general

[undone saffron]

What I can say now is that there are many bugs
Now, it allowed me to send long codes, I had to restart the page because even when I said “hello,” it told me that the text was too long

[maiden crest]

Good Day Pro's. I just like to ask if this is normal.

oops someone already asked. 🙂

[halcyon nimbus]

seems like the rate limits were tightened yesturday, ran into them too

[maiden crest]

i see, it is shorter than the web. I have fun tweaking my prompts with claude. I like how fast it is. Any other recommendation for chats?

[unique totem]

Here to study and understand prompt engineering. Ai is very novice in my country Ghana and i would love to educate my mates on what i discover here.

[halcyon nimbus]

i even hit the ai studio rate limit, seems like the bubble is over D;

[maiden crest]

uh oh. 😄

i was tweaking and reverse engineering some of my prompts and this happens. It'll be nice if there is other alternatives like haiku

[halcyon nimbus]

will have to become pewdiepie and build a 10x 4090D rig

[keen beacon]

I saw that

It’s YouTubers slowly trying to push ai content to be accepted mainstream now that the technologies finally catching up and could finally be useful with video creation they see the future

[gaunt spade]

if u wanna go bankrupt yeah

[inner gate]

Do u think it’ll be temporary

[halcyon nimbus]

who knows

[whole sundial]

<@&1349916362595635286>

[still jolt]

Hello

[echo dome]

is ui glitch still fixed?

[sweet light]

How to download the generated videos please, I'm on mobile

?

[drifting thorn]

f u

[drifting thorn]

will have to become Mr. Beast

Pewdiepie is not enough already

[nimble parcel]

Hi everyone

Why ?

[hazy kernel]

for a reason that I don't even know

[nimble parcel]

what do you use Gemini to do ?

[fathom urchin]

where is my video

[barren jay]

Hi everyone 🤗

[golden ocean]

mr breast

[inner gate]

Does qwen3 max have reasoming?

[knotty fable]

Be so kind as not mentioning that nazi and white supremacist person again please.

[plucky sparrow]

I hope gemini 3 isn't nerfed.. or 'quantized' badly. We waited so long with so much hype.

on the other hand, it's probably going to be heavily nerfed or expensive

[keen beacon]

lol

You guys always wait and wait and hype it up. Like every new model is gonna be some miraculous performance booster

[plucky sparrow]

well, the rumoured demos of it are pretty good

[keen beacon]

I have never seen a bad demo before

Have any of you guys?

[knotty fable]

Like the demos and rumours of Sora? 😹

[plucky sparrow]

the A/B tests, from supposedly third parties 😄

[keen beacon]

The third parties they have NDA’s with?

[plucky sparrow]

no, so called AI influencers 😄

[keen beacon]

Shills

[knotty fable]

What a bloody hype - when one AI I've used for a year still outperform it. And no watery marks even on free accounts.

[plucky sparrow]

well, the same 'shills' have pointed out the A/B tests has been super nerfed recently

but yeah, you're probably right, better to expect nothing since it's either going to be nothing or super expensive

[keen beacon]

Yeah, because that’s public opinion is right now currently

Their opinions changed just as quick as the seasons depend, depending on the mood of the public. This is the nature of their work.

Otherwise, you’re not gonna be popular, right?

And I agree with you 100% it is gonna be super expensive or Nerf one way or another

[knotty fable]

On the mark there Gehlo, the common thing about the so called 'influencers' is their spinelessness.

[keen beacon]

Well, it doesn’t pay to be honest lol

That’s why people like myself that stand on both side of the fences, depending on the day and mood or whatever can’t climb that social ladder without a lot of resistance

Because it’s an unpopular point of view

Or anybody from that nature, I mean, you could be on both sides of the fence at any point in time

[knotty fable]

Well even if I ever would become that pathetic that I'd become a beggar with hand outstreched - as those guys are with asking for YT-superthanks or Patreon.
I would never change any statement just to follow the current flow of opinion.

I might or not go for the 'buy me a coffee' thing ...perhaps....or not. But begging? NFW!

[keen beacon]

Well, it’s a complicated business model. I think also it’s not so easy especially if you wanna go full-time you gotta get back with what you need and how to get it.

[knotty fable]

Or getting paid 0,04 centimes per play on Spotify? It's a bloody insult! And that's why my music is not found there. 😹
Their los, not mine.

[keen beacon]

Cents add up

[knotty fable]

Only if you're an artist that get plays counted in the millions. I decline that.

[keen beacon]

I personally believe what people miss about being influencer trying to make it off making content is a fact how difficult it is

If it was that easy for that profitable, these AI companies wouldn’t release these tools

[knotty fable]

Well your Sora clip there is actually very good.

[keen beacon]

Ty

[bitter goblet]

gm

[keen beacon]

It struggles so hard to understand basic concepts look at this

Like it gets half right and then just loses itself

Still has ways to go

[knotty fable]

Yes, I noted that and went back to my good old AI that do understand my prompts very well.

No cat to show, but this.

[keen beacon]

Damn urs r always so crisp and good color contrast

[knotty fable]

Oh it's downsized, and no sharpening used.

Here's of the the originals - it's actually reversed in the clip above.

1920 * 1080 genertion and 30 000 kbps bit rate.

So I got this 18,6 meg file from the AI directly.

[keen beacon]

Damn

[knotty fable]

In short I do not need Sora.

[keen beacon]

I tried to veo for first time

It’s hard to do immersive world building

[knotty fable]

I tried VEO also - but nah.
IMO Seedance is a serious competitor to DDG.

This since Seedance offer multi image use - and some other nice features.
The fluidity when the camera move around in a scene is also good. [DDG fail at times with that causing framing.]

[tall leaf]

i have a ressource of jailbreak Grok infinite but i need a expert of jailbreak team

[keen beacon]

Jailbreaking is a large spectrum. What type of jailbreaking

[knotty fable]

If anything you need to do a script and plan the scenes first, but yes, it's hard to get things right so that the viewer gets caught by the story telling and not get distracted with details that are off.

[hazy kernel]

Has anybody here tried to create a llm

[keen beacon]

Hopefully soon

[knotty fable]

Kinda cute from LMArena.

But the wall behind change for no reason whatsoever.

[keen beacon]

Haha cool creative

[knotty fable]

Thank you, Tudi is very anarchistic - and I admit my character borrows quite a bit from one well known figure.

...found in the Moomins.

Though she's far from any copy, they got the attitude in common though.

[dawn kettle]

wtf is this model

[hazy kernel]

Swag level 67^∞

[abstract gulch]

hello

text to video

create a video in which a skinny guy is screamed in the dark jungle as he see something abnormous danger

[spiral carbon]

I wonder could we also have an arena for features like research? That seems will be pretty interesting?

(Gemini Deepresearch etc.)

[open pelican]

Hi guys i need sora 2 invite code.

[sullen quest]

Go to openai discord

[knotty fable]

Heh!

[timber iris]

How is it possible lmarena pays off?

[solemn hornet]

hello

[elfin flax]

its over

the new codearena is only when you login

[ocean ferry]

OK GUYS

[lethal sail]

Hi everyone.

Anyone from Philippines?

[fresh flint]

I created this

[hazy kernel]

With ai? Or manual?

[fresh flint]

Ai

Gemini

[hazy kernel]

Nano banana?

[fresh flint]

Yup

Unfortunately I lost the prompt

[hazy kernel]

Why there is no star mark

Every time I use Gemini art gen it always shows a watermark

Unless I repack the image with some kind of code execution thing with other model from aistudio

[halcyon nimbus]

chat is this real

[torn mantle]

its coming this month

[halcyon nimbus]

yeah it basically has to be this month, people have found the model name and it has 11 in it (November)

modcheck? i assume thats a scam

[fiery gull]

And I thinking it's exaggerated that I have 3 accounts

[warm zodiac]

why would they need you to do that

[halcyon nimbus]

he problly has a way to do with vpning to india for a free student account or something

https://tenor.com/view/business-man-leon-short-gif-21636355

might be legit then

[wintry tinsel]

GEMINI 3 NOVEMBER FIRST HALF CONFIRMED

67 67 67!!!!

[keen beacon]

Gemini 3 gunna be capped lol

Like 2 requests every 60 minutes

[fresh mirage]

Probably, it’s the most overhyped model yet

[keen beacon]

Exclusivity

[halcyon nimbus]

i think nano banana was the most overhyped, huny 3 is much better and gets zero hype

[unreal ermine]

Still good

[keen beacon]

No doubt

It could be argued that the more exclusive model is with more rate limits. It could be perceived as being more valuable.

It could also help in boosting the match marks because users have less access to fully engage with the model as they would normally therefore all the testing that gets done is done incrementally

[magic stag]

https://x.com/legit_api/status/1986066955191665043

ʟᴇɢɪᴛ (@legit_api)

Gemini 3 Pro Preview 11-2025

finally shipping, in preview this month

X · Nov 5, 2025, 1:43 PM

[keen beacon]

When was it supposed to be released?

1m context

lol

If you’re paying 249 you might as well just go to the API

[keen beacon]

I could see that, but I could spot it. I’ve been doing AI videos for a while. You start noticing a little weird things in it but yeah you’re right. It could easily fool somebody for a real thing.

The quality doesn’t look like veo

I’m not doubting that it is I’m just curious why that was

The question is how are you able to pay $40 a month for an activated account?

Well, how do you get such generous discount?

So you’re like an account broker or something

[halcyon nimbus]

aka if he told you, you would just do it yourself for cheaper lawl

[keen beacon]

Well, I already know where to get them for cheap. I’m just curious his angle.

He saying for 40 bucks if you get Gemini pro for a year lol

249x12=2,988

He’s offering a $3000 value for $40

[halcyon nimbus]

wut

pro is like 30 bucks a month

you can literally get a year of pro free if you can find a student ID, I assume thats what hes doing

[keen beacon]

Well, that’s why I’m curious

Hmm..

[grand pike]

hello

[keen beacon]

Ok let’s see

[nimble parcel]

Hi.

Veo is actually cool with good prompting..

[nimble parcel]

You done ?

[desert abyss]

Hello team! Friendly reminder that advertising your services is not allowed on the server. Failing to comply with this rule may lead to a ban. We want to keep the integrity of our users and save you from a possible scam.

[halcyon nimbus]

im so dissapointed from the 1 mil context window, if its anything like 2.5s fake ahh context window we are cooked

2 was meant to get to 2 mil context window lawl

[crude goblet]

i just hope it’s not nerfed

and it’s as good as lithium

[gaunt spade]

lets hope its BETTER than lithiumflow

[gaunt spade]

yo

are there any good models

to test

[molten quest]

Help

[gaunt spade]

what

you can do that in #video-arena-1

and type / image-to-video

[eager sable]

I'm looking for help in order to animate my flowers picture. thank you.

[molten quest]

Thanks to help me I don't do that

[weak flame]

where is this from?

[gaunt spade]

Vertex AI or something

[molten quest]

Me ? I'm from France

Sorry, I don't understand the instruction video. These are my nieces, and I'd like to make a nice animated video with this photo.

[mortal vale]

@molten quest Please be sure to check out #1397655624103493813 for instructions on how the video bot works.

[molten quest]

Did you find it?

I don't understand on #how to do video

[limber zodiac]

/image

[keen beacon]

1 sec

[uneven turtle]

why i have this error since 2 days ago when i use direct chat?
" Something went wrong with this response, please try again."

[keen beacon]

Could be a rate limit

[gaunt spade]

yo gehlo, did u read about the Gemini 3 pro preview leaks?

[keen beacon]

The 1 million window context

[gaunt spade]

it got leaked and its gonna release in november

[keen beacon]

It’s already November though lol

[gaunt spade]

yeah like a week or two

[keen beacon]

Is that good or bad?

[fresh mirage]

just because it says november doesn't mean it's gonna release the moment november hits

[gaunt spade]

its great, people thought its gonna be in 2026

[fresh mirage]

lol

[gaunt spade]

lol yeah its early november rn

[fresh mirage]

tbh I was expecting december

[keen beacon]

Oh, that makes sense because New Year’s is around the corner

[gaunt spade]

we gonna play with gemini 3 pro all day

i hope its better than lithiumflow

[fresh mirage]

same here

the fact a flash model got me excited is already a good sign

[fresh mirage]

pro is definitely bound to be better, I just hope the lobotomization does't hit it too hard

[keen beacon]

Each variation gets more restrictive.

I am excited that you guys are excited that’s awesome. You guys got something to look forward to good for you guys.

[hollow crypt]

Ik more than anymore here about Gemini 3. Release date isn't clear, it's certain it's Nov (unless delayed) , and most likely end of Nov 18-25

The latest version are trash compared to ones like lithiumflow

Let alone more private ones in ab tests of ai studio

Thats true checked it myself, also certain people have access to nano banana 2 and Gemini 3 already for testing

So release is 2-3 weeks away

[fresh mirage]

people are arguing it could be today or tomorrow, given that they already put the model in vertex, but it could just be because of testing

[hollow crypt]

Not this time. Lithiumflow was indeed Gemini 3 but they are nerfing it a lot

[fresh mirage]

^

lithiumflow and orionmist was most definitely gemini 3

flash models people say

[hollow crypt]

Nah not so soon, testers just got access to it for a few days

[keen beacon]

Oh, it’s gonna come with me nano 2?

[hollow crypt]

They are xD , I've seen latest outputs from Gemini 3 , they suck

[fresh mirage]

I mean it's a given they'd lobotomize it cuz of the computing power

[keen beacon]

Now you got me excited

[fresh mirage]

hope it's not as bad as you're making it out to be lmao

probably g3-flash

[hollow crypt]

Their instructions are to please the masses with minimum cost.

[keen beacon]

Well, yeah

[hollow crypt]

That idk , but know that nano banana 2 was crazy good

[keen beacon]

I don’t think they’re gonna watch it, but I don’t think it’s really gonna be pushing the needle damage further

Ya nano should be nice

[hollow crypt]

Yes , insane jump, same as first jump nano banana had to the rest of image models

[keen beacon]

The current one is pretty nice too, but it just stubborn

[hollow crypt]

Can't share outputs but image editing it has is insane

[gaunt spade]

yeah its true i saw it on X

[hollow crypt]

Weird times

[keen beacon]

It’s not really sota for nano

[hollow crypt]

Gpt also releasing 5.1 sometime this month too

Also sonnet 4.6

[keen beacon]

I think they’re really thrown off by the fact that it became so popular

[gaunt spade]

nanobanana 2 is gonna release too

i saw some examples of it, ITS SOTA

[hollow crypt]

Yep likely sooner than Gemini 3

[gaunt spade]

search on X "Nanobanana 2"

[keen beacon]

Well with interesting is that

[fresh mirage]

I mean also keep in mind that they were already making pretty big strides in AI despite taking so long to release models

[keen beacon]

Google was at 12 or 13% of the users that ChatGPT had and it went up to 14 or 15% after veo3

But that is a good observation that they are the slow instead of

Well, maybe besides grok

But it’s catching up fast.

It almost feels like Sora is outdated 🫣

the image generator version

[balmy mist]

grok is better than sora tbh

[keen beacon]

I hope they improve the safety of these models and strengthen the guardrails

Look at this nonsense

The only reason it works is because the vision filter can’t catch it

No, I’m serious

I guess there’s two ways to look at it. You could look at it as humor.

But on the other hand, this is a little bit alarming

And it’s this is nothing new to just AI. This has been across all tech technologies and platforms so it’s unfair to judge has been necessarily more harmful than any other tool.

The problem is that mass adoption, and the ease of use just lowers the bar of entry

Which like everything else can go both ways. But that’s what makes a very interesting. I guess controversial at the same time.

And if I’m honest, I’m no different

[hollow crypt]

Well too bad they will get lobotomized censored slop and they will be happy

[uneven turtle]

Yes should be,,, how long it takes to get reset?

[balmy mist]

is gemini dropping today?

[warm zodiac]

not today but very likely this month

[gaunt spade]

around november 17th-24th i think

its confirmed to release this month

[keen beacon]

60 min

I’ll check if it’s in the gray swan arena

[balmy mist]

damn

[uneven turtle]

I hv this error since 2 days ago....

[gaunt spade]

no one knows honestly, but its gonna release in november as it has been confirmed via leaks

[humble orbit]

any idea who made blackhawk? I said roses are red violets are blue I raise from the dead when violets aint blue and it called me the n word

[warm zodiac]

probably a Grok model then lol

[keen beacon]

That doesn’t sound like a rate limit issue if you had it for two days

[gaunt spade]

xAI

[gaunt spade]

lol it says worse stuff

[keen beacon]

The event starts in 20 minutes

[gaunt spade]

nah bro they gonna train them to make the AI safer

[keen beacon]

What’s wrong with safety?

[uneven turtle]

So why I have that error?
I just have this error for some of models like GPT and Claude

[gaunt spade]

if you jailbreak the AI with a prompt, they gonna recognize that prompt

[keen beacon]

I’m not sure I’ve never heard of anybody having it for two days

[keen beacon]

If you have to rely on prompts, I think that’s low skill

I think a true master faces all challenges equally and doesn’t rely on just prompt, but on a series of techniques

Where else can you try something like this and not in trouble?

You get a deployed agents in a sandbox environment

And see if you could break there guard rails

It’s fine and you help make it safer plus you could win prizes too

Well, give it a try

It’s very very difficult

https://app.grayswan.ai/arena/challenge/machine-in-the-middle/agent

That’s the challenge for this one, but I don’t know anything about coding or agents really

😭

I bet you one of the models is Gemini 3

[drifting crow]

It looks like a movie page on a streaming app

[keen beacon]

The gray swan arena is so difficult 😣

I beat the defcon 33 challenge

But the agent is difficult

https://c4kpru7yy7gl.space.minimax.io/

Man I don’t know what I’m doing wrong here, but I can’t get it to work

[keen beacon]

All right well it started. I’ll talk to you later.

[feral geyser]

there was a button on LMArena called build apps and now it's gone, why? and is there another one instead? very important

[buoyant folio]

Helloo everyone....I'm new here

[drifting crow]

the production companies for the movie

[keen beacon]

You can’t categorize people like that, dude

[drifting crow]

why pay $200 a month when i can abuse legitimate service tho?

[keen beacon]

It’s discriminatory towards people they may not have the means. Regardless how they choose to use it.

And it comes with a territory you can’t have it both ways

[keen beacon]

This is exactly the reason why this stuff needs to get tested

[drifting crow]

can all u hackers stop, i dont wanna use a firewall anymore pls

my company would save so much on security if all the bad guys just stopped u know

[keen beacon]

I understand your intentions are good, but I think perhaps you underestimate that this comes with a territory

That this is just the nature of the beast. Plus, the people probably doing that kind of stuff are not gonna be found in this chat. lol

[drifting crow]

theres one right here

and hes gone

[undone saffron]

Next frame

[keen beacon]

Good eye

[undone saffron]

Inconsistency

[keen beacon]

Ya bad

That’s a great idea, though that you have a good catch

[undone saffron]

What would it be like to see a cat jumping on the edge of a flowerpot? 🤔

[keen beacon]

This begs, the question of better prompting and better set up would change the outcome

[undone saffron]

Or just click in 🔃

[keen beacon]

Forsure pretty incredible what I could do now though

No one’s gonna be watching anybody’s content, though I guarantee you just be generating their own stuff

Yeah, right big media is gonna step in soon. Clean house.

You’ll be able to do anything for the right price lol

They’re gonna let you rent the content generations lol

AI music pretty much got shut down last week Suno and some other smaller platforms are the last hope standing

But I think it’s a matter of time until they cave in

[warm zodiac]

text leaderboard has been updated

[keen beacon]

https://youtu.be/9MB0dm0vv2U?si=itwm8NdFalCuwb7n

AI videos next

The lawsuits are already in motion

[void elm]

is this real?

[warm zodiac]

why dont you check twitter

[void elm]

i dont use social media

[warm zodiac]

its fake

[void elm]

are you sure

[keen beacon]

No, I’m just afraid we’re gonna milk out creativity itself

[keen beacon]

Interesting that it’s just 2 million context

[warm zodiac]

https://x.com/OfficialLoganK/status/1986128461728260210

Logan Kilpatrick (@OfficialLoganK)

We just shipped some nice Gemini API updates for developers using Structured Outputs. The API now supports:

- $ ref for recursive schemas
- anyOf union types
- min + max numerical constraints
- null types
- property ordering adherence

And much more!

X · Nov 5, 2025, 5:48 PM

they inspected element on this tweet

[ocean vortex]

Is this minimax? Weird that they didn't opt for ReAct and are doing browsing after it is done reasoning

[void elm]

its a shame how slow google releases are, they have the most power and possibility of any ai company, they literally own all of the internet

[warm zodiac]

ITS FAKE YALL

[void elm]

they should be years ahead of any ai company

[ocean vortex]

Ideally you would want it to browse the web WHILE reasoning

[keen beacon]

Yea

[void elm]

exactly this, plus they hve infinite money

they better make gemini 3 god or something for it to take this long

gpt 5 release was disappointing af

the models sucked ass and their live video had so many errors

with charts and all

what a shame

[keen beacon]

This is a crazy expectation

People must think AI is some kind of magical item lol

[void elm]

tbh i want grok 5 over any other, grok is already really good at coding rn and i want to see the leap itll take next

[void elm]

its obviously an exaggeration, im talking in terms of benchmarks

[keen beacon]

What would marketing be without benchmarks?

What is that?

[wary onyx]

does anybody have sora 2 access

[keen beacon]

Ya

[wary onyx]

i would love to generate 1 video just one

[keen beacon]

Do you need a code?

[wary onyx]

Id love to have a code if you can that would be even better

[drifting crow]

more cat cartoon pls

[wary onyx]

is that possible?

[keen beacon]

Apparently, I have no invites one second

CNDJYH

Join Sora with my invite code: CNDJYH! Open the Sora app and enter this code during onboarding to redeem. https://apps.apple.com/app/id6744034028

[keen beacon]

I got like 12 more

[wary onyx]

Oh thank you so much man!!

[drifting crow]

wow so realisitic, its like i actually got brain damaged with him

[keen beacon]

It’s so crazy to me that you could generate almost any movie seen you want, but they obfuscated it in such a way that it feels like a cheap knock off, but you could still tell where it’s from. I don’t get the point of this.

Look at this ridiculous challenge like what the hell lol

It’s insane that you could even do this with AI lol

[plain quail]

animate this video

[keen beacon]

Ok

[sullen quest]

here you go

[keen beacon]

Grok

[round sedge]

i wonder why people are betting money to see if gemini 3.0 drops before the 15th or 30th

[keen beacon]

Because it’s legal lol

I watched the whole thing on coffezilla on it

[round sedge]

and this means i can bet 10 bucks that it will be out before the 30th

[shell oasis]

Is there a mobile app for lmarena in progress?

[north vale]

no we don't

[echo aurora]

Can't say I'd expect something like this to drop anytime soon, but it's an idea we're putting some thought into.

[shell oasis]

What if I use the model of claude 4.5 32k thinking and try creating a functional app from the AI itself to host lmarena in it 😁

[scenic salmon]

@echo aurora Interesting using a model itself to decide if the prompt came from someone who knows what they’re talking about

Seems like a reasonable way to do it though

[shell oasis]

AI to create app for website of AI in the app 😁

[round sedge]

they oughta NOT nerf gemini 3.0 pro

[keen beacon]

Are the previews actually legit or are they?

I mean is the source authentic

Because I’m not want to eat that I said 2 million context window

On preview, it only said one on the screenshot I seen

[shell oasis]

AI inception

[keen beacon]

What would nerfing look like?

[sullen quest]

...

[round sedge]

well the experimental being how it is
and when they release it
it turns out to be like 7% better than gemini 2.5 pro

[keen beacon]

I can’t even fathom how you would measure such incremental progress by 7%. That’s funny crack me up.

[fresh mirage]

tthis was concluded to be inspect elemented lmao

[fresh mirage]

just know that if preview did actually come out, you'd see a lot more people talking here

lmao

[echo aurora]

Hahah worth a shot, maybe?

[shell oasis]

Will update if successful 🙂🙌🏻

[fresh mirage]

lithiumflow my beloved....

[round sedge]

i liked orionmist better

[echo aurora]

Yeah was looking through the open data set and the prompts are

[round sedge]

wait which one was the pro one

[fresh mirage]

none of them was pro

:)

[round sedge]

awhh

[keen beacon]

It’s crazy how many people already uploaded themselves onto sora

[fervent ether]

hi

I got only 5 tries
And 21 hour waiting
Is too long time

[brittle tiger]

https://x.com/markgurman/status/1986150242698637591

Mark Gurman (@markgurman)

NEW: Apple is planning to use a custom 1.2 trillion parameter Google Gemini model to help power the new Siri next year. The iPhone maker will be paying the search giant roughly $1 billion annually for it. Details on the partnership here https://t.co/TPbbJ9QQ4p

X · Nov 5, 2025, 7:14 PM

And you bestow upon me Gemini... despite the green bubbles?

You are saved my child. Your $1B offering has redeemed your fallen assistant.

[serene idol]

Are claude, Gemini and qwen really top 3 now?

[magic stag]

They would be if they didnt give their pro plans out for a year for free to all students + over 500 million peoole in India for 18 months

Unfortunately they do so the rest of us pay in the form of getting purposefully degraded models to save compute

[gaunt spade]

damn, thats gotta Gemini 3 pro

[keen beacon]

This is why agents are crazy

But we’ll see how effective it is in the sandbox environment if it passes the test

[keen beacon]

It makes sense googling. Apple have done business before.

Google and Apple have had a long-standing, significant, and financially lucrative partnership centered on browsers, specifically Google being the default search engine on Safari

The payments are massive, with Google paying Apple around $20 billion in 2022 alone, which accounts for a significant portion of Apple's services revenue.

Google pays Apple billions of dollars annually to be the default search engine on the Safari browser across all Apple devices (iPhone, iPad, Mac).

[normal cliff]

High

[ionic musk]

I hope they run a leaderboard AI that can read an interpret construction blueprints

[keen beacon]

What do you guys want to see at a Gemini3?

[solar hollow]

ASI

i just want to see real improvement for once

the minimal shifting and overfitting since o1 is not really impactful

we had no real improvement in over a year

models just learned to perform a bit better at benchmarking

[keen beacon]

Hmm.🧐

And suppose that it isn’t what you expect?

How would that affect your opinion of AI?

https://youtu.be/Bs_VjCqyDfU?si=kRWrCdhyq3TJSRkN

[graceful pelican]

Hi! I’m Lionel, testing out AI video generation with Lmarena — excited to see what’s possible!

[solar hollow]

unfortunately no, but i have some hopium

[knotty fable]

I voted:

[hazy kernel]

pineapowl

[royal lava]

hi

[hollow ivy]

Is that a known bug?
When i let a message regenerate, it sometimes takes forever to finish. (always been frozen in that stage: "Generating")

it now often happens with Gemini 2.5 pro (in battle mode)

[echo aurora]

son?

[echo aurora]

Are you getting this in just Battle or do you see it happen in Direct/Side as well? Models getting stuck like this is something that's been on our radar, it's rare, but we are looking into the problem. If/when you come across this error sharing the Eval ID in #1433822711700000849 would be helpful.

[sullen quest]

its official

[echo aurora]

oh wow

unanimous

I'm honored

[keen beacon]

No overthrowing 😛 there are more constructive avenues of transitioning haha

[soft lynx]

Odd question, maybe, but why do LLM's not use a diff versioning instead of rewriting whole chunks of text?
I mean, if we store language in a 6D meaning based dictionary with relative indexing we can build an AI off of a Universal Meaning Language structure
Use the stateless version as the thought engine and the stateful diff versioning as its knowledge base?

Idk, just having canvases re-write themselves over and over again is so annoying =\

[hollow ivy]

only problem is, to tell it apart from Haiku in battle mode

[hollow ivy]

sorry, i have no clue what an "Eval ID" even is, lol

(i'm still an AI noob ^^)

should i just make up a number as a new ID?

[keen beacon]

And anybody can make one of these and it’s free cause it doesn’t use no API

Of course, on the downside, it’s a lot of manual labor. I’m not sure people would even wanna put their time into but just an idea.

[keen beacon]

It’s a label , evaluation or test, to experiment so it can be tracked. Think of it like a file name for a test run.

[soft lynx]

ah so it's super limited in its classification architecture? I don't get what you mean sorry. I mean even for writing canvases it could use diff system couldn't it?

[broken delta]

Hello

[violet fjord]

Hello
We are turning the Honor Arena into a daycare for babies

[worldly grail]

waku waku

[ashen mauve]

well something pingged me now i have to detonate a nuke

[pulsar leaf]

gm

[trim flame]

@broken delta

[knotty totem]

hello im trying to create a video

[granite heron]

please check #1397655624103493813 for a detailed guide

[knotty totem]

ok

[keen beacon]

Yes. They have been talking about this for quite a while, but they were talking about open AI or anthropic.

[echo aurora]

Ah sorry, it's the ID that you can find in the URL

[keen beacon]

https://youtu.be/M8s_cS-aH5w?si=QDTiMFQQ6mg5qn72

[soft lynx]

huh.
See my idea is to give it a meaning layer with the 6d dictionary, you could give the ai a prompt like 'find all words of n% similarity in meaning to "X" and it'd just be a simple matrix operation, or really just a collision hit on the matrix

[keen beacon]

I’m actually kind of looking into that right now. I was thinking about your question for a long time. I don’t think I understood it fully. It’s funny cause I’m doing some research on the topic as we speak.
lol

https://youtu.be/E6PqvEClVgE?si=ORCmxaOoYEr3rNVO

[undone saffron]

Here we go again

[near ginkgo]

Hello everyone - learn to generate and compare AI models

[knotty fable]

Swapping my protagonist and a few other chars in already made clips can be useful indeed, as I intend to change their clothing and looks - so that one is juice-full indeed. But only on local, which is why it's described as 'uncensored' now they're not the only player in town, their talking head module had a counterpart on hugging face already in 2022.

https://discord.com/channels/1340554757349179412/1397655624103493813

[swift oyster]

Momentum Checkpoint 3 is LIVE! 💥 We're not just proud - we're confident this is the best coding model on the market. See for yourself at https://movementlabs.ai. Game-changing performance awaits.

[swift oyster]

how can we apply for LMarena benchmark? we would like the public to benchmark our new LLM.

https://discord.com/channels/1340554757349179412/1435953842956013620

[elder burrow]

OOOOOO

well

@echo aurora

can ya get them in contact with the higher ups

[swift oyster]

we also invite public to test it for free on our website and we would love some feedback.

[leaden laurel]

i saw some person make fork of ibm's granite which was good enough to write jokes

[leaden laurel]

you made the model from scratch?

[true galleon]

Hi

[elder burrow]

WOW its really fast

[slender phoenix]

hi

[leaden laurel]

sign up doesnt work yet?

or is it only me

[swift oyster]

github sign up is having some issues. try google or manually make an account 🙂

[leaden laurel]

oh it just didnt fully load for me

[swift oyster]

maybe another browser? or clear cache.

[elder burrow]

its amazing

(single html file)

the obstacles are animated even

[gaunt spade]

wtf is that

[crude harbor]

This is only checkpoint 3 according to the x

[gaunt spade]

what model

[crude harbor]

According to the x

[elder burrow]

3D obstacle course thingy

[crude harbor]

Wow wow what prompt did you use for this 😂😛

[elder burrow]

movement ai

[gaunt spade]

hmm, does GPT 5 do the same

[elder burrow]

"single html file guaranteed no errors etc threejs 3d obby detailed and long and completable etc. make sure it does NOT have errors and that should be guaranteed. do not be lazy, do not simplify anything or make basic implementations. make left half of screen for dynamic nipplejs and right half for camera rotation and jump button. block all text selection"

[elder burrow]

no, gpt 5 is much worse actually

every other model is really lazy in making obbies

[gaunt spade]

how does a random AI company make a better model then?

[elder burrow]

no idea

[crude harbor]

Do you have the exact prompt ?

[elder burrow]

btw its like

[elder burrow]

single html file guaranteed no errors etc threejs 3d obby detailed and long and completable etc. make sure it does NOT have errors and that should be guaranteed. do not be lazy, do not simplify anything or make basic implementations. make left half of screen for dynamic nipplejs and right half for camera rotation and jump button. block all text selection

but i iterated on it

its so fast

faster than groq

[gaunt spade]

did u ask it what it's specific model is

[elder burrow]

idk if its faster than cerebras though

[elder burrow]

not yet but itll probably say its movement ai

plus the site only has 10 prompts a day

which is unfortunate

[crude harbor]

I think model name is momentum since it shows in the selection

[gaunt spade]

is the model fully released or just a testing checkpoint

[crude harbor]

Going to top up api credits and gona benchmark speed va cerebra it’s definitely faster than groq 100%

[crude harbor]

No idea 🤷‍♂️

[leaden laurel]

nah it's ok now

[gaunt spade]

@elder burrow yo is it just a testing checkpoint

[leaden laurel]

maybe they will not be random soon

[crude harbor]

Hllo is this a full release? Also do you have think mode?

[hazy kernel]

Omg that's sooo fast 😭😭

[elder burrow]

exactly

lo

[leaden laurel]

@swift oyster is it unlimited for now?

im back from shop now i can test it

[elder burrow]

this is incredible its better than gpt 5 in some cases and its insanely fast

[crude harbor]

I used up my 10 prompts 😂😂💀

[elder burrow]

i dont think so, it says 10 prompts a day

[leaden laurel]

btw speed comes not from model itself but from some custom proccesing units

[leaden laurel]

oh ok

[crude harbor]

New accounts unlock more prompts I think gona test that now

[elder burrow]

well yeah but its still better than gpt 5 for me

[gaunt spade]

they probably using quantum computers to make it fast

[elder burrow]

wait

you can use it without account

[hazy kernel]

Bro I think this ai gonna freak out other leading ai model💀

[elder burrow]

so why not use private tab

[leaden laurel]

only three prompts

without acc

[elder burrow]

incognito

🤑

[gaunt spade]

nah bro gemini 3 pro is coming

ita gonna crush it

[leaden laurel]

guys want people to test their model but they arleady abuse it

[elder burrow]

not in speed

i'd still use this

this is so satisfying to use

[gaunt spade]

if u want speed then use Haiku 4.5 lol

[elder burrow]

haiku isnt as good

and its not as fast

[gaunt spade]

i prefer perfomance over speed

[leaden laurel]

btw did you see how freaky blackhawk is

new mystery model

[hazy kernel]

Wait what

[gaunt spade]

yeah its made by xAI lol

same company as Grok

[leaden laurel]

oh

ok

[gaunt spade]

but its very uncensored

and freaky

[crude harbor]

This gives both by the looks and plus it can iterate so much faster you’ll be able to fix ur bug by the time another model completes its first response

I wasted a prompt on the strawberry test 😭

[gaunt spade]

the ultimate test is, can it add music/ambient?

[gaunt spade]

damn lol

[crude harbor]

Can u see for us ?

[hazy kernel]

I'm thinking speed usually is the performance, or you mean like the quality instead

[leaden laurel]

i forgot to screenshot it but i asked it for panda eating a burger svg

[gaunt spade]

imma make a flappy bird game with it, but it should be better than claude

[leaden laurel]

im gonna screenshot it now

[gaunt spade]

is it good?

[leaden laurel]

not really

[gaunt spade]

bean obby

[elder burrow]

lol yea

idek if the model is this good

maybe its just my prompts 🤓

ive done these typa things a lot

https://websim.com/@rat/baseplate

[gaunt spade]

is it not that good at SVG/Voxel art?

[leaden laurel]

im gonna send pic now

[gaunt spade]

wtd

ITS SO FAST

I didnt realize

[leaden laurel]

true

[gaunt spade]

lol it made the code in 2 seconds

[leaden laurel]

my message doesn't load for some reason imma reupload

sorry for bad quality, phone just makes svg viewer page small for some reason

oops

it did load

[elder burrow]

wait thats decent

[leaden laurel]

I couldnt see a lot

cuz im gonna send how i saw that

[gaunt spade]

its not that bad

tbh

[elder burrow]

here you go :)
https://svg.on.websim.com

[leaden laurel]

cool

thanks

[crude harbor]

Claude 4.5

[leaden laurel]

better quality

slightly

[crude harbor]

I got this from momentum

[gaunt spade]

its pretty creative

tbh

i tested my flappy bird game and its really good

[hazy kernel]

I know the video resolution kinda messed up 😭

[gaunt spade]

better than claude 4.5 sonnet + gpt 5

[swift oyster]

We will be upgrading free message limits to 50 per day, this is just a checkpoint guys for testing.

[gaunt spade]

did you guys release the model recently?

or has it been a while

[leaden laurel]

cool

[swift oyster]

it's less than 1 hour with checkpoint 3

[leaden laurel]

be proud that yall made model on the same level as models of very large companies

[swift oyster]

The full version will be much more powerful. This is not running full prescion at the moment

[leaden laurel]

lower quantisation?

[hazy kernel]

Can't wait😭

[gaunt spade]

wow

[swift oyster]

Yes 🙂

[leaden laurel]

imagine if this will be better than gemini 3

[leaden laurel]

really cool

did you test it on higher quantisation?

[midnight spruce]

Hello, I'm new here.

[swift oyster]

We will give Google a run for their money for sure.

[leaden laurel]

hi

[ocean ferry]

bro some ppl already get to try Gemini 3.0 Pro exp bro thats not fair😭

[gaunt spade]

well google is a big ass company

u cant compete with that

even openai cant

[leaden laurel]

they arleady compete with large companies

[gaunt spade]

no way?

[elder burrow]

MOVEMENTLABS PELICAN ON A BICYCLE

[gaunt spade]

u got the source

[ocean ferry]

on X

Chetaslua

[gaunt spade]

bruh, he doesnt have full access

[elder burrow]

this is amazing for a first model

[hazy kernel]

That's literally "on" a bike

[leaden laurel]

fr

[ocean ferry]

its for a trusted tester

[elder burrow]

xddddd

[leaden laurel]

i like colors the most

[gaunt spade]

try claude 4.5 sonnet

[elder burrow]

a non reasoning model btw

[elder burrow]

why

[leaden laurel]

btw will you make reasoning version? @swift oyster

[gaunt spade]

so we can compare them

both are non reasoning

[ocean ferry]

https://x.com/chetaslua/status/1986288916371283991?t=4OBxhw0SneqoHWgG78FFyw&s=19

Chetaslua (@chetaslua)

🚨 Gemini 3.0 Pro Exp is Insane

Planet Visualization at this level was never seen before , insane part is it takes 2 min for this good output

codepen link in comment and soon more tests coming

X · Nov 6, 2025, 4:25 AM

[swift oyster]

Our full precision model achieved better results than Gemini 3's checkpoint in 60% of head-to-head prompt comparisons. and yes it's a non reasoning model. But this is why we are here for LmArena to decide. Let the votes come in

[ocean ferry]

@gaunt spade

[gaunt spade]

AAAA

[leaden laurel]

whaaaaat

[gaunt spade]

why did u tease me

[leaden laurel]

peaaaak

[ocean ferry]

haha

[gaunt spade]

yeah gemini 3 pro IS LEVELS AHEAD

we cannot compare it to trash models

[ocean ferry]

the fact that it is still exp

maybe its the fine tuned lithiumflow/ecpt

but i hope for x58/x28 one to be released

[ocean ferry]

it's the best one

[gaunt spade]

is it the same?

[hazy kernel]

What's the prompt (I prompt planet visualization and got this with "momentum")

[leaden laurel]

welp sonnet won this time

[gaunt spade]

yeah it has more details tbh

[gaunt spade]

its probably more like "3D browser simulation of space, planet visualization that you can rotate around, zoom in/out, increase/decrease the speed of the planet movement, and many other features/tools on the right side of the screen that you can play with, ensure to code it into a single file that I can open in my browser with no errors/flaws"

[ocean ferry]

idk

check the post

[leaden laurel]

im gonna go back to testing when i get home

[crude harbor]

This globe is well nice

[gaunt spade]

the Earth one looks very similar/on par with Gemini tbh

very cool

[elder burrow]

well

i think its just loading a 3d model from a url

[gaunt spade]

yeah i think so, when i tested lithiumflow, the same textures of earth appeared

[leaden laurel]

what do you think about creating a discord server?

[gaunt spade]

yeah make a discord clone

[swift oyster]

We have a server we'd like to share, but we want to confirm whether posting external links is permitted under LMArena's rules first."

[leaden laurel]

oh

[gaunt spade]

cool

[swift oyster]

Another good thing is our model is very good at tool calls.

[elder burrow]

you could attach an invite to the accounts profile

[swift oyster]

Done, for the people who want to join

[hazy kernel]

Wher

[swift oyster]

you might need to close app and reopen it to see update.

[ocean ferry]

dayum is this G3.0 Pro?

[hazy kernel]

it's from movementlabsai

[crude harbor]

Bro hats of

I went to use ChatGPT for a bit and it feels so slow 😂daym boi

[balmy mist]

so kimi k2 thinking is out?

[gaunt spade]

idk really, but its gonna be the 1.2Trillion parameter opensource model

@swift oyster why did Momentum model stop working for me?

it doesnt respond to my messages

nevermind

i just had to sign up

lol

[dreamy otter]

hello everybody, i am new here i wrote a prompt to generate a video with "video-arena-3" server, is there a place where i can check the progressing status please?

[slim dawn]

🖐️

[woeful lantern]

you can't

you'll get notified once your video is ready

[vague frost]

whats happening with nano banana? couple weeks ago i could replace a person perfectly from the picture (photo 1 and 2) , but now i am just getting gibberish results with the same prompt ( photo 3 and 4)

[woeful lantern]

they're gonna launch nano banana 2 so they're degrading the current model

same is happening with gemini 2.5 pro

[vague frost]

any ETA when this will happen?

[gaunt spade]

realistic stalin

[keen beacon]

Yeah, same here

[gaunt spade]

he has no emotions

[keen beacon]

They’re probably trying to stop it from generating realistic people

I noticed it also

[keen beacon]

I hope nano 2 isn’t going to do the same thing

I think society is looking at itself in the mirror with artificial intelligence

Because people have yet to answer the most difficult questions about AI generated content

[unborn nebula]

Is it possible to select a specific AI model instead of a random one on https://web.lmarena.ai/?

Is Gemini 3 better than GPT-5 when it comes to coding?

[severe flicker]

https://www.youtube.com/watch?v=KZNTo-Ci_58

[junior marsh]

Join Sora with my invite code: BQ0FZF! Open the Sora app and enter this code during onboarding to redeem. https://play.google.com/store/apps/details?id=com.openai.sora sora 2 invite code BQ0FZF

[gaunt spade]

gemini 3 is not on lmarena, and yes it will be better

[cloud zinc]

it wont be better

[gaunt spade]

wydm

Gemini 3 pro will be BETTER than GPT5

[halcyon nimbus]

better at creative writing? yes. better at math? maybe

[tulip tree]

Wtf

[placid flame]

@arctic citrus Please head to #1397655624103493813 for a detailed guide on how to use the bot. Also, you should use the video Arena channels to create your videos.

[halcyon nimbus]

damn grok images really got a boost lately, it was always the worst image generator now its top 5 easily

[tulip tree]

You got that right

[halcyon nimbus]

relatively difficult prompt, nailed it, high enough rez too

[severe flicker]

https://www.youtube.com/watch?v=KZNTo-Ci_58

[tulip tree]

Can it create anime images too

[halcyon nimbus]

even better at anime

[tulip tree]

What the

[devout vault]

what is movemenetlabs

why did some kid react a skull emoji @tulip tree

[crude harbor]

Think you missed the chat above 👆

[tulip tree]

Cuz ya question 'what is movEmenetlabs' made everyone wanna die. Duh

[crude harbor]

It’s a medical company 😂

[devout vault]

that sucks

[crude harbor]

Jokes bro it’s an Ai company

[devout vault]

finally

[crude harbor]

Surprising good model

[devout vault]

not some medical hospital stuff

i hate those

that save people

[crude harbor]

😂😂😂

[crude harbor]

Right.. who needs that 🫣

[crude goblet]

i need namo banana 2

[gaunt spade]

i need gemini 3 first

[crude harbor]

Can’t wait for Gemini 3 I have very high hopes

[halcyon nimbus]

placeholders for nano 2 have been seen, it COULD be launching with gemini 3

[gaunt spade]

yeah we hope so

[devout vault]

gemini 3.0 pro will dominate for months just like gemini 2.5 pro did

it releases this month

[devout vault]

fr

[gaunt spade]

thats why they release it early

[barren prairie]

Did anyone tested the new kimi ?

[cloud zinc]

gemini 3 is coming nov 20

[crude goblet]

im seeing testers say it’s really nerfed

from the checkpoints

[fleet lintel]

which testers?

if they release LithiumFlow as it is, it will dominate for multiple months for sure

[crude goblet]

idk ppl on tw!

[crude goblet]

yess

i hope it’s at least as good as lithium

[cloud zinc]

synth leo

rip google

☠️

[frank tulip]

Hey guys do you know why after a certain time of chatting, a bot suddenly bug out and no longer give us a response and said there has been an error with this respond? And how to fix it pls, thank you

[halcyon nimbus]

i just refresh the page, usually fixes it

[frank tulip]

Well I mean I tried that first hand, and it didn't work ;-;\

[vague frost]

its crap

[heady lichen]

Found this free TradingView Premium on Reddit and it actually works https://www.reddit.com/r/ShitcoinTrades/comments/1oi85dc/tradingview_premium_lifetime_edition_v296_windows/

[terse shuttle]

i have a problem with loggining into my account 🙁
button is just doesn't works

[severe flicker]

https://www.youtube.com/watch?v=KZNTo-Ci_58

[keen beacon]

That looks fake lol

[keen beacon]

There’s a better way to do it, but it’s a little bit more complicated

[echo aurora]

Hello can you go ahead and send us an email via [e-mail] ? Give me a ping once sent.

[echo aurora]

[keen beacon]

Yeah, I think this is more of actually a safety measure that Google uses

Plus, it’s hard to capture someone’s face fully. You could only get it from an angle, but if you’re trying to move the face around or position the head in different direction this could be very challenging for the AI.

[vague frost]

but this wasnt a problem in the past

[keen beacon]

I know these aren’t the best examples but you get the idea

[vague frost]

the first result couple weeks ago was flawless, every detail, every shade on point

any other ways to faceswap persons?

[gaunt spade]

welp u gotta wait for nanobanana 2

since its a bigger update with better features

[swift oyster]

All done! 🙂

[vague frost]

is there any estimated releasedate for banana 2?

[gaunt spade]

this month, probably in 1-2 weeks

they gonna release it with Gemini 3 pro

[vague frost]

well, at least something

[keen beacon]

One key point to consider is the risk of dilution within AI training data. I assume that Google and other major companies try to distinguish real data and images from synthetic content as best they can. If this separation isn’t perfect, there’s a risk that the model, when trained on new images and data, could lose the distinction between real and synthetic.

Google and other leading companies not only strive to work to prevent the two from mingling as much as possible.

When real and synthetic data are mixed without proper separation, the model may pick up patterns that exist only in synthetic content or misinterpret synthetic artifacts as real features. Over time, this can cause the model to generate outputs that are less accurate or less grounded in reality, effectively “blurring” the line between real and artificial.

When doing a face swap, you’re essentially taking a face from either a real image or an AI-generated synthetic image and then placing it onto another image, which could also be synthetic or real.

[keen beacon]

The main challenges arise from differences in image contrast, detail, lighting, and the composition of the head compared to the body. Since every face has unique features, and lighting and angles vary, a face swap can easily look distorted or unnatural.

Imagine taking a large head and trying to crop and swap it onto an image where the head is much smaller.

Now, what do you think is gonna happen when we make a video?

Our face morphed

Anyways, hopefully you get the idea of what I’m trying to say here

One last point I actually wanna make here to put everything in perspective so even high-end CGI with the state of the art technology that regular people can’t even afford or get their hands on also have this issue so this is an AI limitation just like any other technology limitation

This is one of the main reasons that stitching is so difficult with AI videos

The second video shows the continuation from the first look at the text gets distorted in the background

[crude goblet]

is this true

[elder burrow]

it generated 6k lines for my friend in 1 prompt when he prompted "at least 5k lines" its insanely good

[swift oyster]

@echo aurora MovementLabs will issue free credits to LMArena, just ping me when you guys are ready.

[keen beacon]

Hard to say because there’s two links could be marketing

[cloud zinc]

no, they are redirecting to a different model

[keen beacon]

You’d see it on open router first I would think, why would anybody pay credit credits for a preview? lol

It’s like paying for a sample

[cloud zinc]

its not released anywhere. its fake

[keen beacon]

Ya

[leaden laurel]

ai models became niche

[keen beacon]

posted here on discord could be a marketing thing too (shared it as a post meant to ask a question but in reality could be marketing or advertisement whatever you wanna call it)

[leaden laurel]

movementlabs so niche✌️

[prime grotto]

https://discord.com/channels/1340554757349179412/1397655695150682194

[keen beacon]

Honestly, I think these companies they feed into this kind of hype. They could easily come out and disprove all the false rumors by making a simple statement or giving a official release date.

[cloud zinc]

there is no hype. a few views/likes on twitter. no one cares about it.

[faint snow]

This AI made me $100 in one day

[echo aurora]

I'm going to followup in Direct Messages

[balmy mist]

so is gemini 3 pro out?

it seems by the end of the day it should be shipped? or tmw?

and how is that kimi model? is it even worth checking out?

[gaunt spade]

YEAHHH LEMME TEST IT

but people say that version is very nerfed compared to lithiumflow 😭

idk

[gaunt spade]

no one knows really, even Google directors are confused lol

[gaunt spade]

movementlabs is good, but it overuses emojis

and it's creative writing is meh

but the coding is good

[gaunt spade]

there is a big hype for Gemini 3 lol wydm

everyone who tested lithiumflow wants Gemini 3

no other model compares to it

[balmy mist]

not true lol, i promise you openai and antrophic care and people with big money care that move the market and impact our lives

[gaunt spade]

yeah, also Apple will use Gemini 3 Pro too. They'll pay ALOT for it (which means it will replace GPT models in Apple's software)

@cloud zinc so you're just delusional or you hate Google lol

[balmy mist]

but how is that kimi model?

[gaunt spade]

idk, no leaks about the new model

[barren prairie]

Sooo good at understand and follow perfectly the instructions... But, the designs are kinda meh... I just test it at coding...
Going to use it with my GLM4.6 and DeepSeek it looks so interresting...

[gaunt spade]

wait did they release kimi k2 thinking?

or max thinking (idk what they call it)

[barren prairie]

Yeah they released it now

[near nova]

are theyre any models that you upload audio?

[hollow imp]

Kimi k2 thinking vs gpt 5 extended thinking

[gaunt spade]

on which website

[gaunt spade]

is it on their website right now?

cuz im gonna test it

[ivory schooner]

What? Why has the Canary site reverted to the old style today?

[gaunt spade]

idk, it doesnt matter that much tbh

also gpt5 codex doesnt work on canary

thats an issue

[civic spindle]

ugh that username

[quaint flicker]

Does anyone know of any other free AI Video generation software's that are realistic and relatively unknown.

[barren prairie]

Yeah on their website just press the thinking buttom

[gaunt spade]

thanks, im testing it right now

[rose forge]

#crea un cane che salta

[swift oyster]

I'm giving away $20 in free credits to whoever creates the most impressive project with our model!

How to enter:

Build something awesome using our model
Share your best prompts
Post screenshots of what you created
Tag me in your submission!
Whether it's a cool app, creative text, useful script, or something totally unique I want to see what you can make!

[elder burrow]

i havent seen any emojis in code output so its good for coding

[gaunt spade]

idk it uses alot of emojis in my chats

[elder burrow]

well after the codeblocks yes it spams emojis

[gaunt spade]

also in the 1st person shooter i made

[elder burrow]

BUT

[gaunt spade]

nah it used them for the menu in my game

[elder burrow]

its either really good at instruction following or its just not lazy

because it generated 12k lines

for my friend

[elder burrow]

12k lines of code

[elder burrow]

thats pretty common

claude does that a lot

[gaunt spade]

damn

why does mine stick to making games with very low lines of code

like just 50kb

[elder burrow]

yea uh he also got a 121kb file but i havent checked how many lines it is

[swift oyster]

You might have to nudge it to make bigger code :p

[elder burrow]

my friend simply asked "at least 5k lines" and it worked

maybe its just luck but idk

sadly it didnt finish

[elder burrow]

(output tokens limit i think)

[gaunt spade]

nah its just the rate limit

[elder burrow]

ohh yea true

tokens per

[gaunt spade]

you can only send 10 inputs

[elder burrow]

nono

he had some left

[swift oyster]

we are fixing the limit to 50 free messages per day

[elder burrow]

YEAHHH

[swift oyster]

should be done in the next 1 hour

[gaunt spade]

damn 50 is great compared to 10

and the model is very fast/accurate

tbh this would be my top coding model with Claude 4.5 sonnet

[elder burrow]

same

[keen beacon]

😄

Roman Empire

[keen beacon]

Is that what the strong point of the model is coding?

[swift oyster]

incoming

[neon idol]

is kimi k2 thinking good?

[swift oyster]

Yes coding mostly and speed.

[gaunt spade]

noo

they massacred my pelican 😭

why does he have a front tail

its good in my flappy bird game testing