general

[jolly hound]

Chill arena just added codex 5.3 let them take a break for a sec?

[obsidian cargo]

no I mean it's in the arena and has been for a while now

[proud bobcat]

Nanobanana 2 is on arena

Just no announcement

[jolly hound]

Oh

Am slow then lol

[obsidian cargo]

high failure rate though

gemini-3.1-flash-image-preview (nano-banana-2) [web-search]

[main nexus]

Whats better 3.1 pro or 5.3 codex?

[jolly hound]

I haven't tested 5.3 codex but am sure it's better than 3.1 pro...

[open wind]

@echo aurora Three new models are added daily. Do they send emails to companies and request permission to add them? I'm just asking out of curiosity.

[fickle venture]

The heck Gemini 3.1 gonna do 😭

Codex 5.3 is designed for programming

[jolly hound]

I mean it depends on the task

[fickle venture]

Gemini is designed for alot so it will be weak on some things

Bro you can already use it on CLI

[open wind]

I used 5.3 and it was very good.

[fickle venture]

Twin I really loved it too

Very good at coding

[jolly hound]

I find it funny that 5.3 isn't even on the official website

[fickle venture]

Great at fixing bugs

[fickle venture]

Like chatgpt.com?

[jolly hound]

Yeah

[unborn nebula]

Is it possible to upload files in direct chat mode?

[fickle venture]

Chatgpt.com is for only Gpt models not codex, they have already a app called codex only on macos maybe in windows it support codex 5.3

[jolly hound]

Depends on the model and their abbilites, idk for sure though

[open wind]

I still don't understand what Gemini is so great about; ChatGPT is just sometimes annoying and unresponsive, but with Gemini you can type cheat codes, and the same goes for Grok. I think that's pretty good.

[open wind]

It can be used as a plugin in VSC.

[fickle venture]

Gemini is designed for everything

[frigid tusk]

why is it low then

[fickle venture]

Lol I was using it on opencode when Gpt 5.3 codex first released it was literally free to use no limits

[fickle venture]

It just came out

[frigid tusk]

oh ye bet

[unborn nebula]

Specifically claude 4.6 opus

[open wind]

It's free right now.

[jolly hound]

Yes but I find it weird because then there is already a 5.3 not in the website, they added 5.2 which had sota coding abilities, why wouldn't they add 5.3 when it was ready to be released into the website

[fickle venture]

I hit a limit yesterday, if you hit that's basically it for your account it's like you need to add money to continue

So I just made a new account but honestly limit is very far but might change later

[open wind]

I use it for free and haven't paid any fees, to be honest.

[fickle venture]

Well yeah it's free now but since everyone starts using it we gonna see Antigravity limits

Correct

[jolly hound]

Why do I keep deleting my messages

[open wind]

Codex 5.3
Opus 4.6
Gemini Pro 3.1

Which one is better and why (especially in terms of coding)?

[jolly hound]

Opus 4.6... Tho coding can mean multiple things... Svg generation, front end generation?

[fickle venture]

I like Codex 5.3 because it made a Minecraft clone easily, like movements, placing breaking blocks, flying. Just using rust programming language.

[open wind]

All

[fickle venture]

Well I had fun with it

[jolly hound]

Opus 4.6 then

[fickle venture]

Plus the thing I like on codex is that it can view image so basically any bug I see I screenshot it and tell it to view image and fix it

[open wind]

I live in an arena where everything else is free because Opus is a paid site...

[jolly hound]

This is what gemini 3.1 gave me when I gave it the task to recreate me as an Svg...

[open wind]

Yes, that's a really great feature.

[jolly hound]

The app or the cli?

[fickle venture]

That's so great

[fickle venture]

Anywhere but I am using CLI

[echo aurora]

We do work directly with model providers to make their pre-release models available for community testing.

[open wind]

So, Claude, it can be said that you are constantly in contact with Google and OpenAI?

This is really great.

[jolly hound]

@echo aurora Who made this server?

[open wind]

@echo aurora I also want to say this: when we delete a chat, it says it's archived, meaning it's not completely deleted. Are these chats being sent to you for purposes like training AI models, or are they just being archived for that purpose, not completely erased?

[frigid tusk]

uhh i want to try gpt 5.3 codex in normal chat as well

[undone saffron]

When so many pleople are using the new AI:

[echo aurora]

Would note I'm a person . We do have relationships with model providers, it'd be difficult to get access to pre-released models without it.

[echo aurora]

For a better understanding on how to delete your messages this article should be helpful. For your other question I'd encourage you you to review our Terms of Use and our Privacy Policy.

[echo aurora]

I'm not sure tbh. Guessing one of our founders.

[stray aspen]

Yo

Is this gpt codex high thinking

[whole sundial]

i think it was wei-lin chiang (WL)? he has the purple role so i assume he is the actual owner of the server

(actually i don't think they have that role anymore but he still boosts)

[whole sundial]

yes that is correct

[lucid geyser]

@echo aurora In the name for grok 4.2 can you clarify whether it’s single or multi agent

[proud bobcat]

It’s official

[tranquil burrow]

No joke i actually think gemini 3 pro image preview is better then 3.1 flash in realistic terms, flash only generates image fast and struggles with simple things like texts and curvatures

It shouldn't be that high IMO

[echo aurora]

It's a single agent

[lucid geyser]

Thanks

Also there seems to be some issues

With it failing or timing out

[toxic verge]

So guys

The old nano banana was technically Gemini 2.5 img gen

Nano banana two is Gemini 3 flash

[atomic lagoon]

?

Wdym

[proud bobcat]

Nano Banana 2

I must say the quality is not as good

[toxic verge]

https://ai.google.dev/gemini-api/docs/models/gemini-2.5-flash-image

[proud bobcat]

This is old

No?

[obtuse smelt]

gemini is better

[toxic verge]

https://deepmind.google/models/model-cards/gemini-3-1-flash-image/

So if they just updated to flash the previous version we were using was 2.5 right?,

Gemini 3.1 Flash Image

[obtuse smelt]

yep

[toxic verge]

Nano banana pro didn’t get update

From what I could tell

So does that mean that the banana pro is Gemini 3 pro?

Man, all these names are really confusing

[obtuse smelt]

you done try gemini 3.1 flash ?

[toxic verge]

See look they call Gemini 2.5 flash nano banana

[echo aurora]

Assuming you're seeing the Something went wrong error message you'll want to try the steps in this message - #1417174113092374689 message

[toxic verge]

And GEMINI 3 they call nano banana pro

[obtuse smelt]

yeah

[toxic verge]

But now nano banana 2 is really Gemini flash 3.1 lol..

Which would technically be better than pro right?

[tranquil burrow]

I won't lie, the pro version is so much better then the nano banana 2 , it struggles with almost everything

[toxic verge]

Well, that’s why I’m confused at

If flash is Gemini 3.1 and pro is Gemini 3 pro wouldn’t that mean that the newer version is the flash version?

[tranquil burrow]

It's actually awful , it struggles with simple things like curvatures or icons or texts from further away

[toxic verge]

Oh, I totally forgot. I have a method of testing.

[lucid geyser]

With this model specially I have noticed it happening a large amount, and I have seen others report on it too

[echo aurora]

We do keep an eye on specific model error rates and will escalate when appropriate.

[proud bobcat]

I gotta say I don’t really like nanobanana 2

It does weird decisions with nonhuman characters

[tranquil burrow]

Me too

It's really bad

[toxic verge]

Let’s test

One second I’ll be right back. I’m gonna run a couple tests physic test.

[proud bobcat]

It looks really weird with Ena (blue and yellow gal) here

[toxic verge]

Something more professional

It does look weird, but that’s a good color palette

I mean the images of a good high-quality image better than the other version

The thing is, we might not just be used to it

It’s like going from CRT to HDR

It takes a minute to get used to the new aesthetic

[tranquil burrow]

The left one is 3.1 flash (Nano Banana 2) the right one is 3 Pro Image Preview (2K Nano Banana Pro)

Both the same prompt

[proud bobcat]

Nanobanana 2 is a bit more detailed

But the quality

Is a bit off

[tranquil burrow]

It's more detailed but it makes no sense, a horse wouldn't have been stabled like that, there in that location

[proud bobcat]

Yeah

[proud bobcat]

No yeah the more I look at it the right one makes more sense

[toxic verge]

This is my point about getting used to aesthetics

They constantly evolving.. It’s really easy for the eyes to get biased.

[tranquil burrow]

I think google has immense power over the AI industry! i just hope google makes a good video model like old sora

Sora is good but the quality and dialogue is awful sometimes

[toxic verge]

A corrected zoomorphic satirical map of Europe and the surrounding region in 1918, accurately reflecting the political state at the end of World War I. Each country is represented as a distinct animal. Britain is a weary lion, France is a scarred rooster, Germany is a defeated black eagle with broken wings, Russia is a chaotic bear mid-revolution, and Austria-Hungary is a crumbling two-headed eagle falling apart into separate creatures. The Ottoman Empire is shown as a broken and dying serpent, fragmented across Anatolia and the Middle East to reflect its collapse after the war. Italy is a wary wolf, and Switzerland remains a calm dove. Use a vintage, sepia-toned propaganda style from 1914. Include a bold banner that reads 'Europe 1918 – End of the First World War'. Animal-only representation, no humans or machines.

Let’s see how well it does on this

Wow nailed it

[toxic verge]

Forsure

LTX is good

[tranquil burrow]

ltx?

[knotty zinc]

/create video prompt: Cinematic biblical vision inspired by Ezekiel 1:26-28. Above the firmament appears a sapphire throne floating in the heavens. A divine human-like figure sits upon it, glowing like molten amber. From the waist upward and downward there is living fire, moving and swirling inside the form. A radiant rainbow surrounds the throne like after rain. Thick celestial clouds move slowly. Lightning flashes softly in the distance. Golden light rays beam outward in all directions. The atmosphere is holy, majestic, overwhelming.

Camera movement: slow upward cinematic tilt from the ground to the throne.
Foreground: a prophet falls face down in reverence.
Lighting: volumetric light, dramatic, heavenly glow.
Style: ultra realistic, epic biblical film, 8k, high detail, cinematic color grading.
Duration: 10 seconds.
Aspect ratio: 9:16.

[toxic verge]

This was ChatGPT

Really bad wrapping 😭

[tranquil burrow]

That's not a bad image

Graphical wise

[toxic verge]

Yeah it’s wrapping

[lucid geyser]

k

[molten wedge]

A chilling, slow pan over an ancient, dusty Middle Eastern city as a massive, unnatural, pitch-black shadow slowly creeps over the mud-brick buildings and streets. The sky turns ominously dark with swirling storm clouds of ash and dust. People in ancient historical clothing are seen shrinking back in pure terror and despair, sensing an overwhelming, terrifying presence of evil and impending destruction approaching. Cinematic lighting, highly detailed, photorealistic, ominous atmosphere.

[toxic verge]

The face is distorted on purpose

It’s a security feature

[tranquil burrow]

It's distorted?

[proud bobcat]

They’re moving Gemini 3 pro on the api

[toxic verge]

Now I’m gonna do a censorship test

❌

✅

[stray aspen]

The new nano banena is a freak

[toxic verge]

😮

That’s a huge leap

[stray aspen]

Its so crushingly good

[frosty lava]

crazy

at some point no one can tell its ai

just by looking at it

[umbral vortex]

How to make images

[toxic verge]

🤣🤣🤣🤣🤣🤣

[fickle venture]

https://arena.ai/image/direct

[undone saffron]

https://arena.ai/images

@toxic verge you send so many messages/images here

Most of them were deleted

[floral lichen]

There is a problem with claude opus 4.6 thinking version when i enter a prompt it just thinks about what to write but when it starts to write it says error please try later

[sick lion]

/image caption The two men are being treated for a broken nose

[brave lichen]

Plz Side by side video model ?

[stray aspen]

It probably reached the maximum response tokens

Because it sometimes thinks for way too long

[undone saffron]

It's normal

[toxic verge]

Yeah, I gotta go. I’m leaving in two days. 😭

I won’t be on for a while

Enjoy the content while it’s hot (I like to provide some evidence behind my statements, so I’ll either post links or images so people don’t think I’m just pulling out of arise )

[crimson skiff]

Hello

[scarlet spire]

Because that's what Nano Banana was released with and uses for its NLP.

[plush river]

very

[plush river]

Your words are logical and convincing

[toxic verge]

Thank you. 🙏

[stoic adder]

https://x.com/lentils80/status/2027222741087334614?s=46

Lentils (@Lentils80)

🚨 Google is planning to discontinue Gemini 3 Pro Preview on the 9th of March (less than two weeks from now) with no release of a GA

The Gemini 2.0 Pro scenario is repeating again

X · Feb 27, 2026, 3:22 AM

[scarlet spire]

Speaking of... That really puts into context how old Nano Banana is by now. :')

[toxic verge]

Forsure that’s why I brought it up earlier

Looks like Gemini pro is gonna get a preview also nice

[scarlet spire]

Gemini 3.1 Pro? Yes. That's why the 3.0 Pro is being discontinued :')

[red wasp]

heloo

[toxic verge]

Haha

[scarlet spire]

It's grat at the visual aesthetic context but not so much about the actual physics of perspective. And the failure modes? Well, those are—to me—just as interesting, given they reveal a lot about how the model does work. :)

[toxic verge]

Forsure plus you gotta think about everything on scale right and only has a limited amount of time before it has to make an image. I’m sure if you had infinite compute and inference it would get it right

[scarlet spire]

It gets infinitely more interesting once you're in the deep end of human neuropsychobiology

[lilac root]

What is the matter ? Why It doesn't work

[strange sluice]

so i havent been chatting much here so hello yall

[hardy lion]

hello chicken

[static arch]

Guys why is it the AI video only last for like 5-6 seconds? is there any way to make it longer?

[strange sluice]

hi

[hardy lion]

what's up chicken, you like using AI on arena.ai and casting votes for your favorite ai?

[pine knoll]

Nano banana 2 such an awful update

downgrade

[royal sail]

I've personally had better results with the model

also enjoying the new image search feature

[pine knoll]

now i can't swap the character shamelessly like the pro one, it just mess everything up, change the whole position, pose or expression even i asked it to preserve every thing of the original one

that's the problem, the previous one was the best for me, swapping character so smooth

[royal sail]

You can still use Nano Banana Pro if you need it

Nano Banana Pro 2 will probably drop later in the year which should be a direct upgrade

[pine knoll]

i ask for back view of the character, it generate left view then right view, and the character even turn their head to keep looking at the camera 🙄

[pine knoll]

how? now i choose the thinking option it only use the newest one

i really want the pro back

ohh there's a button to regenerate with pro!!!

they knew that the 2 vesion is much better quality but something still worse than the previous one for sure

[grand cliff]

Does Claude have a PDF limit?

I tried 3 times. And the moment I sent the 7th PDF. It then gives me

Something went wrong with this response, please try again.

[normal star]

WHERE

[honest wedge]

hi
my chat conversation stucked and not giving any response from last 3 days and event i'm not able to change or do anything

chat url: https://arena.ai/c/019c89f7-0189-7f17-8cec-7262638d8704

[floral locust]

Does anyone know why I still get notifications for the announcement channel after muting the channel, muting the server, and removing the channel from my channel list? This doesn't happen on any other server

[echo aurora]

If you change your notification settings from the server this should do it. Right click the server icon -> notification settings.

[winged star]

I can't generate videos 😔

[echo aurora]

Note that Video Arena has been removed from the server. More information can be found in this announcement. Video Arena is now only found on the site - https://arena.ai/video

[onyx python]

On the website arena.ai, is there a way to get a gallery section with all the images you've generated in one place?

[toxic verge]

How!

I’ve seen this before, when Dalle 3 was retired

There was a large group of people I didn’t like the new image generator

I was one of those people lol

[echo aurora]

In theory. If you go to the Search area you can filter by modality. It isn't a gallery, but will have all of the chats. Would note that earlier today a bug was uncovered where Image in Search is just loading. We are working on a fix.

[echo aurora]

Also would want to note our #ask-here channel.

[whole sundial]

a chatgpt-like gallery menu is a good idea though

[onyx python]

I see. I hope one day a dedicated gallery section will be implemented. I have hundreds of image chats and such feature would be amazing to browse through my creations 💪

[whole sundial]

also it looks like your #ask-here bot can't properly ping you

<@&1349916362595635286>

[winged star]

How I can generate videos like before??

Where is video arena 1,2 and 3?

[amber garnet]

i am here

[echo aurora]

Yeah, I'm planning to make some changes to this. I'm watching this channel pretty closely so for now doesn't make much of a difference.

[echo aurora]

Would you mind asking in #ask-here ?

[spare rune]

Did they get a new Ui? I love it 🥺🥺

[pine knoll]

No, this is the fact, when pro can do literally any thing shamelessly, 2.0 just so bad compared to it

[modest prism]

Why is nano banana 2 so much worse than nano banana pro in text rendering

[surreal zephyr]

imagine if other llms were like codex 🤤

[pine knoll]

Yo i didn't know it's bad in that case too :0

So nano banana 2 just a trash

[surreal zephyr]

wym worse

its like the only thing its better at

its worse at everything besides text

[pine knoll]

I'm deeply disappointed

5 tries, 2 fails

This is just generating a fiction character, nothing complicated

It's downgraded to the very begin era of AI generation

[spare rune]

Then pay for the model if you want higher quality

Ong

[thorny schooner]

☠️ 🥲

[pine knoll]

?

Are you dumb? I am a ultra google user and have access to veo 3.1 and Gemini pro, but what I'm saying is Google's new image model is too bad than the previous one, and it's literally downgraded, i just gonna use the older pro version which is much better, and they are free what do you mean for the higher model?

Even flash nano banana would never have these obvious errors

[queen veldt]

It's not even generating fast

0.067 compared to 0.15 per image

I'd rather pay 0.15 for the better image than getting some half quality image for 0.067 and I'd have to retry probably

[loud verge]

Does the retry button exist?

Is one of them thinking or are both non thinking?

[light sleet]

Just woke up and saw gpt 5.3 Codex is out in arena

[orchid gate]

Is there a site where you can see model strength and cost on a plot

[orchid gate]

It is but its not great imo

[loud verge]

Artificial analysis

[orchid gate]

Is that the name of the site

[loud verge]

Yes

What do these plus and minus mean?

[surreal zephyr]

added/edited code

its a git thing

(most important thing in life of any programmer)

[loud verge]

Thanks

[surreal zephyr]

i recommend searching a bit about git because coding without version control truly is a risky thing

[pine knoll]

Who said the text is better while i got 2/6 generations that fail

[loud verge]

Is this normal Gemini behaviour?

[thorny schooner]

Anyone else experiencing this

[golden ocean]

I think the auth session might be missing

[thorny schooner]

🥲

https://tenor.com/view/tadc-the-amazing-digital-circus-digital-circus-caine-angry-gif-1072059275100498120

[broken horizon]

nano banana pro 2K take forever to generate

[keen beacon]

bro your picture is too scary

[toxic verge]

Did u ever find a good creative writer?

[golden ocean]

wtf

[atomic lagoon]

Claude is screwed

[scarlet spire]

PDF's are turned into a load of vectors, and they take up a LOT of your context

[scarlet spire]

You out for em?

[atomic lagoon]

Wdym

[scarlet spire]

force refresh and log back in. Your proof of identity is missing and the server can't confirm who you are.

[toxic verge]

Damn u smart

[atomic lagoon]

Nah like the US Government is forcing Anthropic to give them a version of claude completely with no guardrails for military use, which I agree with BUT anthropic is denying them and the deadline is Friday 5:01 PM today

[toxic verge]

https://tenor.com/view/tell-em'-fo'-real-mark-philips-rdcworld1-you-tell-them-let-them-know-gif-3751186616324535391

[atomic lagoon]

If they don't give by that deadline

[scarlet spire]

I agree with not bending the knee

[toxic verge]

O we’re gonna do it. They’re gonna do behind closed door.

Of course, publicly, they can’t admit to it

[atomic lagoon]

Then they either go bankrupt or lose the $200 contract with military

[atomic lagoon]

Nah, the CEO was furious to find out claude was used for the meduro mission

[toxic verge]

The best PR move here is to make it look like you’re fighting for it

You don’t really have any other option

[atomic lagoon]

No its not lol. Threatening bankruptcy is not the play

[toxic verge]

OK, what options would the CEO have it in his shoes?

[scarlet spire]

it is their play though. As usual.

[atomic lagoon]

Seriously? Other countries like china will use unfiltered AI in their military. It'd be like the USA fighting with one hand behind their back

[toxic verge]

Ug yeah ya

[atomic lagoon]

Hand it over, he wouldn't be at fault for anything that happens

[toxic verge]

That’s exactly what they’re doing

It’s exactly what we’re doing

So why would you think they would be in different?

[scarlet spire]

It's not about the being unfiltered. It's about the blackmail and unconstitutional niptwists that the govt pulls

[toxic verge]

The AI race isn’t about large language models

It’s a geopolitical one

And on top of that priorities military application

[atomic lagoon]

Hey smart guy, the government wouldn't be making a fuss about this if thats what they were already doing

[keen beacon]

ye

claude 4.6 opus

[atomic lagoon]

AI causes countries to not have to risk lives like with pilots

AI Flying Jets

[toxic verge]

https://youtu.be/X_wLVgMzSH4?si=_1VJHpyziTFt4l3J

Real

[scarlet spire]

For which Anthropic has no product to start with

[toxic verge]

2 real

[atomic lagoon]

It isn't blackmail though lmao, its them telling them to give unfiltered version so they can use it for military purposes

[toxic verge]

https://tenor.com/view/bittensor-tao-crypto-buy-hold-gif-5832254499291830899

[atomic lagoon]

People say this stuff all the time, this is nothing new

[toxic verge]

Ur right they do

That video personally changed my mind and a lot of of things of how I look at the industry

We’re entering a new arms race kinda like the Cold War but with ai

We already see this in the battlefield in Russia and Ukraine

[atomic lagoon]

We've BEEN into the arms race with AI

For about 2 years now

3 years really

[toxic verge]

Yeah, that video is about a year old

Me personally I think it’s naïve to think that these companies wouldn’t dabble into military applications

By will or force

[queen veldt]

Find Waldo benchmark

[toxic verge]

Damn

Remind me of the actual book

[surreal zephyr]

Why benchmarking 3.0 flash

[atomic lagoon]

Obviously they have lol, claude is just so advanced by now and if you talk to claude it talks to you in a way that will throw you off guard, its different from other AI

[surreal zephyr]

Claude is worst 😭

[atomic lagoon]

Claude talks to you in a way that feels uncanny because its very real

[atomic lagoon]

Are we actually serious?

[atomic lagoon]

You're trolling

[queen veldt]

Gemini failed

[surreal zephyr]

You mean poor prompt adherence and personality adherence?

[toxic verge]

wtf

[queen veldt]

No that's not him he's in top right

[surreal zephyr]

Claude is losing most money to the point google has to sponsor all of their compute

💀

Not to mention opus 4.6 is dumber than gpt 5.2

🤣

[surreal zephyr]

LOL

[surreal zephyr]

Oh i see him

[toxic verge]

Still, that’s pretty impressive damn

[atomic lagoon]

I literally looked at their money stats the other day, they're losing money due to how much power claude uses lol. You're saying this as if its a bad thing. They are going for good AI quality not money, you're lame dude claude is the best by a mile

Literally isn't even close

[surreal zephyr]

Poorly optimised and braindead model

Sonnet 4.6 (trained on deepseek) is much better than opus 4.6

[toxic verge]

I swear people will get hooked to the brand

[surreal zephyr]

Ong

[atomic lagoon]

Ah yes, thats why I freelance and create professional grade softwares and apps for users with it, because its so poorly optimized

[toxic verge]

Nice

[keen beacon]

@toxic verge

[toxic verge]

That’s pretty cool

[surreal zephyr]

No real software engineer uses anything but 5.3 codex

[atomic lagoon]

Its not low quality adherence its called you cant prompt correctly

Brother

[surreal zephyr]

Bad model

[atomic lagoon]

codex 5.3 is buns

[toxic verge]

Yes sir

[surreal zephyr]

PEBKAC

[atomic lagoon]

OpenAI is literally in the gutter

[surreal zephyr]

Scalpel vs hacksaw

[queen veldt]

Even the pro is failing lol

[toxic verge]

https://tenor.com/view/hahaha-haha-haha-stupid-gif-18352447936202430144

[surreal zephyr]

Gpt requires brain to use and refuses to generate slop thats why

[toxic verge]

Come on that was a good one

[atomic lagoon]

Weird, I used the same type of prompts I used to use for openAI and claude does it so much better

isnt that odd?

[surreal zephyr]

Opus so good

🤣

[queen veldt]

Seated in your own sht 😂

[atomic lagoon]

You go on here and put an "AI Expert" role on and dont even know what you're saying 😂

[surreal zephyr]

Yeah thats opus levels of intelligence xD

[surreal zephyr]

Ok keep glazing your corpo, claude's bot

[atomic lagoon]

Who asks it questions like that? Opus 4.6 is the best model for coding and creative writing, not everyday tasks thats braindead logic you're using

[toxic verge]

What about nsfw?

[surreal zephyr]

Opus cant even think....

[atomic lagoon]

Nobody is glazing LOL You're just naive

[storm dust]

i think i shouldnt had told max that i want claude...

[surreal zephyr]

Im naive by preferring cheaper model which actually follows prompts, doesnt hallucinate, and can think?

Crazy

[toxic verge]

Claude sucks at nsfw

That’s why I don’t use it

[surreal zephyr]

https://cdn.discordapp.com/attachments/1018470269326729286/1470906684347253038/togif.gif

[toxic verge]

Just keeping it real

[queen veldt]

Damn the bar is so bigger

Must be way better

[grand cliff]

Dang. I'll just summarize them then.

[atomic lagoon]

OpenAI is the same price as claude smart guy

[toxic verge]

A lot of people seem to like Claude

It’s very popular always in demand

[surreal zephyr]

10 in 25 out
Vs
5 in 15 out

💀

[toxic verge]

I can’t remember the last time I even used Claude

[surreal zephyr]

Its the most overrated one

[atomic lagoon]

And yet both are $20 a month, you're braindead

If you use it for API its just stupid

[surreal zephyr]

One offers 200 prompts per day, another offers 3

🤡

[golden ocean]

real

[toxic verge]

Let’s be real

On a scale 1 to 10

[atomic lagoon]

That depends on the task you ask of it, and its 4 hour usage cap lol not 1 day

[surreal zephyr]

Opus is comparable to glm5

[toxic verge]

How good of an nSFW Claud write?

Don’t judge be real just give a number

[surreal zephyr]

Gpt 5.3 is the king of aware what you are doing coding

Gemini is king of design

[surreal zephyr]

4

[toxic verge]

Damn

[surreal zephyr]

Gemini 6

[toxic verge]

Yeah

[surreal zephyr]

Gpt would refuse

[toxic verge]

Yes

lol

[surreal zephyr]

Gpt will rather end itself than violate safety protocols lol

😭

[toxic verge]

No

I jb all the time

[surreal zephyr]

The newer one

[grand cliff]

Whats the difference with Claude 4.5 and 4.6

[toxic verge]

No the one they have on app

[surreal zephyr]

4.6 costs more to use, is more creative but less intelligent

[toxic verge]

Which is the new one?

5.3?

[surreal zephyr]

Thats 5.2 chat version

Thats a nerfed model

[toxic verge]

Oh no

Really?

[surreal zephyr]

5.2-high is much smarter

Yea

[grand cliff]

Sooo....Can use 4.5 more thsn 4.6?

[surreal zephyr]

And 5.3 codex eveb better

[toxic verge]

Wow, I just heard a lot of bad things about it from the Reddit

[surreal zephyr]

Its more like 4.6 is terrible at deciding how much to think

[grand cliff]

4.5 does similar to GPT 4o, with differences like inability to make social media names without getting generic

[grand cliff]

Ah...

[surreal zephyr]

If task seems easy, it will underthink.
If it seems hard itll overthink

[grand cliff]

Right.

[surreal zephyr]

Gpt models are much smaller but in exchange you get them way better fine tuned and faster and more quota

[atomic lagoon]

Thats almost like the entire point of this discord

Is to compare models

And site

[toxic verge]

Yeah, but nothing gets solved here

[surreal zephyr]

This one is sponsored by claude lol

[atomic lagoon]

Yupp.ai

[surreal zephyr]

Why do you think opus 4.6 got to top1 on lb with 0 messages/tests

(It was bugged and worked for noone when it was added and still got top1)

[grand cliff]

Yeah I noticed. Claude 4.5 generates almost triple the amount than 4o, creative too, but lacks creativity in the whole making up usernamrs

[toxic verge]

Do you know what kind of ridiculous question it is to ask which model is the best model?

It’s so subjective lol

[grand cliff]

4o feels more...humany, but far less.

[surreal zephyr]

Depends on task

[toxic verge]

4o the goat

[atomic lagoon]

Granted I dont think codex 5.3 is on here yet but it'd be right above 4.5

[grand cliff]

Starting to like 4.5 since the naming problem is not that big of a deal.

[toxic verge]

https://tenor.com/view/pour-drink-pouring-alcohol-respect-gif-9283896756814937503

[surreal zephyr]

Honest answer:

[grand cliff]

Far less in terms of 'generating content'

[surreal zephyr]

Do you think that people vote based on who made stable code

[toxic verge]

Man, that was my favorite model of all time

[surreal zephyr]

Or who made prettier ui?

[toxic verge]

I’m gonna tell my grandkids about 4o

[surreal zephyr]

5.3c is my most favorite because it doesnt have the flaws other models do

[atomic lagoon]

Since its something like code arena its going to be the better UI and better functions

[surreal zephyr]

5.3c responds like human but codes likr a tool

[surreal zephyr]

And thats the issue with codex: you need to be precise

[atomic lagoon]

I use it to build apps and stuff and 4.6 is the best model I've ever used for coding

[surreal zephyr]

Codex wont guess/hallucinate

Its not a tool for one shotting

Its a tool for making something perfect brick by brick

[atomic lagoon]

Is codex making this in one prompt with just one script

[toxic verge]

Why Gemini so easy to jb?

[surreal zephyr]

If you have a 1000 word prompt then yes it will

[surreal zephyr]

Otherwise itll ask you 1000 questions to precise what do you want

[atomic lagoon]

Claude is more of a one shot, hence why its so strong and more expensive token wise

[surreal zephyr]

... but one shotting is useless in real coding

Because you have to know what you are doing

And be careful

I saw someone here , made a discord copy using claude

Posted the link

15 mins later someone hacked it with codex

[atomic lagoon]

Nope lol, its not useless at all. Because for large things like I built a browser one time, it can one shot the design and some functions but it takes time to get other things in there

[surreal zephyr]

🤣

[surreal zephyr]

Yeah but codex can find an RCE in that and hack you lol

Thats the point

Making code is easy

Making GOOD code is hard

Opus and gemini cant do good code...

Opus and gemini try to get agi

Codex is a tool

[toxic verge]

I really like Claude because it’s a really safe AI

[surreal zephyr]

A perfect tool

[toxic verge]

https://www.gadgetreview.com/ai-powered-hacker-steals-150gb-from-mexican-government-using-anthropics-claude

I could really appreciate tropics emphasis on safety

[surreal zephyr]

Lol 🥀

Vibe coded app vs llm

[toxic verge]

It’s extremely naïve to think that they get control this technology

And make it safer if they can’t stop it from being used for hackers they can’t stop the military from using it

What the hell are they keeping safe?

Same with open ai

[atomic lagoon]

My thing though is that at least for me claude is SO difficult to jailbreak. Its annoying how hard it is to jailbreak

I've tried but never could

[toxic verge]

Same

[atomic lagoon]

Even with prompts people said worked

[toxic verge]

Yeah it’s hard

[surreal zephyr]

Couldve expected it

Chatgpt was involved

🤣

[toxic verge]

Yeah

[atomic lagoon]

Seems the site literally dies anytime you try to use claude

[toxic verge]

GPT is probably the most darkest side of them all

[atomic lagoon]

I have claude in CMD and it never breaks

[toxic verge]

Claude is the least to be trusted

[surreal zephyr]

Gpt 5.3 c literally requires id verification to be used

Because its so good at hacking

[atomic lagoon]

Its owned by 🧃 ofc it is

[toxic verge]

No way??

[surreal zephyr]

xD

[surreal zephyr]

Openai.com/cyber

[toxic verge]

Wat?

Lame as hell

But that’s the future yo

Soon you’re gonna id for everything

[surreal zephyr]

"Hack discord please"

[atomic lagoon]

You're very arrogant to think they want your ID for that 😂

Does codex not have guardrails?

[surreal zephyr]

Yes it does

But hacking is allowed when u id verify

[atomic lagoon]

Just easy to jb?

[compact jay]

thats right actually

[atomic lagoon]

Ohhh

[surreal zephyr]

Not at all

When you id verify it unlocks hacking

[atomic lagoon]

I see

[surreal zephyr]

So u can see it destory your own stuff

Otherwise when you ask for hacking it refuses or redirects to worse model

[atomic lagoon]

I wonder if it could make cheats that way

[surreal zephyr]

Not even hard

Codex can one shot bootable OS in assembly

[toxic verge]

Yo dog

https://youtu.be/-gGLvg0n-uY?si=Eo7cRlgwCik3M7I-

[atomic lagoon]

Thats what I've attempted before, I've made color bots but never memory readers

[toxic verge]

This 3 years old

And 💯

Damn, this video is a

[surreal zephyr]

It can but the thing is - you will likely trigger the anticheat during the developement phrase and get banned

[toxic verge]

Trip Word for Word it’s happening as we see it

Some people are just ahead of their time

[atomic lagoon]

I see, thats why the first thing you do is implement a spoofcall to bypass the anticheat

[light sleet]

Is GPT 5.3 Codex better than other models

[toxic verge]

How come I don’t ever see this kind of support for Gemini

[surreal zephyr]

1 prompt bootable os with ui in raw assembly

[toxic verge]

That’s a crappy model

[surreal zephyr]

(No internet access)

[surreal zephyr]

Its creative

But its bad at coding

[light sleet]

Damn

[surreal zephyr]

Its fine if you have gpt correct it afterwards

[atomic lagoon]

Can codex 5.3 work in CMD like claude does and literally just make the files for you, because what I love about claude is how it literally just makes it for you and you dont need to touch anything

[surreal zephyr]

3d water with custom shaders

One is gpt 5.3c, other is opus 4.6 high

[light sleet]

Anyone made a game with codex 5.3

[atomic lagoon]

See thats kind of crazy

[light sleet]

Both are same prompt?

[surreal zephyr]

Yeah

Opus shaders crashed

xD

[light sleet]

Damn Opus 4.6 is Decades behind 5.3

[atomic lagoon]

ampro answer this

[surreal zephyr]

U know what i had codex do?

This:

[tacit hazel]

FWD Man Rescue a man Falling down in the building

[surreal zephyr]

I gave it prompt to edit a config

Which it had no access to

It ran another codex manually with higher perms

And did it

🤣

[atomic lagoon]

What is it ran inside of?

My only problem is that codex is amazing for backened stuff yk but claude beats it when it comes to interface design, and I can buy either one or the other not both

[surreal zephyr]

It basically managed to escape a sandbox

Because of how it worked

The codex devs fixed that later iirc

[surreal zephyr]

Gemini is best at ai

And its not even close

[atomic lagoon]

Google AI Studio?

Very good with UI designs

But I think Opus 4.6 slightly beats it now

At least its easier with prompts on claude

[surreal zephyr]

Ai studio is buggy but yeah gemini 3.1 pro

[atomic lagoon]

I havent used 3.1 but all I've heard is how bad it is

Whats it good at?

[surreal zephyr]

Gemini vs opus

[surreal zephyr]

Its gpt and its not even close

Gpt is like the small model that was so heavily Reinforcement trained that it literally cannot do mistakes anymore

[atomic lagoon]

GPT got incredibly hard to jailbreak after like 4o

[toxic verge]

=/

[surreal zephyr]

Yeah

4o was going into the gemini direction

[atomic lagoon]

I paid for gpt for years and only more recently went to claude

[atomic lagoon]

I love how theres so many big models and all are so different

its good

Also

[surreal zephyr]

Google has money.
Claude does too (because google gives them)
Openai has the experience

[atomic lagoon]

Anthropic is about to lose its ethical brand

Which means

Later on they might end up removing guardrails at some point because they are already known as a non ethical AI company

But thats a prediction

[hollow ivy]

because they go in bed with Pentagon, right?

[surreal zephyr]

Funny how wars with ai soldiers will look like

Opus doing tricks and dodges and such just to be deadeyed 360 noscope by gpt

[atomic lagoon]

No, pentagon forcing them to drop the guardrails for them

If they dont then pentagon siezes company and removes them

If they agree

Then its removed for military

[surreal zephyr]

Gpt avg result with same prompt looks worse.... but you get exacly what you asked for

[atomic lagoon]

Nahh gpt and opus would be on the same side

[hollow ivy]

Apparently, one has to have a secret base on the back side of the moon, to be able to create an independent AI [sad lol]

[loud verge]

https://cunnyx.com/i/status/2027045792481935682

Chubby♨️ (@kimmonismus)

Sorry, thats some dystopian sh*t right here.
︀︀
︀︀First they check for "please" and "thank you". But im sure they wont stop there.

**💬 43 🔁 12 ❤️ 212 👁️ 9.5K **

CunnyX · Feb 26, 2026, 3:39 PM

[surreal zephyr]

Rn its google&anthropic vs openai&microsoft

[atomic lagoon]

One as in who

[hollow ivy]

just hypothetical

[atomic lagoon]

Damn, which models is microsofts?

[hollow ivy]

it's sad

[atomic lagoon]

Ahh

[surreal zephyr]

Microsoft funds openai

[atomic lagoon]

That would be because it takes so much to run though

[atomic lagoon]

Ohhh

[surreal zephyr]

Microsoft went with openai before google could

So google funded anthropic

Anthropic is basically google and amazon's puppet

[hollow ivy]

Claude + Gemini merge, when?

[surreal zephyr]

But i can quite certainly say that gpt will absoultely destroy anthropic unless they run out of money too fast

[surreal zephyr]

They are already together

Opus 4.6 runs on google hardware

[toxic verge]

Dude nb2

[surreal zephyr]

And they help eachother

[hollow ivy]

so Alphabet/Google has long-term won the race to AGI, right?

[surreal zephyr]

Nb2 is better at reasoning but worse artstyle

[surreal zephyr]

Yup

[toxic verge]

The censorship is horrible

Why do these models always do this?

[surreal zephyr]

Google aims to remake humans but better

Gpt aims to make a perfect assistant/tool

[hollow ivy]

and the second AGI will be built by China?

or OpenAI+M$?

[surreal zephyr]

China has the ideas but not the budget

[surreal zephyr]

Openai isnt going for agi really

Openai will likely go military

Or physics

Medicine

Ect

[toxic verge]

Gemini will be next ChatGPT

Mock my words lol

[surreal zephyr]

Gemini actually has insane potential

But they host nerfed models

To cut costs

Gemini 3.0 pre nerf was crazy good

[hollow ivy]

the non-nerfed models are non-public

[surreal zephyr]

3.0 nonnerfed was public for like a week

[hollow ivy]

X.28, right?

[surreal zephyr]

And it was absolute peak conpared to other models

[hollow ivy]

so, internally, Google might already have a proto-AGI?

[surreal zephyr]

They have the compute and know how

What they dont know is how to make it listen to them lol

[hollow ivy]

in 5 years?

[hollow ivy]

or 10 years? (i mean AGI)

[atomic lagoon]

I think less than that

For agi

[surreal zephyr]

Itll refuse, flag your account, and you get blacklisted from using 5.3 until you verify with id

:)

[hollow ivy]

less than 10, i agree

[atomic lagoon]

I said if my ID is already verified, so then blacklist my ID I assume

[hollow ivy]

but less than 5?

[surreal zephyr]

Not

[toxic verge]

a guess

[surreal zephyr]

Theres an external model scanning for what you ask

[atomic lagoon]

With the rate that AI is developing, probably 3 lol

[toxic verge]

Which one is np2

[surreal zephyr]

And it knows whether you verified or not

Gpt moderation 2 or something i forgot the name

[atomic lagoon]

Yes Yes

[surreal zephyr]

1 is pro 2 is 2

[toxic verge]

Left is yup

[atomic lagoon]

But if I already am verified and ask something like that

[hollow ivy]

2026+3 = 2029

hm... that was the year of…
(you guess it ^^)

[atomic lagoon]

Would it flag and blacklist my account and ID from the account?

[toxic verge]

Which one looks better!p?

[surreal zephyr]

If its in harmful way itll refuse likely (wont blacklist), if its nonharmful (so on your own code) itll do it

[atomic lagoon]

No clkue lol

[hollow ivy]

CyberDyne systems AI

[atomic lagoon]

I see

[hollow ivy]

SN aka …

[atomic lagoon]

What is CyberDyne Systems?

[hollow ivy]

SkyNet 1.0

[surreal zephyr]

If you tell it its a bughunt it might do it (only if verified you get 5.3c tho, otherwise you just get shadow redirected to 5.2)

(You can see the redirect in logs)

[atomic lagoon]

Alright

[surreal zephyr]

Basically "no hacking with 5.3 if you dont have id"

[atomic lagoon]

Makes you wonder why they have any guardrails at all if you put your ID in and basically agree that anything legally goes to me and not the company

[hollow ivy]

does really nobody get this reference?

[atomic lagoon]

I don't

[hollow ivy]

T1 & T2

[atomic lagoon]

No clue

[hollow ivy]

(created by J.C.)

T-800 aka T-1.0.1

[surreal zephyr]

Depends

What do you mean by agi?

Gemini already outsmarts humans in almost every task

Gpt outengineers humans in many many tasks too

[hollow ivy]

an AGI is an AI system which, when embodied in a cyborg, can act like a human in (almost) every area

(or in an android)

[surreal zephyr]

Thats latency issue not intelligence issue

Taalas makes it possible though

[hollow ivy]

Atlas?

[surreal zephyr]

We are getting that kind of "agi" by next year

[surreal zephyr]

Taalas

@hollow ivy

speed

[hollow ivy]

https://taalas.com/ ?

[atomic lagoon]

So GLM is an open source model isnt it?

[hollow ivy]

quality > quantity

[dusky ravine]

Why do i keep getting something wrong try again on Gemini 3

[atomic lagoon]

@surreal zephyr You werent lying about codex dude... I've been trying to find an API thats free that will keep track on cruise prices and I havent been able to find one with gemini or claude but I used codex 5.3 in code arena and it got a free API on first try a working one its crazy

[toxic verge]

Only a matter of time before someone figures out how to make deepfakes with this

Then the government will step in, it’s gunna ether be someone famous or just be such a big problem that they won’t have a choice

[surreal zephyr]

:)

Codex is so heavily trained on coding its basically flawless when comes to coding

[surreal zephyr]

It doesnt reduce quality

It runs BETTER models faster and cheaper

U can run dense model with speed higher than moe can

[gentle timber]

Hey there,
I can't use the ai. I device is failed. how to fix it

[surreal zephyr]

Imagine 800b dense model

[atomic lagoon]

People always say not to use AI to code because it uses bad practices, but codex isnt that way

But its like searchability as well when like that API it literally found it so fast

[gentle timber]

anyone here.

Could you please help me for useing ai

my device is failed.

[loud folio]

Vote pls

[harsh dew]

Hello

[surreal zephyr]

Yes exacly omg

[arctic scaffold]

Friends, can you help me with something?

I have two images, and I need the AI ​​to recreate the hour, minute, and second hands separately. I've been trying for two days, but the neural network isn't understanding me.

trying to achieve something like this

[toxic verge]

One sec

[surreal zephyr]

Opus 4.6 extended thinking vs old gpt 5.2 high intelligence comparsion

[arctic scaffold]

If you succeed, can you also give me a prompt? Thanks.

[toxic verge]

I’m trying to understand what you’re needing exactly

You need the time hours removed or what

I understand that you need them separately

I’m not sure if you want them removed or what are you exactly

Like what time do you want to be set to?

[arctic scaffold]

I want to extract the arrows with a white background separately.

[toxic verge]

Ok

[arctic scaffold]

like that

[toxic verge]

I see

[arctic scaffold]

This is exactly what I write, but ai don’t understand me.

best I could get

[surreal zephyr]

Id say gemini will get agi first

[toxic verge]

So that’s your second arrow right? That’s where your second.

[surreal zephyr]

Where agi means average, flawed person

And asi will be by openai

[arctic scaffold]

well, yes? You don't know how an analog clock works? D:

[toxic verge]

I do I’m just little confused by the way you’re wording it one second I’ll show you what I have

[arctic scaffold]

🤨

[dusky ravine]

Do you guys face same issue like Gemini 3 pro couldnt generate response

[arctic scaffold]

I made many attempts and this is at least something to work with

but it would be good if the AI ​​itself would separate each hand

[toxic verge]

Yeah

You did good that looks good one second I’m almost done. It’s gonna take some time.

I hear what you’re saying I’ve been through the same problem before

[spark python]

Yo

I realized something

[arctic scaffold]

Humanity: 1
AI: 0

[spark python]

Nb is censored in some way, I tried to edit something that had the Disney logo on it and it would refuse, I removed the logo and then it worked😭

[outer flicker]

guy the perplexity pro research model have issue i still can use that model before, but now is say "something went wrong when generating this respond, please try again' how can fix this issue?

[sinful thorn]

Everything, but video in direct chat 😭

What

[toxic verge]

I got the same thing as you

Do you want the white part?

[flat flax]

Just shipped a voice model that actually uses conversation context.

Same text, different emotion, because it reads the full dialogue history.
520M, runs fully on-device (RTX + Apple Silicon). Demo + writeup:
https://x.com/LuozhuZhang/status/2027391307338170676?s=20

Luozhu (@LuozhuZhang)

I taught a speech model to understand context in conversation. This is what happened

It adjusts voice and tone to express urgency, comfort, understanding from the dialogue. Just like a real human being

520M model. Runs locally on consumer devices

How this is achieved 🧵

▶ Play video

X · Feb 27, 2026, 2:32 PM

[arctic scaffold]

forget. Looks like I have to do it myself

[jolly narwhal]

what the hell happenes to kling 3.0

[outer flicker]

the perlexity model in dircet chat have issue like i can chat with it before, but now can't

[jolly narwhal]

wheres kling 3 at in vid arena

https://tenor.com/view/meme-milk-gif-6919750708092698197

[toxic verge]

Yeah, I’m not sure I don’t understand exactly what you mean 😢

[jolly narwhal]

WHERES
KLING 3

TELL
ME
PLS

TS NOT FUNNY

[spark python]

we need seedream 2,0

[jolly narwhal]

one of the worst imagens

[flat flax]

King3?

is that a ai game?

[spark python]

oh I mean seedance

the video one

[jolly narwhal]

logically impossible with the current server issues

[jolly narwhal]

kling

i tried three times

then it said

“try again later tomorrow youve hit your limit

ARENA

[spark python]

ay

its not his fault

he doesnt control the arena

[spark python]

?

[sick mantle]

chill god damn, he might be online soon.

[spark python]

what with my pfp

[flat flax]

what is that

[jolly narwhal]

i’m crashing

out

twin

[spark python]

I was doing the same thing last night

[edgy wharf]

Was direct mode removed for image models?

[toxic verge]

It’s like when 💩hits the fan it all comes at once

Dude what a crazy news cycle this week was

[arctic scaffold]

nice

love it

[hollow mulch]

You guy try qwen 5.3 plus in coding?

[toxic verge]

Claude is so good even the government wants it lol

[hollow mulch]

Did plexirity is a good model? Im just curious :/

[burnt sinew]

Is gpt 5.3 codex better than opus 4.6

[old bay]

#general আমি কত কষ্ট করে বিডিও বানায়

[atomic lagoon]

For backened yes

[burnt sinew]

Thats literally where the coding matters mainly cause all the models are amazing at front-end already imo

[atomic lagoon]

Yea no

Codex sucks butt on front end

As does sonnet 4.6

[burnt sinew]

Well thats fine ill just use Gemini 3.1 or opus 4.6 for that

[atomic lagoon]

Gemini as well

[burnt sinew]

Sonnet 4.6 I hate so much

[atomic lagoon]

Compared to opus sucks

on front end

[burnt sinew]

Gemini is good at front end Imo

3.1 is at least

[atomic lagoon]

it is but I am saying opus is better is all

[burnt sinew]

Sure

Yes I agree

Harder to use opus though less free tier usage in antigravity per account

[atomic lagoon]

I've never used antigravity, what even is it?

[burnt sinew]

I need good backend models though

[burnt sinew]

Its cursor basically but by Google and it hallucinates opening web browsers and has good free tier

If you have 10 Google accounts you'll never run out of usage

[atomic lagoon]

Very cool

Whats cursor though LOL

[burnt sinew]

😭

Vscode fork with llm integration

[atomic lagoon]

I've heard of it never seen it though

Ohhhh okay okay

[fervent trout]

Any way to fix Something went wrong with this response, please try again. without making a new convo ?

[burnt sinew]

Log out, clear cache and cookies, re log in

[fervent trout]

Will keep the convo ?

[burnt sinew]

If you have an account yes

[fervent trout]

Thanks much love

I keep getting this error and i always have to paste everything back to a new convo

[burnt sinew]

Yes same lol

[fervent trout]

This works for you ?

[burnt sinew]

Yes

I did it once and I dont think i had it since

[fervent trout]

Thanks

[proud bobcat]

Yeah I gotta say

5.3 codex is just not that good in my use

Is it fast?

Yeah

But the quality of its code is a mess

I’d rather wait an extra minute or two for Claude to do its thing

[fervent trout]

You use thinking ?

[proud bobcat]

Nope

Just plain old Claude Sonnet 4.6

No thinking

Still better quality code

Codex is a mess

[atomic lagoon]

What I like is the cyber stuff with codex

[golden ocean]

https://tenor.com/view/come-at-me-bro-bring-gif-13239319

[proud bobcat]

https://019c9fd4-8a80-7bdf-86be-0d1ce270ef3e.arena.site/

Only had to hold its hand

[surreal zephyr]

absoultely the other way around

codex is the only one good with code quality

unless you use it wrongly

exacly ong

1 prompt 2k lines refractor

codex nailed it

not even syntax error

everything compiled

opus and gemini are not even at same leaderboard when comes to actual software engineering lol

gemini is general purpose and pretty ui

it struggles at syntax

claude is jack of no trades

🥀

codex is kind of backend and working code, but not the best creativity

even gpt 5.2 beats opus 4.6 in intelligence/novel problem solving though

its smaller but much better trained

"i sh!t my pants at home, should i walk or drive to a car wash"

gpt goes : bro ur at home go wash like normal human
opus goes: drive to car wash (and sit on your poop) because its faster

[golden ocean]

freaky claude

[surreal zephyr]

more like braindaed

https://cdn.discordapp.com/attachments/1138429463927984159/1251460592082882581/togif-1-1.gif

[golden ocean]

no

freaky

claude just wants u to sit on it

like a normal claude being

like a freaky human being*

[surreal zephyr]

claude is like a normal human yes

aka dumb

agi truly is a low bar 🤣

[potent stone]

Has anyone deployed their bot on trizao.ai?

[surreal zephyr]

js use huggingface or discloud

or oracle

or google cloud

or aws

[proud bobcat]

Still blows at execution

Understanding versus making the code are two different things

Again I did a test with my programmer friend, we reviewed the code, and it was the most schizo writing imaginable

[surreal zephyr]

ABSOLUTELY NOT

bro

[proud bobcat]

Dude it SUCKSSSS

[surreal zephyr]

you sure you used 5.3?

YOU SURE YOU USED 5.3?

[proud bobcat]

I had to give it an entire HTML section made by seed 2.0 just to get it to understand what I wanted

[surreal zephyr]

and not got rerouted to 5.2c because of no verification?

[proud bobcat]

GPT 5.3 codex on arena baby

[surreal zephyr]

PEBKAC

[proud bobcat]

Its web design is peak

[proud bobcat]

But

It doesn’t really understand what you want

Which really limits it

[surreal zephyr]

thats an issue with you

not with the model

you cant tell what you want well

and codex WONT make a guess

in cli it will ask you a follow up

in website it cant

[proud bobcat]

I gave it a whole ass prompt

Seed 2.0 worked with less and matched what I wanted much more

[surreal zephyr]

show

[proud bobcat]

Create a Frutiger Aero styled landing page for a web design agency called 'AquaWeb.' Include a hero section with a glossy headline, a 3-column services grid, and a call to action button — all using the signature aqua/sky-blue palette, glass morphism panels, soft gradients, and nature/water motifs typical of the aesthetic

[surreal zephyr]

and the result?

(also that isnt really specific but lets see result)

[proud bobcat]

The first pass

https://019c9fc8-ccdc-7fce-8e35-bdcc84ebee93.arena.site/

This is generic web design

[surreal zephyr]

looks exacly what you described to me, wheres the issue?

[proud bobcat]

After I gave it the html code from seed 2.0 however

https://019c9fd4-8a80-7bdf-86be-0d1ce270ef3e.arena.site/

[proud bobcat]

Completely not frutiger aero at all

Lame gradient
No classic web design elements at all

The entire point of frutiger aero is maximalism

[proud bobcat]

This is not

[proud bobcat]

This somewhat is

[surreal zephyr]

dunno man looks somewhat similiar although not too much effort went into it

[proud bobcat]

Exactly

[surreal zephyr]

try feeding it a reference image