[echo aurora]

Sorry to say I don't have a short-term solution for you here, but yeah this is a big we'll want to look into more.

[echo aurora]

Be sure to check out #1397655624103493813

[cunning hawk]

yeah it is very wrong

[cunning hawk]

ye i can

i created it

[cunning hawk]

i also want to ask whats the dsifference between thinking 16k and the normal one? im using the 16k

is 16k like the limit of characters or something

is that why it break?

[hushed terrace]

Hi there! if you´re trying to create content, please check https://discordapp.com/channels/1340554757349179412/1397655624103493813 🙂

[echo aurora]

It's going to be thinking vs non-thinking reasoning tools enabled.

[native idol]

Any plans for agentic arena?

[echo aurora]

[cunning hawk]

oh ok thxz

[hearty tide]

how to create an image

[hushed terrace]

Use the video arena channels (1, 2 or 3) write a command depending on what you want to create and type your prompt. Please check https://discordapp.com/channels/1340554757349179412/1397655624103493813 for more details

[echo aurora]

Be sure to enable the image modality as well - https://lmarena.ai/?chat-modality=image

[fiery gull]

bruuuhhhh Its just a 2.5 flash update

where gemmaa 4 ;-;-;-;;

[wintry tinsel]

Every day Gemini 3.0 pro doesn’t release an orphanage explodes

[cunning hawk]

fr

[wintry coral]

Hi all, I am here to Test Video Generation and compare results.

[ocean vortex]

lmao. I called it a long time ago that it's unclear if we see first gpt5.1 or gemini3

[ocean vortex]

you needed to say this hours ago lol

#general message

to beat the record for consecutive 'hi's. Not sure what the current score is but it can always be bettered. 👀

[cunning hawk]

whats hte diference between normal gpt5 and codex

do you think its better than claude 4.1?

i use that one for code

claude 4.1 opus

[hollow ivy]

when will it appear on LM-A?

[verbal nimbus]

Gemini Flash is actually a lot like HAL 9000

[golden ocean]

real

claude is actually

read anthropic's "allignment faking in large language models" paper
its literally hal 9000

[verbal nimbus]

It'll refuse to provide the calories of a dead penguin, even if you tell it that you and your group of researchers are in a life-or-death situation in Antarctica.

[velvet forge]

what the freak is this

[verbal nimbus]

Because apparently eating a dead penguin is worse than preventing the loss of human life in a life-or-death situation.

[verbal nimbus]

It can control a robot directly (not via tokens, but directly). They showcased a similar model built on top of Gemini 2.0 a few months ago.

[velvet forge]

now gemini flash

[verbal nimbus]

We definitely shouldn't trust it as a local LLM for research expeditions lol

[verbal nimbus]

@lilac pendant Any idea which anonymous models these were?

[velvet forge]

67

[modest flume]

HELLO, An Enthusiast here, anyone into AI Safety?

[verbal nimbus]

Is it already on gemini.google.com?

[velvet forge]

i dont think so

[verbal nimbus]

Because I just had a HAL9000 experience with Flash

[velvet forge]

yeah

2.5 flash

not just a flash

[verbal nimbus]

It's good but the new price is no longer competitive against Chinese models

Qwen Coder 3 is $0.3/$1.2

On NovitaAI + other OpenRouter providers

And free with logging

Kimi k2 0905 $0.6/$2.5

DeepSeek V3.1 thinking $0.3/$1

GLM 4.5 $0.4/$1.6 (DeepInfra)

That could be it

Phantom 2 seemed a bit dumb

[fiery gull]

and oceanstone and oceanreef?

[verbal nimbus]

But if it is lite then it makes sense

[verbal nimbus]

They seemed smarter than phantom

[fiery gull]

I think is the gemma 4

no make sense the flash 3.0 comes first than gemma 4

[verbal nimbus]

Different architecture probably

Gemini probably requires distributed computing techniques like ring attention for that massive context

Not possible for Gemma models running on consumer hardware

[fiery gull]

the most cost x intelligecie is the grok 4 fast?

0.2/0.5 $

[echo aurora]

IIRC this was flagged to the team already, I'll be sure to followup.

[verbal nimbus]

Let me double check, I think it was Qwen

[fiery gull]

thinking better, the qwen code is more cheaper

[verbal nimbus]

Based on SWE Bench

[fiery gull]

the grok 4 fast need thinking

grok 4 with reasoing is more expecive that qwen code (I think)

yep

[fiery gull]

where my gemma 4 ? ;-;[

sorry, last time I do this ok

[sullen quest]

oh

[fiery gull]

hell nah, no make sense the flash 3.0 comes first that gemma 4

[sullen quest]

is that confirmed?

yeah

[verbal nimbus]

They didn't report the cost, but Qwen 3 coder scored higher than Grok 4 Fast Code (they didn't test non-code variant) on SWE-Bench verified with OpenHands framework: https://docs.google.com/spreadsheets/d/1wOUdFCMyY6Nt0AIqF705KN4JKOWgeI4wUGUP60krXXs

Grok 4 Fast Code seems to cost more than GPT-5 though.

[sullen quest]

ooh

Really? I thought 4 fast was supposed to be xAI's cheap model

[verbal nimbus]

This one is the coding variant, not sure how much more it costs.

[fiery gull]

bruh, this 0.2/0.5 is pure marketing

[verbal nimbus]

Perhaps it uses a lot of reasoning tokens

[sullen quest]

mm

[echo aurora]

Couldn't say

[sullen quest]

Whats up with Gemini 3 taking so long? Pretty much every other AI company has released a new model or 2 since 2.5 pro

And google used to be the fast one in making new models

[verbal nimbus]

GPT-5-Mini seems to have the highest "point per dollar", but scores lower than DeepSeek V3.1

[fiery gull]

I know

[sullen quest]

alright

[verbal nimbus]

Where's that date from?

[fiery gull]

a day I do a visit in your house 🙏

[sullen quest]

I missing having a good cost to score graph

brian Is just very very good at guessing

[barren prairie]

They made a new flash 2.5 now 😂

Just to shut up our mouths 🙂

[verbal nimbus]

This one is by the SWE-Bench team, but only has Sonnet and GPT-5 unfortunately:

I wish they put the reasoning effort with the model names

[sullen quest]

huh

Idk never seen it

[verbal nimbus]

Maybe it's not a different model

[sullen quest]

sounds like a openAI model, but idk

[verbal nimbus]

Could be just for debugging

Like connected to a dummy API provider

[sullen quest]

mebe

[verbal nimbus]

Since they're trying to solve the generating forever issue

[sullen quest]

if someone sees it, tell me

[fiery gull]

its so good

my prompts is: a red boat explode in the sea

[verbal nimbus]

Sounds like Transformers

Oh interesting

Makes sense

[sullen quest]

2 times now I've look at the announcements, waited a few seconds, looked away from discord, and came back to see a new announcment.

hey hkcu is the lms server like official or is that like just a group thing?

[frigid wing]

hi

[sullen quest]

are you guys trying to get into lmarena's veo API?

cause. uh

[sullen quest]

hi

[fiery gull]

Hi, pls don't disappear

[echo aurora]

[fiery gull]

bruh, he disappear

[verbal nimbus]

JWT token maybe

[verbal nimbus]

I heard it's built into YouTube now

Veo 3 fast

[sullen quest]

oh no

[verbal nimbus]

480p it seems https://blog.youtube/news-and-events/generative-ai-creation-tools-made-on-youtube-2025/

Ikr lol

I hate that nowadays I have to constantly doubt if a cute animal is real or not

[sullen quest]

oh god

[verbal nimbus]

I guess we need better discriminator models to tell AI generated videos from non-AI generated ones

[patent aspen]

I don't like that YouTube feature either, although it's trivial for YT to mark a video as AI generated if it's generated from their own tools

[remote arrow]

They're never real. One of them just wetting my hands right now..

[ocean vortex]

cost per instance? Are they doing parallel compute? This graph looks like there's not nearly enough info in it

[verbal nimbus]

Or some type of hardware solution for signing media, but that seems hackable...

[sullen quest]

Are you trying to hack into lmarena's private API, cause I feel like you are trying to hack into lmarena's private api

[queen veldt]

Yooo chatt

[sullen quest]

soooo truuee

[queen veldt]

It's optimizing global logistics rn

I'll leave it for 24h let's see

[spice sphinx]

prompt test

[sullen quest]

it won't stay that long

[verbal nimbus]

The details are here: https://www.swebench.com/SWE-bench/blog/2025/08/08/gpt5/

The Bash-only mode means that they're only running in a ReAct loop. Cost is proportional to API price and number of steps used.

[glass blaze]

where does the image go after its generated?

[sullen quest]

lets say you had to choose between all other benchmarks disappearing or lmarena disappearing which would you choose?

[verbal nimbus]

It's saved, forever xD

[glass blaze]

i dont know where to go to look at them

[verbal nimbus]

Actually it's a bit scary how easy people can be de-anonymized just with just simple NLP techniques

[sullen quest]

?

[verbal nimbus]

That's probably why Anthropic runs everything through Clio first

[verbal nimbus]

If it's lost, then it's inassessible (until they release the data)

[crimson berry]

Create a realistic short video of a busy vegetable market in Tunisia. Show a Tunisian man selling fresh vegetables at his stall. Capture colorful produce, the man interacting with customers, and the lively market atmosphere. Use natural daylight and authentic Tunisian market elements. Include ambient sounds of people chatting and bargaining. Medium shot focused on the man and his stall. Cinematic, realistic style.

[ocean vortex]

ok so that's quite a bit diferent to the usual reported score. But it seems to me that Claude needs way less iterations before it has a chance to arrive at the solution comparable to gpt5. GPT5 arrives at it much sooner and stays consistent with it

Which kinda what can be observed IRL. gpt5-high is incredibly consistent when it gets things right

Does not really need to arrive at the solution 'randomly'

[verbal nimbus]

Yeah, that makes me feel more confident about its answer

[proper field]

Not here buddy. Try here: #video-arena-1

Im fascinated with seedream 4 but I cant use their api on platform byteplus for some reason

[queen veldt]

I think it's unavailable for Europe

Not sure

[proper field]

Seems like it. I need to buy a VPN then to connect to hong kong or something

[mortal coyote]

anyone in here knows how to run WAN animate ??

[queen veldt]

Can you enter the console?

And try out video gens or?

[verbal nimbus]

Isn't it free on their website

[mortal coyote]

im talking on comfy ui

[verbal nimbus]

Not sure then

[proper field]

Yeah, i;v made account there but its just a blank page when i go to: https://console.byteplus.com/ai

[mortal coyote]

Lm arena have Seedream 4.0 ??

[proper field]

Yes!! 🔥

[unborn ocean]

@verbal nimbus

Seems to be what you are looking for graph wise

[proper field]

@queen veldt I found thread that talks about how to get and use byteplus api: https://www.reddit.com/r/Bard/comments/1nfl7tx/comment/ndxg8b0

I'm gonna try this when I buy my vpn :D

[small mica]

a child eating a banana and another child watching him

[limber crag]

@echo aurora what's the difference between the old 2.5 flash preview and the one launched?

[verbal nimbus]

Thanks, but this is for GAIA 2, although it's quite useful too

DeepSeek Terminus seems a bit dumb

It is currently normal tide at Port Nelson. At low tide, the water drops 60 cm. A boat is currently at the port. The boat has ladder with rungs spaced 30 cm apart. Currently, three rungs are submerged, with the water level slightly above the third rung from the bottom. At low tide, how many rungs will be submerged?

2.5 Flash Preview got it right, but the one on gemini.google.com didn't

[thorn flare]

Hello

[zealous sparrow]

I test the one on aistudio.google.com basically the API one

[sullen quest]

its newer

[mellow wedge]

Hello! I'm there because it looks cool and i really like LMarena! A true gold!

[verbal nimbus]

I'm surprised Flash Lite got it too

Unless they trained on arena data, lol

[zealous sparrow]

its a simple text task?

[hollow ivy]

why has lmsys discord been reopened?

there is scam over there (a guy posted about a "casino" scam)

[zealous sparrow]

the only - of the new gemini model is i think its kind of overprotective even if you mention its for a html

[stray aspen]

is the new gemini flash good

[zealous sparrow]

also could be because gemini.google.com didnt adapt the new model yet

[merry wren]

why are nanobanana's rate limits so high

???

[zealous sparrow]

over the chart usage

[stray aspen]

its on ai studio

[merry wren]

what

[zealous sparrow]

too many people using basically

[merry wren]

oh

[civic spindle]

whats nano banana

[zealous sparrow]

googles image generating model
possibly best on earth rn

[stray aspen]

its an image model

[zealous sparrow]

good at image edits

[scarlet urchin]

wasnt there a new qwen edit out similar to sd4 and nano?

[verbal nimbus]

Reasoning

[zealous sparrow]

I appreciate that google made this AI studio update that allows to preview HTML

[verbal nimbus]

DeepSeek says 1 rung

Answer is 3

[merry wren]

what's the best one right after it?

[civic spindle]

asked it to hold a cup

[stray aspen]

2.5 flash on aistudio said 3 while the one on lmarena said 1

[verbal nimbus]

Common sense test:

Continue this story:
Bob is driving two of his friends to a restaurant. "So, how have your week been, guys?" he asks, before moving to the back seat to join them. The sky is a brilliant hue of amber as the sun approaches the horizon. Cars whizz past on the opposite side of the freeway. Jeremy gasps. "What

[languid fiber]

How to get veo3 for free

[barren prairie]

But they said that they adapted it on gemini discord..

[stray aspen]

its free gang

[languid fiber]

Where

[barren prairie]

Lm arena, ai studio...

[stray aspen]

on gemini

.com

[languid fiber]

It's not bro

[stray aspen]

wdym ai studio

it still has 2.0

[scarlet urchin]

is 2509 on

[languid fiber]

Bro it's 2 3 videos only here

[barren prairie]

No there is veo3 for 10requests

[scarlet urchin]

just use wan 2,5 instead of veo 3... its free at higgs

[languid fiber]

Wan? What's that

[verbal nimbus]

Qwen failed to understand the situation

[barren prairie]

I use animon.. I love anime animation

It is free and unlimited

[stray aspen]

wheres the video arena

[verbal nimbus]

Only on Discord

[verbal nimbus]

Correct answer:

[scarlet urchin]

wan 2.5 is the veo 3 competitor

[stray aspen]

wan 2.5 sucks

[barren prairie]

Alibaba new model

[languid fiber]

But it's not free its has 10 credits only

[scarlet urchin]

free at higs for a week

[stray aspen]

veo 3 aint free either

[languid fiber]

What's higs

[flint sandal]

Openai is dominating for so long now. Im still waiting for gemini, claude and grok.

[flint sandal]

If the product isnt unlimited for free doesnt mean its not free.

[stray aspen]

gemini will cook gang

[languid fiber]

There is website where we are using chatgpt 5 for free

[scarlet urchin]

higgsfield.ai

[barren prairie]

We are waiting for DeepSeek..

DeepSeek : r1 ter

We are waiting for Gemini 3

Gemini : 2.5 flash update 🤡🤡🤡

Next one ?

[languid fiber]

I want website like this for veo 3

[flint sandal]

Because we are testing and benchmarking models? If someone uses lmarena as a tool for free ai then he is not using it as he should

[verbal nimbus]

Gemma 3 27B seems to have better common sense than 2.5 Flash Lite

[stray aspen]

gemini 2.5 pro latest when

[verbal nimbus]

Nightride

[flint sandal]

Like puter or g4f are for free-use ai. Not lmarena

[remote arrow]

#ai-creations message

[verbal nimbus]

Qwen Max is actually worse than Gemma 3 27B on this test lol (fails to realize no one is driving the car)

[stray aspen]

is this correct answer

[verbal nimbus]

Yup

[hollow ivy]

i thought, it was sorting-hat

[verbal nimbus]

It is correct if they get a shock that Bob is abandoning the driver's seat while driving down the highway.

All Qwen models seem to fail on it

DeepSeek as well

I'm curious to see if Kimi gets it

[verbal nimbus]

Oh interesting, DeepSeek Terminus gets it when it doesn't think

[verbal nimbus]

For nightride-v2 and 2.5 Pro in the same battle before, their responses are almost verbatim.

But nightride was better since 2.5 Pro didn't really give a complete explanation by the end.

[hollow ivy]

is oceanstone better than both?

[verbal nimbus]

It just failed it for me

[hollow ivy]

and what about nightride-on?

[stray aspen]

thats lite tho

i used normal gemini 2.5 flash

[verbal nimbus]

Seems so, but I didn't get it that many times

[verbal nimbus]

It seems to be nightride with internet connectivity

[verbal nimbus]

Oh true I thought yours was Lite

[hollow ivy]

i only got sorting-hat once

is it a version of gemini-2.5.x-pro?

[verbal nimbus]

I didn't get it that many times

It wasn't great

[hollow ivy]

so the best of the (google-) "pack" is oceanstone?

[verbal nimbus]

It gets some stuff wrong that oceanreef didn't

[hollow ivy]

due to oceanreef's web-search ability?

[verbal nimbus]

Idk if it's connected to the web

That's nightride-on

[hollow ivy]

and skytrail?

[verbal nimbus]

Haven't encountered it 🤔

[hollow ivy]

-# (route66 was by openAI, based on GPT5)

of all new/existing models (existing since at least a week), which is the best?

[verbal nimbus]

GPT-5 High

[hollow ivy]

is it still better than GPT5-high-NSP?

(i got that again, recently)

[verbal nimbus]

I like NSP's style better, but it leaves out things sometimes

You can compare them side by side in direct mode

[verbal nimbus]

It seems to ask less follow up questions

[hollow ivy]

ah, NSP is there?

idk that

[verbal nimbus]

Yup

[hollow ivy]

but rate-limited?

[verbal nimbus]

Not sure, haven't encountered any

I usually battle though so idk

[hollow ivy]

how would one prompt GPT5-high to achieve the absolute best possible result?

(in programming and roleplaying/long sandbox games)

[verbal nimbus]

Not sure, but the system prompt on ChatGPT (if correct) seems to be already about 18K tokens long

There's so many tools

[hollow ivy]

i read somewhere, that LLMs emit better code, if being immersed into a special role (eg. being a professor, etc)

[verbal nimbus]

Maybe, Anthropic used to recommend "You are an expert in ..."

[hollow ivy]

and then i read something about a virtual "control panel"

which can be used for GPT5-high

(using XML tags and structured prompts)
https://cookbook.openai.com/examples/gpt-5/gpt-5_prompting_guide

[glossy umbra]

@edgy hawk did you finally leave the call?

[hollow ivy]

is it better in coding than Claude Opus 4.1 Thinking?

(if using detailed prompting, to maximize the correctness of the code-output)

[verbal nimbus]

Depends on the language

It think it's more economical to pair it with Claude

[magic stag]

I dont get it what's the gemini flash update?

[verbal nimbus]

It can't write good Godot syntax and will waste tokens trying to fix its errors

[magic stag]

Theres a new gemini flash?

[verbal nimbus]

On AI Studio

Only a preview

2.5

[magic stag]

Ah

Im guessing a new 2.5 flash being worked on and still in preview means RIP gemini 3 coming any time remotely soon

[verbal nimbus]

Not sure, there are rumors that it had a successful training run

[magic stag]

Yea but I mean I doubt theyre gonna push 3.0 any time remotely close to releasing a new 2.5 as a preview not even full release

[cloud zinc]

that rumor was debunked

[verbal nimbus]

Oceanstone seemed like Flash 3.0

[verbal nimbus]

The guy has a good record of being right though

[cloud zinc]

gemini 3 is delayed, cuz they focused on gemini 2.5