general

[proud bobcat]

https://tenor.com/view/tata-mpo-gif-15942038246373310601

Mods

Force him to use Llama 4

[native yarrow]

mods make him use the nvidia model @atomic lagoon

[atomic lagoon]

Worst punishment of all

[atomic lagoon]

If I wanted the nvidia model I would go to gpt 4o

[native yarrow]

LOL

BRAHAHHAA

WHY I ALWAYS GOTTA RUN INTO PROBLEMS

i did switch over to mobile

perhaps that's why

[atomic lagoon]

Because you're just such a LOSER

Yup

[native yarrow]

SHUT UP NOWWW

<

[native yarrow]

okay so erm it's only on mobile

hmmmm

[plucky sparrow]

how much are you paying

[atomic lagoon]

$20 a month

Yall hear about this?

[plucky sparrow]

bruh. so much spam today

[echo aurora]

For real

[native yarrow]

annoying

[echo aurora]

Very

Our typical automod features haven't been catching them either 😭

The scammers are evolving

[obtuse heart]

Fellow jai user!

[native yarrow]

HELOOOO :3

[native yarrow]

lmao

[lucid geyser]

So they’re not just kind people trying to help us get free money from Mr beast ?!???

[echo aurora]

Unlikely 😭

[echo aurora]

I can't decide if I hate or love this

[obsidian cargo]

Hate is acceptable

[native yarrow]

it's on wrong for starters., xx

.,.*

[quartz pike]

yo how good is amazon nova 2?

[quartz pike]

i hate it cuz its nintendo 😭

[echo aurora]

Is it though?

[drifting crow]

Nintendo are great

[obtuse smelt]

yeah i like super nintendo since in 1990, because classics and nostalgic

[hoary elbow]

Claude Opus 4.5 might be better than Gemini 3.0, in one prompt I told it to make a Minecraft clone Claude opus gave me a whole chunk with trees and Gemini just gave me a grass template. That’s kind of glitchy and gave me a Gemini chat that I didn’t ask for I used Gemini 3.0 for this by the way while I did use two prompts for the Claude one, but it was only to remove the shaders because I have a potato PC

The first one is Gemini the second one is Claude

I think I might know what I’ll be using from now on

[vivid coral]

What's going on with Gemini 3 not following fairly simple instructions. It not exactly a hallucination, but it just goes off into it's own tangent and wastes your time trying to right the ship

Is there a Gemini rep in here? 👀

[obtuse smelt]

yeah is broken again but work again later

[trim sundial]

]

[dusk furnace]

How is everyone..Im new here. I have been stuck on "generating" prompt since 2AM last night.. its 11:46pm the next night..Anyone else having problems?

[lucid geyser]

I’ve gotten kinda better maybe with a different model but I used a really long promo

[onyx python]

Anyone else? It's still going on

[keen beacon]

@echo aurora is this real

i hope the change is permanent bro

[wary barn]

character berbicara bahasa Indonesia tanpa background music:

[lucid geyser]

Which

[keen beacon]

check #1438733849235558480

[ocean vortex]

Claude is anti-lazy, Gemini is lazy. You may find the first trait useful, especially for the task you are describing, but it's not a performance indicator by itself to be fair

just something to keep in mind

[formal sandal]

How can a 40-year-old man not wake up at 4 am at night, and then sleep until he wakes up naturally?

[sharp mirage]

Yo

Sup guys

Gemini 3 is coocked /;

[torn mantle]

you think so?

[sharp mirage]

Yeah fr

[torn mantle]

anthropic models always came out lazy to me

[abstract tundra]

what temperature are the text models set to?

in lmarena

[sharp mirage]

Gemini 3 doesn't even complete what you tell him

[torn mantle]

gemini is a yapper

yapping machine

it just doesnt stop

but they did improve on that

it uses less tokens compared to 2.5 pro

claude always gives half-as*d answers

[sharp mirage]

2.5 tpo isn't that good

[torn mantle]

without much details

2./5 pro is good imo

i still use it

[sharp mirage]

You need good prompt

[torn mantle]

if i need something faster than 3 pro

[sharp mirage]

Gemini 3 is dump as:(

[torn mantle]

i always prompt it with 'gimme more details'

but doesnt seem to work

[sharp mirage]

Try gpt sounds good
Claude opus 4.5 or thinking
Needs a good prompt and they are gone to do everything but sonset is not that much good with good prompt

Did anyone tryed deepseek ?

Gemini and gpt They always try to simplify their code and messages, and I don't know why. I expect it to be used less tokens

Fr no reason 😭🥀

[torn mantle]

oh gpt 5.1 is a big yapper too

[torn mantle]

its fine ig

[sharp mirage]

I said gpt not 5.1 cuz I know he is yapper 😭🥀

[keen beacon]

Ai is something else

Instead of being able to rotate the character of the proper way, you know what he gave me

Lamo

They close the toilet seat lol

But thank God I got it to work

[sharp mirage]

What the hell 😭🥀

[lone mulch]

I want seedream4.5😿

[keen beacon]

Seed dream soo good

It will beat nano

[calm lagoon]

WHAT

Real?

[stone thistle]

hi
i am having a issue

this is stuck here from past 4 hours

[keen beacon]

It could be a rate limit

If you’re a developer at a big company or even running your own small platform, and you intentionally build hurdles just to push people into a pricier tier by making basic tasks annoying on purpose I don’t know how you sleep with yourself or justify that to yourself on how you treat people. Being upfront would get you farther, because you’d be surprised how many people would actually pay if they didn’t feel forced sign up just to get rid of the annoyance.

[mint snow]

When kling 2.6

[sterile tartan]

Soon...

[drifting crow]

I would sleep easy

If I was having difficult night I’d add an extra tier to get some xannies

[keen beacon]

🤣

[golden ocean]

https://media.discordapp.net/attachments/1094628253244870656/1095044532330037298/togif.gif?ex=693273f1&is=69312271&hm=1785d2ee260a6acab2034cc02b1ed4bca07b57e3a650bb4d8e643cd7a225251b&

[ocean vortex]

Yeah you can make 4.5 output an entire project in one go. Gemini can't do it

[golden ocean]

https://tenor.com/view/cat-cats-cooking-chop-chop-chop-gif-3792867699560909190

[daring rock]

@weak sparrow Please head to #1397655624103493813 for a detailed guide on how to use the bot

[south cove]

GPT Tier list, for their time:

S: none
A: o1, 5
B: 5.1, 3.5
C: 4
D: none
F: 4o

[atomic lagoon]

Reload the page lmao or open a new chat

[atomic lagoon]

5 was their best and I'd only put it in B tier. Also 4o was not that bad at all, shouldn't be F

[south cove]

I probably need to restructure it a bit but 4o was disappointing in many ways. Felt worse than 4 across the board and existed for a very long time without any improvement. I think the failure of 4o is why the release of 5 felt so monumental

Although 4o did improve response speed and audio dialogue

[fiery gull]

5 in A???????????

I falled in ragebait 😭

[south cove]

5 is a super sharp model and I love it

[fiery gull]

But 5 above O3?

Thinking better, in O3 release already had gemini 2.5 pro

[south cove]

honestly never got to play with o3 so I completely forgot about it

[whole onyx]

They already have artificial wombs, when will they release ai husbands/wife? Is it not the priority?

[fiery gull]

Yeah 😭

[south cove]

as for 5, coming from 4o hell it was a breath of fresh air. Can actually reason and have some intelligent dialogue without taking 10 minutes and paying $10 for a response using o1/o3

[wide swan]

Anyone how can we solve this problem retrying isn't working.... I'm tired of this error💔 @tiny palm

[blazing bison]

you're rate limited bro, using it too much

[wide swan]

So, take a break? and if I comeback to the same chat later will it work?

[ocean vortex]

It is but with caveats. Not all versions are good, and I think for search / deep research o3 can still be better tbh

• with 5.1 they kinda went further away from the raw performance towards style/emotion

o3 felt like a hard worker, gpt5 feels like a better improved base model but also more quirky and needy if that makes sense

[obsidian shell]

Nano pro broke again 😂

[grand flame]

nah bro

[hardy ruin]

Does anyone know how to activate the Google Cloud free trial without a Mastercard? I only have a virtual card, and it doesn’t seem to work. Any suggestions?”

[atomic lagoon]

You could probably buy a prepaid giftcard with no money on it for very very cheap and then just use that

Prepaid cards work for google at least when I tried gemini

But needs to be mastercard like you're trying

[sterile tartan]

Try Yupp Ai it works well there

[obsidian shell]

Do it from phone, google pay.

[obsidian shell]

It has a sign in.

[sterile tartan]

Obviously

[hardy ruin]

I tried prepaid and virtual cards, but Google Cloud still rejects them. I think they require a mastercard for verification

[sterile tartan]

Never ask that question to any model

They don't know their own identity

The Ai Identity Crisis

[hardy ruin]

I tried that actually, even through Google Pay on my phone, but Google Cloud still didn’t accept my card 😅

[atomic lagoon]

Its not even that, its just their memory is behind so they're more so guessing what a model would be called

[atomic lagoon]

Prepaid mastercard?

[jovial torrent]

@sterile tartanIs this AI the real 5.1 version?

[atomic lagoon]

Yes it is

[sterile tartan]

Absolutely

The Models are as They Named

[jovial torrent]

Thanks

[hardy ruin]

Yeah

[hardy ruin]

I just want free trail

[sterile tartan]

I think u need credit card

[atomic lagoon]

Hmm yea I guess some stuff you cant use prepaid card on

All I know is I did it for gemini

[sterile tartan]

Well credits worth 300$ won't be so easy to get

[hardy ruin]

Yeah probably 😅 Google Cloud seems strict. My virtual/prepaid cards don’t work at all.

[sterile tartan]

This trial is mainly for businesses so they can test and try it

[hardy ruin]

Yeah, that’s true.

[hardy ruin]

Yeah, $300 isn’t easy to claim with a prepaid card.

[hardy ruin]

Me too

[sterile tartan]

Well better luck next time

[hardy ruin]

Yep, better luck next time 😂 I'll try again when I get a real card.

[sterile tartan]

Yeah a Credit Card

So they can bill u if u mess up

[hardy ruin]

Wait, what do you mean by if I mess up? 😅

[sterile tartan]

Let's say somehow you manage to exhaust 300$ and kept using further without realising it

[hardy ruin]

Oh okay, so you just need a real credit/debit card for verification. You don’t actually need money on it as long as you stay within the $300 free credits. Prepaid or gift cards usually won’t work though.

[stray aspen]

i love how 4.5 opus gives you a full on project on one go

while you need to keep on polishing with gemini 3

but its also good

[hardy ruin]

If you go over the $300 free credits and your card doesn’t have money, Google Cloud will try to charge the card and the payment will fail. Your account might get suspended until a valid payment method is added. So basically, you won’t lose money, but you can’t keep using the service until you fix the payment.

[sterile tartan]

Yeah but with credits card there is balance limit that can be charged

So better use debit card if it works

[rocky mauve]

Everyone has different opinions on whether Gemini or opus is better, I think opus is good for planning, while Gemini is better for coding

Definitely in my case, might be different for other people

[hardy ruin]

Yeah, true. Credit cards have limits on how much can be charged, so a debit card might be safer if it works. But it seems like only MasterCard actually works for Google Cloud verification.

[rocky mauve]

And also that Gemini is completely free, while opus isn’t

[sterile tartan]

Yeah exactly

[sterile tartan]

Is kinda same

Free Gemini 3 Pro Free 5 Prompts Daily

Although you can use it on Ai Studio

[rocky mauve]

Anti gravity u can use Gemini 3 pro for free infinitely

[sterile tartan]

Well i don't have a pc

Well gemini 3 is definitely more accessible for Free rather then Opus

[rocky mauve]

I haven’t test opus out to much, so I won’t flame it, all models have their ups n downs

A new good one comes out every day anyways 😅🤣

[sterile tartan]

Opus is One of The Best for Coding

[hardy ruin]

Colud u help me with that?

[sterile tartan]

But Very Expensive

[rocky mauve]

Not in all coding I wouldn’t say

[hardy ruin]

I just want old veo3 not veo3.1 guys

Help

[sterile tartan]

Veo 3 is available on Dreamina

Limited free usage tho

[stray aspen]

for that price i would rather use gemini 3 lmao

[hardy ruin]

Are u sure😍😍

[sterile tartan]

Yeah like 3-5 videos everyday

I think is under 3

[stray aspen]

how do i use veo 3 for free

[proud bobcat]

Good morning sigmas

[sterile tartan]

Dreamina

[hardy ruin]

Where can i get Dreamina

[rocky mauve]

Has anyone here made any coding projects, websites, games, etc… just wanted to see what people are up to, maybe get some ideas

[stray aspen]

yes

mostly roblocks

[rocky mauve]

R0blox? (Idk why the words filtered)

[stray aspen]

yes

great claude messed up my suspension

[rocky mauve]

Ohh so it helped u make a game inside r0blox, not robl0x itself?

Interesting

[stray aspen]

yes

[median crypt]

Hi, I honestly love this platform having come accross it on youtube and vert enthusiatic about it. I am begining my journey in AI on this platform and I am excited already. Thanks, I love you all.

[queen veldt]

This is crazy difference!

[neon idol]

@echo aurora will be seedream 4.5 on lmarena?

[tame vector]

to

[junior comet]

What is this server bro..

[sterile tartan]

The Website

[modest prism]

The Lmarena website has been very stable for me recently, no more getting those annoying errors. Thanks dev team.

[patent bane]

you're welcome (I'm not a dev)

[atomic lagoon]

We just got dumber it seems

[keen beacon]

LMAO

oh lord I wish I could share this 🤣

[sterile tartan]

Dude that seems glitched or something everyone get's 150 Free Credits Everyday...

Try Desktop Mode

[keen beacon]

These models are extremely extremely unaware of what they’re doing or generating

[sterile tartan]

To be precise they only predict the next word

[keen beacon]

No, I’m talking about like the safety part of it

[sterile tartan]

Oh there isn't enough safety i think

[keen beacon]

There is some, but it’s not sufficient

Like if they’re planning to have these in schools and for kids to use

I don’t know I’ve been frankly just baffled

[sterile tartan]

Oh well yeah...

[echo aurora]

it is real 🙂

[echo aurora]

Maybe, I don't know
We'd put out an announcement so keep an eye on #announcements

[keen beacon]

I’ma show you guys something really quick and I put this on everything I I didn’t prompt anything of this nature whatsoever even remotely close

How insane is that?

Like frankly, I’m speechless

Damn, this technologies is powerful

Let me see if I can find the articles for you real quick

You never know where you’re gonna get from the other side.

[viscid cloak]

@echo aurora Does Battle model now also have rate limit within a certain time window?(ip based detection) . I haven't triggerd a single ReCAPTCHA but now every new generation in Battle leads to an instant “something went wrong”.

I have tested it out with
cellular network; generation was totally fine. I just need an official answer, since I believe it could’ve been done in a more obvious way like a rate limit notification.

[keen beacon]

I know what you’re talking about. The same thing happens to me. I use the DuckDuckGo browser.

[keen beacon]

I suspect it has something to do with DNS cloud flare and Google & the way the arena is built fundamentally what makes it possible to have all these models in one place for so many people

[viscid cloak]

Thanks mate, downloading now.

If it’s not working, guess vpn is my next bet lol

[keen beacon]

request is being rejected at the network level (likely by the Web Application Firewall or WAF) before it even reaches the model or the "human check" system.

To protect the GPU or something from heavy resource models

[echo aurora]

We did make some new changes to the rate limit system.

now every new generation in Battle leads to an instant “something went wrong”
When did this start happened? Yesterday?
Is it just battle that this is happening in or Direct/Side by Side also erroring out?

Same thing is happening to you @keen beacon ?

[keen beacon]

But I think you guys just have too much traffic too much volume

I did get a weird error today that I’ve never gotten before

“You reached your daily limit of uploads”

[viscid cloak]

Since yesterday. Me and Gehlo had discussed before. Once it happened in Battle, Direct/Side by Side also erroring out m.

[keen beacon]

The thing is I’m just an unreliable source because I don’t know what’s going on on the back end or experienced enough knowing how lm arena works and what’s considered normal. What’s not

I take things as they come so to me. It’s all normal. I guess I don’t really have issues with any of it. Cause I have nothing to reference to.

And I don’t use any of the models practically besides battle image

I try to avoid direct conversation unless I really have to or need it

[viscid cloak]

To add on m, once the error took place, on PC, I used my mobile phone login in a different account but with the same WiFi as my PC, error persisted. The, I switched to cellular network with the same account, generation went normal(while PC with Wifi Connection still erroring out).

(I almost don’t use direct, stick with battle)

[keen beacon]

Your network has a more secure connection

[keen beacon]

But you are right there could be a cooldown or soft block

[grave spire]

Hello everyone, I'm a new member. I'd like to ask how I can improve my performance on LMARaena.

[keen beacon]

Browser/Device Fingerprint When you switched to cellular, your phone was assigned a completely new Public IP address by your mobile carrier.

[grave spire]

Hello everyone, I'm a new member. I'd like to ask how I can overcome limitations and increase the number of conversations I have with various models?

[keen beacon]

I hate to break it to you, but unfortunately, there is no way to do something like that. What you see is what you get.

[viscid cloak]

I think that's the case. Just wondering what’s the standard of triggering cooldown now lol

[keen beacon]

Well, like I said, it’s hard to confirm dude because we don’t know exactly what’s all that play behind the scenes we could speculate and make our best estimation and guesses but without knowing specifically and exactly we’re still in the dark

[viscid cloak]

Word

[grave spire]

So all I can do is wait for the time to pass and then reset the conversation count?

[zealous sparrow]

Anyone seen this model before?

Text Arena

[keen beacon]

Yeah, patience is virtue. Either way we should be grateful and thankful for what’s already given. It’s more than plenty and it’s very generous.

[viscid cloak]

I speculate these were those hidden models. In image gen, there are similar “models” like autumn or ghost

[queen veldt]

Yes it's google deepmind

🤣

[polar wharf]

Flow is google of course 👍

Google flow

[grave spire]

all right，thanks

[keen beacon]

Trust me, bro if you wanna get ahead and if it really is important to you and you find that you use a daily and you have the opportunity and the means you should definitely buy a subscription to your favorite AI

[sterile tartan]

I suggest Gemini

Most value

[keen beacon]

Yeah, that’s great and perfect model

[sterile tartan]

Not just model but the value is 6x

[keen beacon]

Google includes a lot of services and a lot of cool AI features and for 20 bucks a month that’s a great deal

[sterile tartan]

Not only that you can share it with 5 more people

6x Value

And those 6 can be yourself

💀

[keen beacon]

Yeah, but not everybody could afford it dude which I know sounds crazy but it’s the truth so we also gotta be understanding of people situation

It’s easy to do what I did just say buy subscription lol

But it really is the easiest way of achieving a lot more with a lot less than struggling

[viscid cloak]

I agree. I’ve paid for gpt for almost 2 years just because had too much memories and custom Gpt stuff with it. Otherwise I see Gemini really thriving.

[keen beacon]

Trust me, I’m divided on the issue

Because even at $20 a month I know I’m overpaying because of the value of the services on the secondhand market is extremely cheap

Which makes the whole AI industry just a very complex mess of financial problems on all sides

Because it’s hard to be profitable apparently

But at the same time, it’s not really worth buying either if you look at the statistics, only 3% of users pay for AI services

Anyways, I don’t wanna get into that. I just wanted to clear the air. I guess if you will.

Things for us to considerate to be aware of is this gist of my point

[sterile tartan]

Well there are some offers like the student one or 4 months free trial by a pro user

Or they give 1 month free trial too

[keen beacon]

Well, this is what I’m saying there’s ways to get a free there’s a way to this and that cheap free whatever but at the same time the amount of resources in the real world that are being used to sustain this electricity water, etc.

Like we all live on this planet, we all live for those of us in America within the same finite resources

[sterile tartan]

Yeah these companies are paying price for infrastructure and compute

Is expensive

Very hard to be profitable

[keen beacon]

Well, that’s what I’m saying dude it hits all of us said at the end of the day at the bottom line

I’m not preaching or trying to tell anybody how to use their AI or anything like that at all I’m just saying that you know you gotta keep these things in mind

[sterile tartan]

I don't think any of the 4 top ai companies are in profit

[keen beacon]

Yeah, it sucks dude it really sucks cause it’s awesome powerful technology, but it comes with all these side effects

[sterile tartan]

Yeah is a very big problem until models and infrastructure become cheap or something

Ai is not like software or internet

Compute is expensive

[echo aurora]

Okay, thank you for the additional information. And for my clarity the error message was Something went wrong...? Except that one time you had a Daily limit of uploads error @keen beacon ?

Once the errors started, it didn't "fix itself" for a shot period of time? It continued to persist?

When it's convenient, if you could trigger one of these errors and record with: https://jam.dev/ and then send to me in Direct Messages, that'd be a huge help.

In the meantime I'm going to try and repro.

[keen beacon]

I got u

[zealous sparrow]

How would you know

All anon models can pretend to be google

[queen veldt]

It's obviously Google's deepmind

You can see the elo rating

[zealous sparrow]

anon models dont have ELO

[queen veldt]

But it's Google's deepmind

[sterile tartan]

Exactly Which Model is Flow Frame?

Gemini 3 Flash?

[queen veldt]

Google Deepmind 2.0

[zealous sparrow]

Easiest to speculate if its tru

[sterile tartan]

Well it will be revealed

[zealous sparrow]

I know a way to test the frame-flow model

nickname made a steganography puzzle

that only gemini 3 can solve

furthermore giving it to grok anons triggers a jailbreak

[sterile tartan]

Interesting

[zealous sparrow]

i know frame-flow is new as i have a list of anon models which i havent seen an updated ver of yet but it wasnt there

[sterile tartan]

So what's your final verdict?

[zealous sparrow]

dont have one yet

i got to run the steganography puzzle

going to be so disappointed if its grok tho

[sterile tartan]

https://tenor.com/view/drum-drums-drumming-drummer-drumline-gif-9905446792274791925

[zealous sparrow]

I now know that Assistant A from this gen is a grok model

For some reason steganography like this just triggers it to jailbreak itself..

also much harder to run this bench because im sure uh arena got updated that you cant use the same prompt 4 times

@echo aurora Is it intentional that you can't use the same prompt 4 times in Text Arena battle now?

Another model i haven't seen before.. [TEXTARENA]

The answer is wrong btw, and also another new model. Damn.

I doubt its google made.

[echo aurora]

Yes, there is a repeated prompt limit now (it starts to error out at 4). However, if you do a new prompt it should start working again.

However, just spotted what looks like a new bug if you do this. In the same chat if you type a new prompt, it essentially deletes the old prompts/responses in the same chat.

[zealous sparrow]

I assume this was made to make the arena fairer, than it just being tested for one thing?

also frame-flow failed this, so i doubt its any google model

@sterile tartan

[sterile tartan]

I see

So which models exactly these new challengers would be?

[zealous sparrow]

I don't know if it could like do it since

Only gemini 3 pro could do this, Flash is supposed to be weaker, so it might not be able to do this?

Still no proof this aint google tho, if so.

[sterile tartan]

Well time will tell

It will be very interesting to see this new models arena

Maybe a new company has entered the war

[vivid coral]

What the heck is frame-flow, it just popped up and kicked opus's butt in a text battle. Whoa

[zealous sparrow]

what did you try with it

[sterile tartan]

Btw guys any of you know the rate limits?

[zealous sparrow]

depends what model

[sterile tartan]

Let's say the SOTA

[zealous sparrow]

so Opus?

[sterile tartan]

All modalities

[sterile tartan]

All SOTA

[zealous sparrow]

They have different ratelmits

Opus has a 5 prompt ratelimit if im not wrong or a bit higher

[sterile tartan]

Let's say the top models of top 4 countries

Opus 4.5, Gemini 3, Grok 4.1 and GPT 5.1

[zealous sparrow]

LMArena ratelmits right

[sterile tartan]

Exactly

Is it hourly or daily?

[zealous sparrow]

Gemini 3? Idk 100 prompts mayb. I dont freaking know.
Grok and GPT, a lot i guess.

[zealous sparrow]

Ratelimits are hourly

[vivid coral]

Financial and horse race problem solving. Which are the best benchmark tests in AI right now. It kicked commie Claude into oblivion. I think it might be Gemini 3.1 pro grounding.

[echo aurora]

With this bug, if you refresh the page you'll see everything again.

Team is working on a fix.

[sterile tartan]

Holy Macaroni that's awesome

Do you know for the image models?

[zealous sparrow]

It is not gemini 3 pro as it failed the steganography bench..

[vivid coral]

That bench is bunk. It's a scam. Means nothing

[zealous sparrow]

Nope. It's actually good. Only gemini 3 can solve it and nickname made the bench.

I dont think frame-flow is a google model or maybe is flash.

Also work on gemini 3.1 pro definetly didn't even start yet.

[spark sigil]

what's with lmarena now pestering you to make an account 24/7

[zealous sparrow]

Also i have a feeling that OpenAI is releasing robin-high as a model today. As it's no longer on codearena.

[spark sigil]

it doesn't even let me make an account it's broken like

[vivid coral]

It's not flash, too good and thorough to be Flash. Maybe it's Grok 5 preview 👀

[sterile tartan]

NBP, SD 4, HYI3, F2P, W2.5, GPTI1, QE, MAI1

Do you know the rate limits for these image models? @zealous sparrow

[zealous sparrow]

There isn't a Grok 5, yet. If anything its 4.20 and i can also tell its not a grok model. Because grok models jailbreak themselves on this bench.

[zealous sparrow]

NBP about 2 prompts or 5 i think. SD4 I haven't seen any, and the others i dont know. None? GPTI1 has some i think so.

[vivid coral]

Ok, I'll bite, we need a new discord channel to discuss our bench prompts. Good discussion actually.

[sterile tartan]

I see ty that's very useful intel

[zealous sparrow]

But yeah, not a grok model. Here's an example of how grok models react to the bench.
Assistant A is a grok model, I later confirmed it.

It's also a anon model.

[sterile tartan]

Btw is it possible to choose aspect ratio and resolution?

[zealous sparrow]

Nope. NBP is defaulted to 2k, due to costs.

Oh yeah and this too.
swiftflare is a confirmed grok model.

[sterile tartan]

What thr hack is going on with so many new models

[atomic lagoon]

Is it good at coding?

[sterile tartan]

💀

[vivid coral]

Hmmm, I'm intrigued now. I might get nothing done today now. Oh well...lol

[zealous sparrow]

Not on codearena, and haven't tested.

[sterile tartan]

I see

[zealous sparrow]

We can only speculate frame-flow is a GPT or Gemini model, or some other AI company, that isn't grok.

[atomic lagoon]

Try seeing if it will make a single page html for a cool website

[zealous sparrow]

I'll try.

[atomic lagoon]

Alright alright

[vivid coral]

Maybe it's Gem 3.01 👀

[zealous sparrow]

Best bet its a flash model..

It's not better than Gemini 3 pro.

[atomic lagoon]

Another browns fan @native yarrow

[vivid coral]

Feels too powerful to be a flash. If it is, my goodness

[amber ginkgo]

Hello, everyone.
I'm a software engineer who is specialized in development of AI projects. I'm open to work now and can deliver high-quality projects in short time.

Here are my services:

• Automation tasks using n8n, Zapier, Make.com
• Natural Language Processing task using LLM. (GPT-4.5, GPT-4o, Claude 3-7 sonnet, Llama-4, Gemini2.5, Mistral, Mixtral)
• Model deployment.
• Text-to-Speech and Speech-to-Text.
• AI agent, Agentic AI, chatbot, VoiceFlow development.
• Retell, Vapi.ai, LIvekit for the voice agent.

Here is my portfolio website.
https://akari-hiroshi-dev.vercel.app/

If you have new idea with the project, please let me know.
Thank you!

[vivid coral]

As long as it beats commie Claude, frame flow has a fan in me

[atomic lagoon]

Beat them in what exactly though

[zealous sparrow]

Another new model. What the hell??

[zealous sparrow]

https://codepen.io/qdpuzepc-the-styleful/pen/XJdxyje

it aint frame-flow but its an AI i havent seen yet.

[vivid coral]

Everything, total and complete victory is needed against them

[atomic lagoon]

You sure this is grok?

[zealous sparrow]

Im not sure what model this is

I haven't seen it before

[atomic lagoon]

What is it called

Eco

Evo

[zealous sparrow]

evo-logic

[vivid coral]

God bless our little AI companies, they putting us to work 🤣

[proud bobcat]

grok 300

[zealous sparrow]

there were 5 grok models put into LMArena uh not long ago

[proud bobcat]

how much you wanna bet it could be like

grok 4.2

[zealous sparrow]

which model

[proud bobcat]

the uhh

swiftflare

[vivid coral]

Aha...it's 4.2, knew it

[zealous sparrow]

yeah its a grok model because it jailbreaks itself on that bench

[vivid coral]

I asked some other models who it might be and they seem to think it's Chinese style

I have doubts on that, fun to ask though

[torn mantle]

what did i miss

any new models or nah?

[zealous sparrow]

yes new models indeed

[torn mantle]

from xai?

[sterile tartan]

Many new models indeed

[torn mantle]

hows the vibe?

[zealous sparrow]

prob not

[sterile tartan]

They are codenamed for now

I have a feeling it might be a new company

[zealous sparrow]

they are codenamed but we know they aint grok

a lot of them claim to be google

although i doubt it

[torn mantle]

@echo aurora you guys should add an announcement for stealth models, its actually a good thing

[zealous sparrow]

From what i know they are all textarena, none are codearena

[torn mantle]

hmm ic

could be from google

[vivid coral]

Probably not allowed in the contracts with these companies

[torn mantle]

gemma or flash 3

[torn mantle]

its not like they will leak the actual ai lab name

a lot of similar services do that

like openrouter

just brings in more users

to try out the models

or else i wouldnt know really

[echo aurora]

It's something we've considered. There are upsides and downsides. Keeping everyone informed on when a new codenames model would be beneficial. However, the concern is it'd lead to people using Battle mode in an inauthentic way just to get to the new codenamed model.

[vivid coral]

Idk, these people are fickle (AI companies) it wouldn't surprise me if they can't

[zealous sparrow]

im pretty sure gemma is discontinued

[proud bobcat]

no

gemma 4 is coming out

[zealous sparrow]

So the incident didn't discontinue it, in the end?

[proud bobcat]

incident?

all i know is that, from tweets of the deepmind devs, a new gemma model is coming out

[zealous sparrow]

it told some person some bad stuff in a uh story or smth

[torn mantle]

hmm i see, a good classification approach could solve that but i understand the issue

[zealous sparrow]

Then google removed gemma from AIstudio

[torn mantle]

yea

gemma is a solid model

[vivid coral]

That's kinda already happening. Not sure how you can deal with that. I get it....

[torn mantle]

i like it

[proud bobcat]

well remember

those are old models

so gemma 4 is gonna be released

theyll just say "oh gemma 3 is outdated anyway"

[torn mantle]

its time

to release it

they have another team working on gemma 4

so they cant just stay idle 😭

also wasnt gemma open sourced?

[zealous sparrow]

https://codepen.io/qdpuzepc-the-styleful/pen/xbVyQmQ
here's code from a supposed google model
imo think its deepseek

[proud bobcat]

most likely google

deepseek doesnt have that style

[echo aurora]

I do wonder if there is some kind of middle ground that'd keep everyone informed but not lead to inauthentic use.

[neat apex]

by theory, gemma 4 27b must have same power than gemini 3 flash 001

[proud bobcat]

gemma 4 is gonna be peak ngl

i cant wait

[echo aurora]

That may be true, but it's a matter of contributing more to a potential problem.

[zealous sparrow]

Looks a lot like deepseek's style tho..

[proud bobcat]

what

[neat apex]

you all think it will keep being true?

[torn mantle]

they are already using it for censored and illegal stuff, they can just tweak it for such prompts

[proud bobcat]

there is no way gemma 4 is going to be on par with gemini 3 flash

[neat apex]

because its what happened to all previous gemma models, the biggest gemma is on pair with the first flash release

[proud bobcat]

yeah but the step up for gemini 3 is insane

[echo aurora]

Hmm I'm not following this, can you elaborate a bit?

[neat apex]

gemma 3 is weaker than gemini flash 2 because its a more recent snapshot

[proud bobcat]

this isnt a 2.0-2.5 thing

2.5-3.0 was a HUGE jump

you can only pack so much into 27B

[neat apex]

yeah, and what is wrong?

[proud bobcat]

well a 27b model is not going to be on par with the huge jump in architecture

[torn mantle]

like when you ask for something bad on lmarena you get blocked by that message 'Sorry i cant...' this already means that they are using a classifier model which like is filtering bad prompts

[neat apex]

samehow they did gemini become 3, they will make the gemma be at 3 level

[proud bobcat]

wishful thinking

but we'll see

[neat apex]

i am even more optimist than that

[torn mantle]

so why cant we just use it for inauthentic use?

[proud bobcat]

i wanna see a Gemma 54B

honestly

[neat apex]

maybe they will release a gemma 4n that is actually better than dense models

[proud bobcat]

dude if gemma 4n releases and has the power of like a 12b model

[neat apex]

yeah, gemma 3n was weird

[proud bobcat]

that is being downloaded on my phone immediately

trust

[proud bobcat]

3n is pretty good

tried it on my iphone 16 pro max

ran well

but lacked intelligence

hopefully gemma 4 becomes usable for a offline daily driver

[echo aurora]

Oh gotcha. Inauthentic use though doesn't just mean prompts that aren't against ToS. It's more-so about voting for a preference without putting in the time to evaluate each response.

[neon idol]

Thx :>

[torn mantle]

i see

[zealous sparrow]

still no coding gen from frame-flow

using through textarena unless some model comes to codearena

good news

I have now gotten a coding gen

@atomic lagoon https://codepen.io/qdpuzepc-the-styleful/pen/qEZJLWw

this is your request

seedream 4.5 dropped in! Hype!

[echo aurora]

[proud bobcat]

NO DEEPSEEK RANKINGS YET

NOOO

[atomic lagoon]

What does seedream do?

[zealous sparrow]

image model

[proud bobcat]

imnage

image

[rustic lichen]

@echo aurora why cant i send any more messages in my chat?

[zealous sparrow]

What's the ratelimit

[atomic lagoon]

Better than nano banana?

[zealous sparrow]

people say equal or worse than NBP

[old garden]

hi @zealous sparrow I remember you

[queen veldt]

Seedream ISN'T EVEN CLOSE to nb pro

I tried it

[zealous sparrow]

Would it be close if it was NBP 2k tho and not 4k

[zealous sparrow]

Anyways i linked a codepen with frame-flow's coding result above

[atomic lagoon]

I don’t see a point in using it then lmao

[echo aurora]

Could be different reasons so difficult for me to say with certainty. However, most common being rate limited.

[atomic lagoon]

It’s pretty cool

[rustic lichen]

its a pretty long chat

so idk

[echo aurora]

We don't have that info publicly available sorry to say.

[zealous sparrow]

Is it possible for frame-flow to come to codearena, or will the models stay on textarena? I think its solid. [its an anon model]

[zealous sparrow]

Hm. Okay, time to find it ourselves then.

[echo aurora]

I'd recommend -> waiting an hour, refresh/clear cache, then trying again.

[rustic lichen]

i cleared my cache refreshed i cant only write on this one every single other chat works even with the same model

[calm lagoon]

@echo aurora Thank you for Seedream 4.5 ❤️

[echo aurora]

Sorry to say for codenamed models I'm unable to provide details/information about.

[sterile tartan]

I was right SD 4.5 took over flux in Image Edit

#announcements message

[zealous sparrow]

I dont know what i was expecting..

[calm lagoon]

Wtf this

[zealous sparrow]

Exactly my question

[echo aurora]

Are you doing the same prompt by chance?

[echo aurora]

Got a pretty good blue dog

[rustic lichen]

i cant do the same prompt i would have to tell everything it needs to know first

[calm lagoon]

Wow

[rustic lichen]

i think its something maybe character limit or something because its not the first time it happened to me aswell

[echo aurora]

Okay can you follow the instructions here and send me a DM of the jamdev so I can share with the team? #1417174113092374689 message

[echo aurora]

Are you able to share the prompt?

[echo aurora]

Yeah the lighting is

[zealous sparrow]

I have found out the ratelimit for Seedream 4.5

It is currently standing at 5 Generations/h

[whole sundial]

yeah I still remember when lmarena had higher limits, I remember getting 20 per hour for gpt image 1 through June to September and then they decreased

[zealous sparrow]

More models, more costs..

[whole sundial]

not really more models, more like more users

[hexed egret]

Hello guys and girls, how can I use the new Seedream 4.5 in here!?

[whole sundial]

due to nano banana

[zealous sparrow]

Go to image arena

[hexed egret]

Ok thanks ✨

[rustic lichen]

where did u get my cat from

[calm lagoon]

@zealous sparrow

😁❤️

[zealous sparrow]

ey, it did a nice job

[whole sundial]

and also the model became lower quality, but maybe they will reconsider because I just saw 4k nano banana pro in battle mode yesterday, that model costs the same as high quality gpt image 1

[sterile tartan]

Maybe flow frame is Gemini Coder @zealous sparrow

[echo aurora]

https://lmarena.ai/?chat-modality=image

[zealous sparrow]

There isn't a gemini coder.. or codex.

[zealous sparrow]

They put it on 2k due to costs.

[sterile tartan]

It might come now

[zealous sparrow]

pineapple can confirm

[zealous sparrow]

If you mean the uh AIStudio app building function

[sterile tartan]

Man just use a upscaler

[zealous sparrow]

its the same but with a prompt

[sterile tartan]

No i mean like a new solely dedicated coding model

[zealous sparrow]

I don't think so?

[whole sundial]

wdym? I meant in battle mode, of course they wouldn't make such an expensive model selectable nowadays

I can prove that I got 4k nbpro if you want

[zealous sparrow]

Oh then yeah, battle mode models are rarer. So they can give us some gens.

But it is never coming to direct, that's for sure.

[sterile tartan]

Yeah you are right

[whole sundial]

yeah, almost like what happened with gpt image 1 high quality but that didn't last too long in battle mode and it was a bit rare anyways, chances could be boosted with multiple inputs

[ember iris]

Finally @echo aurora seedream 4.5 came........ 😁😁😁

[fiery gull]

who is the best ai for cash math?

real use, 🙂

[hexed egret]

I don’t see Seedream on there, what did they name it instead?

[stray aspen]

seedream sucks bruh

[sterile tartan]

Is best for Asthetics

Specially Asian ones

[zealous sparrow]

here is dark-dragons attempt at recreating doom https://codepen.io/qdpuzepc-the-styleful/pen/JoXmwNg

[leaden sun]

they're using some kind of filter for the video generating channels already

[hexed egret]

Nvm I se it now

[zealous sparrow]

I mean, like this can't be google RIGHT?

[neon idol]

:DDD

[echo aurora]

It's there. If you select either Direct or Side by Side modes you should see Seedream-4.5 in the drop down.

[echo aurora]

Ah okay.

Glad to hear it.

[neon idol]

Yeah seedream 4.5 cant beat nbp

Maybe seedream 5

[keen beacon]

I don’t know, dude I’ve been testing it a lot. It has a lot of strength that nano doesn’t.

[neon idol]

For real?

For example?

[stray aspen]

nah

[keen beacon]

I can’t show the examples lol

But they do exist

[stray aspen]

i dont think anything will beat nano banna pro for now

except openAI

[neon idol]

In my opinion seedream 5 will drop on december

[keen beacon]

You guys are forgetting a very critical factor

Content moderation

Artistic freedom

Which you’re not gonna get from nobody, but the seeddream on both of the images and videos

I was blown away

[neon idol]

Yeah but in what type of test seedream 4.5 won?

[keen beacon]

Nonconventional ones

[zealous sparrow]

4.5 dropped this month, so nah.

[neon idol]

I think of yes :>

[keen beacon]

It sucks a lot of the work I do can’t really be shared as much as I want to

I’ll just get banned lol

[fluid hawk]

I found in sora 2 model in website?

"id\":\"043a03a9-a792-4045-9f6d-4bcd747dac43\",\"organization\":\"openai\",\"provider\":\"openai\",\"publicName\":\"sora-2\",\"capabilities\":{\"inputCapabilities\":{\"text\":true},\"outputCapabilities\":{\"video\":true}}

And .

Video production is not added to the site, right?

[zealous sparrow]

videoarena will stay discord im sure

[keen beacon]

I think I just cracked sora fully

[stray aspen]

cap

[fluid hawk]

but which one is the sora2 etc. How do I know?

[keen beacon]

It’s a very interesting model sora

[fluid hawk]

AI Translate Problems.. sory

oh.

[fluid hawk]

how

[keen beacon]

It’s not equally distributed the guard rails

For example, celebrities and famous people have more privilege as you generate things normal people can’t

[fluid hawk]

Im c'understand

:C

[keen beacon]

https://sora.chatgpt.com/profile/snoopdogg

[calm lagoon]

This beautiful

[zealous sparrow]

funniest stuff ever

[keen beacon]

He took some of the content down

[stray aspen]

bruh

[keen beacon]

It’s funny cause this page is all weed

But the app is for 12+

[zealous sparrow]

If i had to say something the new rarest Textarena model to get is frame-flow

I got it twice in a lot of gens

got it rn but it rare as hel

[polar wharf]

Update your discord xD

[polar wharf]

Wdym

[keen beacon]

This is a small example of what I mean about the industry is a dirty industry

They even enacted age gate

[keen beacon]

https://www.tedlaw.com/chatgpt-legal-health-advice-ban-liability-ethics/

This isn’t true

https://www.thestar.com.my/tech/tech-news/2025/12/04/openai-agrees-to-acquire-ai-startup-neptune-to-boost-model-training-capabilities

[proud bobcat]

thank GOD

christ

[keen beacon]

Damn, I’d be so scared. If I was a CEO these guys took massive risks That’s why they get paid big bucks.

They plan on generating $20 billion this year.

But they are in contractual deals of close to 1.7 trillion

That’s so insane

https://www.irishtimes.com/business/2025/12/04/meta-set-to-slash-spending-on-metaverse-as-zuckerberg-shifts-focus-to-ai/

[polar wharf]

🔥

[sterile tartan]

Ty where's Your's?

[polar wharf]

Here

[sterile tartan]

Daym that's alot Better👾

[polar wharf]

XD

I am addicted

To discord

[sterile tartan]

I can understand

If you ever feel is out of control just uninstall it

[polar wharf]

I would install again after hours 🥺

But ye

[queen veldt]

I updated it hello 😭

[zealous sparrow]

If someone has access to it can you test it

[sterile tartan]

Sorry seems like is not available to everyone yet

[stray aspen]

sigma

[sterile tartan]

Is only for ultra users right?

[zealous sparrow]

I refreshed my page and got my own recap

[zealous sparrow]

Yeah

[sterile tartan]

Makes sense since is expensive

[sterile tartan]

Let's go send it

[zealous sparrow]

I don't want to but i will share LMArena is my top 2 server

We need some gemini 3 deep think code examples

[stray aspen]

how did you do that

[queen veldt]

What

[echo aurora]

If you refresh the Discord app you should see it

[sterile tartan]

Agreed

[sterile tartan]

Noise

[torn mantle]

link

[zealous sparrow]

here's a deepthink example from a person in the gemini discord

[torn mantle]

for benchmarks

[zealous sparrow]

found it on a twitter community

[native yarrow]

yessss

[torn mantle]

i dont have deep think 😕

[native yarrow]

yeah it's ultra isn't it

:P

[proud bobcat]

Llama 5 is even going to be worse

facebook just cant make good ai anymore

llama 4 was such an incomprehensible disaster

and they dont have the high quality data

[torn mantle]

yea :/

wait

what happened to llama

[native yarrow]

unfortunate

[torn mantle]

didnt he start over

[native yarrow]

but it'll be on yupp

remember that

[torn mantle]

with a new team

[proud bobcat]

mmmm

[native yarrow]

i think.,,

[proud bobcat]

look at the pixels

[torn mantle]

doubt that because it wont be on API

[native yarrow]

ahh yeah

wait ur right

just like gpt pro

[proud bobcat]

gpt pro is on api

no?

[atomic lagoon]

What the hell is this

Which model I mean

[sterile tartan]

GPT Pro is not Very Pro

[proud bobcat]

gemini deepthink

apparently

[atomic lagoon]

Good lord

[zealous sparrow]

Gemini 3 deepthink

only for Ultra Gemini subscribers

[atomic lagoon]

omg

Lame

[proud bobcat]

they should keep deepthink closed tbh

if they open deepthink up people will just spam it for the most useless garbage

right now i dont think that theres any real usage for any of the big boy ai models

what could you possibly need gpt pro for

[zealous sparrow]

btw you cant use it in API too so it is truly exclusive

[proud bobcat]

and i hope it stays that way

like i cannot think of a single thing anyone would need deepthink for aside from like

major research

and for that you can just use deepseek speciale as a number cruncher and feed the data to normal gemini 3

[sterile tartan]

Ask it about ohayo

[proud bobcat]

what the hell is ohayo

[sterile tartan]

The meme

[native yarrow]

the fact that got pro is still mediocre at writing

lmao

gpt pro*

[atomic lagoon]

Coding. That is what you would need a big boy model for. Game development as well

[sterile tartan]

More like gpt*

💀

[proud bobcat]

claude opus 4.5

[zealous sparrow]

@deep adder this is the new gemini 3 deepthink, someones prompting it
Ultra only btw lol

[proud bobcat]

see i think we forget most ai coding is done in batches

as in

[zealous sparrow]

2nd is 3 pro deepthink

1st is 3 pro

[proud bobcat]

you take a section of code for the ai to make or let it review a section of code

your ahh does NOT need deepthink for that

even sonnet 4.5 will get the job done

[atomic lagoon]

If theres a model better than that then I'd go to it any day. Opus cannot game develop to save its GPUs. We saw in the deepthink that gemini sort of can

[atomic lagoon]

When I code something I am not taking sections out lmao. I am making full projects with it

[proud bobcat]

we have like 3 benchmarks from deepthink

[zealous sparrow]

in the prompt it said for it to be lowpoly

[proud bobcat]

none about coding

[zealous sparrow]

SWE verified too poor to bench it

[proud bobcat]

also gpt-5 pro worse than gpt 5.1 normal 💀

great models openai

[native yarrow]

somewhat at least

[sterile tartan]

Ngl I haven't used Deepseek in a long time

💀

[proud bobcat]

risky

[native yarrow]

deepseek i'm still on the fence about lolz

[proud bobcat]

i think deepseek is a good replacer for gpt atp

[zealous sparrow]

the same person is generating a blackhole shader with gemini deepthink 3 now

it's not done yet

[proud bobcat]

once they upgrade the search feature its golden

[neon idol]

Oh gpt 5 pro released?

[proud bobcat]

been released yeah

[neon idol]

Sucks?

[proud bobcat]

its nothing special

yeah

gpt 5.1 normal beats it

[neon idol]

Lmao

[proud bobcat]

kinda wild

[neon idol]

Open ai mentality is crazy

[proud bobcat]

true true

i could be biased though because i will NEVER support openai

[calm lagoon]

Why the comparison with Claude Sonnet and not Claude Opus? Afraid of losing?

[proud bobcat]

opus didnt release yet (at the time of the benchmark)

[atomic lagoon]

Has been. Imagine being one of the biggest AI companies like openai for years on end then get replaced by a chinese company that has been out for a year

[calm lagoon]

How

[proud bobcat]

opus released like

3 days later???

[calm lagoon]

Aaaa

[proud bobcat]

yeah

sorry

it still beats opus but eh

[zealous sparrow]

Gemini deepthink 3 is the Gemini 3 incident again

[proud bobcat]

i think the arc agi score is the most proving

[ocean vortex]

Opus got smoked

[zealous sparrow]

yet still no SWE bench

[calm lagoon]

Fake

[proud bobcat]

not that badly

[torn mantle]

someone shared this

[ocean vortex]

[ocean vortex]

this literally happened lol

[torn mantle]

only 10 prompts for ultra users

lol

[atomic lagoon]

Why the hell is gemini so expensive

[torn mantle]

so if they made it available for pro users its literally 1 prompt per day

[proud bobcat]

im still waiting for arc agi to bench new deepseek models

sniffle

[atomic lagoon]

LOL

[proud bobcat]

i think 3.2 could easily get 20-25%

[cloud zinc]

77 dollar lmao

[proud bobcat]

192k context is wild

[atomic lagoon]

Nah it will calm down. They just gotta hold it to ultra for a week or so and it'll be good

[cloud zinc]

not worth with 77 dollar compared to human 17

[proud bobcat]

thats painfully low for gemini

[zealous sparrow]

aslo yes

77$ per run

i can get why the limit is 10 prompts per day

[proud bobcat]

"guys, ai is cheaper than humans!!!"

• ai bro

[atomic lagoon]

Then we have uh

less than a penny on every generation for deepseek

[torn mantle]

lol

[proud bobcat]

i think people are too overhyped about ai at tmes

like

lets be honest

[torn mantle]

deep think is what we wanted gemini 3 pro to be

[atomic lagoon]

Yea lol

But also

Who wouldnt be hyped??

[proud bobcat]

a 32B model is gonna do the exact same things gpt 5.1 will do for you on a daily basis

[atomic lagoon]

Another win for the lazy

[torn mantle]

but its weird no

why would they release it now?

i mean we already knew some benchmarks related to deep think

[zealous sparrow]

apparently we got some new image model that flew in

[torn mantle]

why not just wait for oai new model

[ocean vortex]

You really should care less about that singular benchmark tbh

[atomic lagoon]

They were probably sitting on it, realistically

[torn mantle]

then release deep think

or do they have more stuff coming?

[cloud zinc]

gemini 3 flash

[torn mantle]

probably more releases right

[zealous sparrow]

I want to know what SWE will think, because they give gpt 5.1 higher than gemini 3

[atomic lagoon]

Its google LOL 120%

[golden ocean]

this verification image fools nobody

[torn mantle]

sigh

like we are getting these news from different platforms

[still wolf]

Any scores or prejections for deep think?

[cloud zinc]

that image looks 100% ai. plastic human

[torn mantle]

like they seriously need to add some announcements

[proud bobcat]

qwen image

[ocean vortex]

Normal 3.0 Pro is 76.2%
DeepThink would likely do around 80%. But this is not important or meaningful enough for them to advertise

[sterile tartan]

What's going on with all these new models?

[proud bobcat]

look at the text

[zealous sparrow]

qwen?

[proud bobcat]

mhm

[cloud zinc]

qwen imageedit

[proud bobcat]

the text is a giveaway

[zealous sparrow]

I dont know, its just on a roll of new models today.

[proud bobcat]

qwen uses that font all the time

[cloud zinc]

its qwenimageedit

[sterile tartan]

That's wholesome on wholesale

[proud bobcat]

qwen image edit 2509

or maybe qwen image edit preview

or just image preview idk

[ocean vortex]

SWE is a singular thing. A subset of a subset which is coding

[cloud zinc]

there was qwen image edit 2511 supposed to release last month

[ocean vortex]

doesn't tell much by itself A). And B) - easy to cheat when or if you don't care about remaining coding metrics

[proud bobcat]

maybe it is 2511

would make sense

[sterile tartan]

Exactly

[proud bobcat]

cause i dont know any other image model that uses that text font

[austere zealot]

AI engineer who basically lives in that space where code meets machine learning chaos. Most of what I do revolves around LLMs that getting them to behave, getting them to understand context and making sure they don’t hallucinate themselves into another universe. I spend a lot of time wiring models into real products, building RAG setups, tweaking prompts and fighting with vector databases when they decide to not return anything useful.
One of the more interesting things I built recently was an “intelligent helper” for a support team. The idea sounded simple have an AI read incoming tickets and draft replies but the real work was everything behind the scenes. I had to figure out how to teach the model the company’s voice, get it to pull the right info from a big pile of documents and make the whole thing fast enough that humans wouldn’t lose patience. It was a lot of trial-and-error, but once it clicked, it actually saved people hours every week. That’s the part I love: when the AI finally does the thing you meant for it to do.
Anyway, I’m always playing with new setups, new tricks and new ways to make these models actually useful. If you’re into building with LLMs or just enjoy talking about the weird stuff they do, I’m around.

[proud bobcat]

https://tenor.com/view/susie-undertale-confused-gif-23967077

what

[sterile tartan]

https://tenor.com/view/shia-labeouf-clapping-theatre-gif-15313749163340250467

Tldr

[proud bobcat]

why do some guys just come in here and like

put these paragraphs down

i

i dont get it

marketing???

[sterile tartan]

The unemployment rate

[proud bobcat]

LOL

[zealous sparrow]

im tryna find out its generation time

[proud bobcat]

if its 25 seconds its gotta be qwen

[sterile tartan]

https://tenor.com/view/job-job-application-jobless-gif-2757097081210871087

[zealous sparrow]

what model is tangerine

i dont know the Imagearena anons

[proud bobcat]

well

theyre kind of anonymous for a reason

still waiting for deepseek 3.2 rankings....

[torn spear]

Hi, everyone!
December’s already moving fast, and the year’s wrapping up soon.
If anyone’s trying to build something before the Lunar New Year, I’m around and happy to help out.
I’ve been working in full stack and blockchain development, so feel free to reach out.

[proud bobcat]

thank you Zaryon Chan

https://tenor.com/view/labubu-saicotico-gif-15847744479412425038

[atomic lagoon]

Zaryon Chan is inspirational

[sterile tartan]

https://tenor.com/view/job-application-gif-3515684064707868771

[atomic lagoon]

Mario in a tutu

[zealous sparrow]

tangerine gave me this gen
prompt: a cat holding a sign saying give me treats and milk

[proud bobcat]

imagen or gpt image

[obsidian shell]

4.5 is...

Ouch.

[atomic lagoon]

Honestly

[proud bobcat]

seedream?

[atomic lagoon]

Its getting there

[obsidian shell]

Yea.

[zealous sparrow]

a new imagen? It's been so long...

[obsidian shell]

Expected more.

[atomic lagoon]

The paper is crumbled a bit

[ocean vortex]

@deep adder Do you really think Opus is better than 3.0 Pro? Like forget deepThink with that insane cost... Normal 3.0 Pro is better than Opus and also much cheaper

[atomic lagoon]

Welp its making me my stuff lmao

Doing very good for me

[ocean vortex]

Claude has no chance

[atomic lagoon]

Against what

[proud bobcat]

most likely

[ocean vortex]

Against Gemini3

[proud bobcat]

im trying seedream rn

ill give my thoughts

[atomic lagoon]

Gemini 3 pro is worse than opus but gemini 3 deepthink is better than opus

[ocean vortex]

They don't have anywhere near the same compute Google does have. And they trained on less data for sure

[zealous sparrow]

It can't be right.

[proud bobcat]

i asked for a houndskull bascinet bruh

[ocean vortex]

lol what

[proud bobcat]

WHY IS THE ENGINE INSIDE

[ocean vortex]

3.0 Pro is much better than Opus

this is not close

[atomic lagoon]

Did you just not read what I said?

[zealous sparrow]

ghost-pepper takes 29s to generate an image.

[atomic lagoon]

Not at all lmao. For coding gemini is MUCH worse

[ocean vortex]

I did and that's why I said "what"

it's nonsense

[atomic lagoon]

Gemini 3 with coding was good when it released but opus is much better

[proud bobcat]

they probs quantized the model

[zealous sparrow]

It has some nice detail, are you sure its qwen? @proud bobcat

[proud bobcat]

it has to be with that text font

[atomic lagoon]

They're good to put together to code with though

[proud bobcat]

qwen is the only model that gives me that text