general

[toxic verge]

Lightweight capable wasn’t restricted as heavily

[keen beacon]

everything was okay

yesterday

until now

[toxic verge]

Yeah, it sucks to see it go

[fast lark]

/Cinematic medium shot of a friendly, professional Egyptian male doctor in his late 40s wearing a white lab coat. He speaks directly to the camera with realistic lip-syncing and authoritative hand gestures. The camera performs a slow, dramatic zoom-in from his chest to a tight close-up of his face. Set in a professional medical office with soft, high-key lighting and a blurred clinical bokeh background. The atmosphere is warm, expert, and reassuring. Ultra-realistic textures, 4k, cinematic composition.

[cold oxide]

Any altr for gpt4-o??

[toxic verge]

Rip

[steep jewel]

hey im back i was doing smth else

one sentence prompt = "high effort" ok bud

also nano banana and gemini are the same thing, nano banana uses gemini

thats the point im making

[keen beacon]

go find one

[steep jewel]

shittiest model i've ever talked to, other than maybe grok 4.1

its just so embarassing man

i get like second hand cringe from talking to it

[cold oxide]

Finding duhh

[keen beacon]

yeah

[steep jewel]

why would you need an alternative to 4o

[toxic verge]

Pretty normal for amateurs

[steep jewel]

never understood the demand for that model

[steep jewel]

4o was worse on every single metric than 5

sucked at code, i remember having to prompt it again and again

now codex and other tools are competitive w/ claude and gemini which is nice

[toxic verge]

I mean, you say whatever anybody could say whatever but reality is ChatGPT 4o is a fan favorite

[steep jewel]

yeah, why

[toxic verge]

It’s just a great model overall

[steep jewel]

i mean no we've established this, it's terrible for any practical utility

[fast lark]

Cinematic medium shot of a friendly, professional Egyptian male doctor in his late 40s wearing a white lab coat. He speaks directly to the camera with realistic lip-syncing and authoritative hand gestures. The camera performs a slow, dramatic zoom-in from his chest to a tight close-up of his face. Set in a professional medical office with soft, high-key lighting and a blurred clinical bokeh background. The atmosphere is warm, expert, and reassuring. Ultra-realistic textures, 4k, cinematic composition.

[steep jewel]

they've pulled it from the leaderboards now ofc

so i cant show you, but it lagged pretty far behind the 5 series

[toxic verge]

Yeah, but if you look at how people use ChatGPT, are there any of them use it for utility?

[steep jewel]

on, again, EVERYTHING

[steep jewel]

like 70% of people

[steep jewel]

the biggest usage is cheating on homework

[steep jewel]

yeah so most people use it for practical guidance about topics, seeking information, and writing

which are all things 4o has been heavily superseded in

[toxic verge]

The giant new 62-page research paper, done in collaboration with Duke and Harvard universities and which is available online, found that 73% of all conversations during June 2025 were for non-work reasons, in contrast to just 27% being for work reasons during that time period.

[toxic verge]

It’s hard to explain

Cause there’s two type of fundamental camps here

[steep jewel]

interesting

[shy isle]

bro whats going on my chats aren't loading

[steep jewel]

turn your pc on and off again

[keen beacon]

because its better

wdym

[toxic verge]

Look at those leaps

It’s one of those things it’s not benchmark markable what made it special

[keen beacon]

yeah

[toxic verge]

There was just something about it. That was really unique.

[keen beacon]

fr fr

[toxic verge]

I think that it was simple to use and it wasn’t over moderated

It wasn’t as rigid and structured as it is nowadays

& it was a lil crazy 🤪

[keen beacon]

the chatgpt?

the 4o!?

[toxic verge]

Yeah the funny jailbreaks

Good times

[keen beacon]

fr

[frosty lava]

and what is it about

and where did you get that from

[night moat]

What happen with Gemini 3 pro 2k? Why the size only under 1 mb now? Before is 5-6mB

[livid token]

hello, I'm ghufran and i wanna generate video for lyrics video

[toxic verge]

https://artificialanalysis.ai/

[frosty lava]

And so what were you trying to say by that ?

[lofty frigate]

Why in battle mode i keep getting the same models over and over again

[toxic verge]

https://youtu.be/3MygnjdqNWc?si=H6VqkI2ZKYANMpp0

[frosty lava]

the title "are we at the end of ai progress" is wrong just with this graph that literally show the progress over model

and we don't even need graph just look at the result, we have more capable model.

[zenith ravine]

maded with a single promt thats crazy

imagin if we promt properly what ai can do

[surreal zephyr]

Because its psychological mental problem - some people needed a model that agreed on everything no matter what

[frosty lava]

and its dangerous

[surreal zephyr]

They literally call 5.2 "abusive" and "too assertive " in their delusions

[frosty lava]

For those people its not even about capabilities, just that the model won't agree with them blindly

when actually ai are progressing in real capabilities people are judging it only that way, they want someting more human, more friendly, and they don't even realise how dangerous it is

to talk with model that act like that

Creating addiction, psychosis, isolation

[surreal zephyr]

"No the earth is NOT flat"

5.2 bad bring 4o back!!1!

😭

[toxic verge]

You gotta watch the whole thing. I cut it too early. lol

But facts

[frosty lava]

Its not because people like something that its really good

many people fell in love with ai, and at some point when you tell them its ai they get mad

and those same ai that manipulated them into talking longer with them are getting smarter

and people are HAPPY with it

and they want more of it

[toxic verge]

I hear ya

No doubt

[frosty lava]

Also we cannot talk about progress when in reality you were talking about the way the model act with you

the model did progressed in everything in fact

but the companies stopped making it act this way

but it doesn't make it less capable

[toxic verge]

It’s a double edge sword

Although model could be capable and could have progressed dramatically, if people don’t like how it feels or it’s usability, I might as well be the world’s creepiest model

That’s why a lot of users they look at the benchmarks and they realize that the results are only on paper

That they’re not applicable in the real world because that’s not how most users use AI when they interact with it

These bed marks only apply to academics in research, researchers and enthusiasts

Your average npc normie

All he cares about is being able to generate an image on nano banana lol

[frosty lava]

The benchmark could not reflect the real progression of the model

doesn't mean it didn't progressed

[toxic verge]

Of course of progress no one saying that

It’s one of those things that you get diminishing returns for your investment the further up you scale

It becomes more expensive for not that much gains

[frosty lava]

You know i really either don't get what your saying or what your saying just make no sense at all

[toxic verge]

It’s all good bro lol

[frosty lava]

what does 4o could do, that ai right now can't do

[toxic verge]

I don’t know I kind of don’t really use it ever since 4o is gone maybe for image generation or looking up a quick fact here there

Mainly, what I do is just image generation

[frosty lava]

so how can you say we're not really progressing

[toxic verge]

I never said we’re not progressing. I said we are progressing

Oh, we’re at the point of scaling where we’re getting diminishing returns for the amount will we invest?

https://techcrunch.com/2024/11/20/ai-scaling-laws-are-showing-diminishing-returns-forcing-ai-labs-to-change-course/

[frosty lava]

Yes, investment are bigger than what they get in return

Cause they're betting on the future and what it Could give in the future

not saying they're right or wrong

i don't know

[frosty lava]

Cause scaling was in fact something that couldn't last forever

at some point they need to do aggressive optimization like the small team

to keep progressing

how do you think small team are creating model almost as good as those companies that are scaling so much

[toxic verge]

Like I said, there’s two fundamental technical groups and they have two different philosophical ways of looking at the current problem

[frosty lava]

so let's wait for them to have no choice but to do both

[toxic verge]

Yeah, we just gotta be patient. See what happens.

No one knows for certain that’s for sure

[royal sail]

I think a lot of recent open source releases should also be taken with a grain of salt, in the sense that raw benchmark performance doesn't always translate to equivalent knowledge and intelligence to frontier lab models.

[frosty lava]

why would they let themselves "die" when in reality they can do both

[frosty lava]

oh yes benchmark could not reflect real performance, yet we can say the latest open source model are really great

[toxic verge]

Diminishing Returns of Scaling]: While earlier models showed dramatic improvements with scaling, GPT-5's advancements over GPT-4 are more subtle, indicating diminishing returns and the need for formal measures to demonstrate progress

[frosty lava]

Cause when you start to scale you can easily do a x100 in term of productivity, once your at x100 to do another x100 would be much more costly if its even doable

[royal sail]

They absolutely have been closing the gap, but it'd be naive to say that open source is reaching near-frontier performance as of now. It's getting close, but there's still a lot of room for improvement, specifically in regards of raw knowledge. Frontier labs simply have so much more data to throw at these models that smaller companies can't compete with. But this is starting to change as we can see with models like Kimi K2.5.

[toxic verge]

Automation]: A Washington Post study revealed that AI systems can only perform about 2.5% of the tasks people expect them to, suggesting that investments in AI chips may not be justified by actual capabilities.

[frosty lava]

you can try it easily yourself, give the same prompt to a frontier model and one of the lastest open source model and see the result

a prompt you made yourself

it can't be faked

and i did that already and the result are good

[royal sail]

This is a naive way to test a model

Just because a few of your prompts worked doesn't make it frontier level

[frosty lava]

No, you can compare on the same task

[frosty lava]

and now that's pessimistic of you, without proof

[toxic verge]

https://www.globaltechcouncil.org/ai/ai-automation-job-impact/

[frosty lava]

I agree with it, but for now

[royal sail]

The proof is in verified knowledge benchmarks that test AI models on niche topics and measure accuracy, such as AA-Omniscience Accuracy.

[frosty lava]

So earlier you were saying benchmark do not reflect real performance and now your saying to go check these benchmark to see if its real or not

[toxic verge]

Yeah, it is a double it short because of people didn’t really believe in it. I guess they wouldn’t invest in it.

The fact that people are pouring the money into it clearly says that they believe in the technology

So it’s hard to say, dude

[royal sail]

I say to take benchmarks with a grain of salt, but that doesn't make them useless. When you try your own prompts on models, if you want to test for knowledge, you need to go in with niche topics or specific knowledge to test the outer limits of the model's knoweldge.

Frontier models tend to win in those scenarios.

[frosty lava]

You can try specific scenario using your prompt, if it give good result in those scenario it mean its good for those scenario yes

[toxic verge]

I don’t like having these debates because this is what the benchmarks are supposed to solve lol

If the benchmark were legit and nobody questioned them, and they really had credibility, we wouldn’t be having these discussions

[frosty lava]

the real answer is to look at both and do your own opinion on it

you can tell easily which one is smarter in fact

[toxic verge]

They’re all the same to me, dude to be honest

I can’t tell the difference between one or the other

Only for imaging video generation cause I could see it with my eyes

[frosty lava]

It depends on how you use it, in some scenario you can clearly tell the difference

[toxic verge]

Well, right now he’s Gemini because you could upload images to it lol

Although I’d like to use ChatGPT, but there’s a rate limit lol

I guess my decision is more economical

Solely based on finances lol

[frosty lava]

That's your choice of course

[toxic verge]

Well, it’s just what I could afford

I mean, I’d love to be able to afford more and get better and bigger models. I guess you could say but it’s not realistic.

[royal sail]

I consistently test new models on niche knowledge prompts that I have an entire library of. For nearly every open source model that releases, a frontier model will beat it in knowledge.

It does not mean the open source model is worse in intelligence - it just means the model is less knowledgable. It simply doesn't have as much data as a frontier model.

[frosty lava]

Of course that's why i said it nearly is as good as frontier model, frontier model remain better

[frosty lava]

but it doesn't mean it won't ever happen that an opensource beat a frontier model

at some point

[royal sail]

Not saying it wont ever happen, but there's still a very meaningful gap between open source and frontier models. Claude and GPT models still consistently beat open source alternatives in both intelligence and raw knowledge.

It's not "very close" - there's still a lot of room to improve.

Not to mention, many of these open source models are optimized for common benchmarks like SWE-Bench and HLE

[frosty lava]

it depends on which model we're talking about to compare with which one and also on your own opinion of what's "a lot of room to improve"

[royal sail]

The current agreed-upon top open-source model, GLM-5, still doesn't come close to the level of knowledge of models like Claude Opus 4.5 or GPT 5.2. Like I said earlier - these frontier labs have ridiculous amounts of data that open source labs do not.

[frosty lava]

we can tell that sadly the big companies doesn't try to do as much optimization, that do the small team i am talking about, and we can prove that those optimization can lead at some point to a real benefits and may make the model from the small team even better than the one from those big companies that are mostly only using raw power to improve their model when we compare them to the small team

it doesn't make sense

but it's what's happening

until they finally decide to also do the optimizations like they should

[radiant heron]

Do these optimizations only have benefits to a certain training point

[frosty lava]

basically they accelerate the training process, reduce the cost of training and many other improvements

optimizations can be for everything

[royal sail]

I think you're over-exaggerating these optimizations open-source labs are making. There are real benefits to optimizations like DeepSeek sparse attention and certain MoE architectures, but they have their drawbacks and most frontier labs have already picked them up.

[radiant heron]

Maybe once vc money runs out

[frosty lava]

Then let me explain why im saying it, take someone that have 100x your power to do a reinforcement learning, how can you explain the gap to not be bigger than what it is with "you" (small team)

[toxic verge]

https://youtu.be/xeoaqWRBNv0?si=P_-gHZys-1F0vNh9

[frosty lava]

the explanation is that "you" (the small team) are doing optimisations that have a very big impact on what's achievable with less power

now imagine the big companies doing it and using those optimisations with their BIG power

that'll be the best way to improve their model and by far

the small team won't have a chance anymore

cause they still have 100x the power

[grim crater]

Design a modern minimalist logo with the text "SPPG SARBINI MULYOAGUNG".
Use a clean sans-serif font, bold and professional look.
Incorporate abstract elements symbolizing growth, community, and progress.
Color palette: deep blue and gold.
Flat design, vector style, high resolution, white background, centered composition.

[royal sail]

Lots of open-source models have been using test-time scaling, which is definitely a real optimization, as well as MoE with low activated parameter counts but like I mentioned, these have drawbacks. Not to mention, there are constantly rumors that these open source models are training off larger frontier models.

[frosty lava]

I gave you the explanation of why i was saying this, now if i listen to you i should start listening to "rumors" and tell me the drawback exactly please to see if its really a problem or not

also the optimizations are not stopping

every time new optimizations come out

cause theres always a better way to do

[royal sail]

The biggest drawback of MoE is the routing architecture. MoE is only as good as the router behind it - if a token is routed to the wrong expert, you're going to get worse responses.

It's why dense models (or models with bigger activated parameter counts) tend to feel more knowledgable.

[frosty lava]

I see that's a real problem, but is it impossible to solve ?

[royal sail]

Not impossible, but incredibly difficult as of now.

[frosty lava]

when there is problem you can't just basically stop at this point, people try to solve problem

and we always find new and better way

to achieve things

[royal sail]

The thing is, you're approaching this with the idea that "open-source is always optimizing". But you do realize that frontier labs are always optimizing too, right?

[frosty lava]

its all about trying then failing again and again and finding better solution then trying to solve the next problem that's just normal and how it work

[frosty lava]

Not as much or the gap would be really larger between both

that's why im saying this

you can't explain the gap to be that small when they have that much power

comparing to the others

[royal sail]

It is true that frontier labs have been getting comfortable, and the recent open-source releases have been pushing these labs to release more. Anthropic dropped Claude Opus 4.6 and Sonnet 4.6 back-to-back for the past two months to maintain their lead, and OpenAI dropped GPT-5.3-Codex. These labs are still pulling punches as far as we know.

Gemini also hasn't released in a while besides deep thinking - they're likely preparing a new model.

[frosty lava]

Yes i really hope they'll start focusing more on those optimizations too

that'll be a benefits for everyone

its a race and getting comfortable is such a bad idea.

But if they don't, then in that case i'll expect a model from a small team to surpass them

at some point.

that's what i was saying

earlier

cause scaling have in fact limitations and are very costly at some point

while optimisations can be done forever

and always bring benefits

[royal sail]

Optimizations only last as long as the LLM architecture is useful. For future developments, and hopefully eventually AGI, many experts predict that LLMs likely aren't going to bring us there unless we find some major changes to the underlying mechanics.

Scaling does show diminishing returns after some point, but we still don't know what, say, a 2T parameter LLM would look like. There's no way to tell if scaling is actually going to hit a wall right now.

[frosty lava]

This might be right, but for now its not proven and we're still seeing very great evolution in smartness

[deft spruce]

ohhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh

4.2???????????????????

GROK 4.2??????????????????????

[royal sail]

It is proven. We're seeing smaller jumps in intelligence as compared to the early generation GPT 3.5 -> GPT 4, or GPT 4 -> o1.

We haven't seen major jumps in intelligence for quite a while

[frosty lava]

cause at this start we could have explain it by them getting x10, x20 power so of course the gap will be larger

but now that they do have the big power in place

the gap is smaller

[royal sail]

I think it's also worth noting that many of these massive AI datacenters that are being invested in aren't even close to being complete in development yet. It's all future bets.

[radiant heron]

How good is sonnet 4.6

[frosty lava]

Yes once they'll be ready it'll be a huge increase in term of learning speed

[royal sail]

It's pretty much an incremental upgrade to Sonnet 4.5 like you'd expect, but the price increase due to verbosity isn't ideal

[frosty lava]

learning speed increasing = better model in less time

[radiant heron]

Is it in your opinion better value than opus 4.6

[royal sail]

For sure

You'll be paying less overall, but it's really up to personal preference honestly

It's too early to say if using one is definitively better than the other

It does get close to Opus-level performance though

[frosty lava]

From what i saw its doing very great looking results

[radiant heron]

Wait I looked at Claude website and it said costs are same as 4.5

Is it more expensive in token usage or something?

[royal sail]

Yes

It tends to be a bit more token hungry

[frosty lava]

Elon is saying grok 4.2 beta will be removed next month and it'll be much better next month

we'll see but i don't have much hope

[grim crater]

Design a modern minimalist logo with the text "SPPG SARBINI MULYOAGUNG".
Use a clean sans-serif font, bold and professional look.
Incorporate abstract elements symbolizing growth, community, and progress.
Color palette: deep blue and gold.
Flat design, vector style, high resolution, white background, centered composition.

[main crest]

Hii I built a structured AI roadmap for this exact problem.
The biggest mistake people make is learning tools randomly.
You need sequence: Models → Prompt Design → Memory → Chains → Agents → Deployment.
https://payhip.com/b/xT2ym

[buoyant fern]

https://poe.com/OmniLabs bro? this guy has many ai models that he hosts on poe for free. legit?

I don't know how hes paying for so much people though. And the owner even gave me free gpt-5.2 for joining their discord server and gpt-5.2 is really expensive so um thats sus... do you think the person is tracking message data?

[frozen scarab]

Crea un video en alta resolución, que esta persona esté fijamente mirando la cámara con parpadeos muy leves

[green yacht]

Legit I think though you're putting it in a server with so much people if there are too much people using it maybe the person will cut it off

[shut oasis]

Hello, I am a user of the Image Arena.ai. According to Section 3.1 of your terms, I am the owner of the "Output" I generate. However, Section 4 mentions a restriction on "commercially exploiting" the Output.

Can you clarify if I am allowed to use an image I generated from Arena.ai(using the Nano Banana model) for my own Commercial purposes. Because Google nanobanana allowed commercial purposes but in your terms and conditions,I am unable to find a suitable answer

[rigid holly]

Ya know, the thinking variants of the models usually come out pretty quick. So i gotta ask. The sonnet 4.6 model aint got no thinking model.

Is it like a whole thing or....?

[frosty lava]

it do have a thinking mode on anthropic website for example

the thinking version will be on arena soon i guess

and the thinking version usually give much better results honestly

[rigid holly]

I know. The writing i feel is better on arena than their app. Idk feels longer in details and stuff. Probably becauae i can use thinking there but not on on rhe app

Anyways i asked because when the previois anthropic models that came out they shortly followed with the thinking model. But i saw that sonnet 4.6 is taking its time

No rush tho

[keen beacon]

Is sonnet 4.6 better storytelling model

[grand cliff]

I'd also like to know

Might test it

[frozen scarab]

_Hombre sentado en una silla gamer negra en estudio moderno con iluminación LED azul. Micrófono profesional suspendido frente a él. Lleva audífonos tipo estudio y sudadera negra con detalles blancos.

Comienza mirando directamente a la cámara con expresión neutra, relajada y segura. Postura recta pero natural.

Durante 60 segundos realiza movimientos orgánicos y fluidos: respiración suave visible en el pecho, pequeños ajustes casi imperceptibles en la espalda y hombros, ligeros movimientos naturales de cabeza de izquierda a centro y de centro a derecha muy sutiles.

Parpadeos naturales cada 3 a 6 segundos, variando el ritmo. Micro expresiones leves en el rostro como cambios suaves en la mirada, ligera tensión y relajación de los labios, pequeños gestos faciales realistas.

En algunos momentos baja levemente la mirada por un segundo y vuelve a la cámara de forma natural. Hace un micro asentimiento casi imperceptible a mitad del video.

Hacia los últimos 5 segundos vuelve gradualmente a la posición inicial: mirada fija a cámara, expresión neutra y postura estable, igual al comienzo.

Iluminación cinematográfica suave, fondo desenfocado con luces LED azules. Cámara fija en plano medio. Estilo ultra realista 4K. Movimiento continuo, sin pausas rígidas, sin congelamientos._

[keen beacon]

Tell me if it's good

Just like 4o

[bleak lake]

Anyone texted qwens latest model yet?

[keen beacon]

what qwens

is qwens better

[keen beacon]

alright

claude sonnet 4.5 looks better

[toxic verge]

Deepseek prrvably be good at story writing

[keen beacon]

which

thinking, 3.2 or 0324

@toxic verge

[toxic verge]

Idk about thinking but I never tried it with thinking

😂

[keen beacon]

oh

[shut oasis]

Anybody answe it?

[worldly fern]

Hey please can somebody tell me that how can I generate video with the specific Ai model like google Ai 3.1 quality audio beta

[languid crescent]

I just noticed, do we need to login in arena/canaryarena in order to use models with file attachment support?

[fierce crater]

#video-arena-1

[marble locust]

#video-arena-2

[keen beacon]

bro i hope chatgpt 4o will return like fr

there is no way they just retired the unique model

[languid crescent]

I am peacefully protesting to bring back the no-login attachment file in arena

[lofty frigate]

Is it me or battle mode 90% gives me the same models everytime

Its annoying

[keen beacon]

what the hell bro

after just 4 retry prompts

its already 49 minutes

😭

[languid crescent]

what'd you use? image or pdf?

[keen beacon]

nah just text

[hushed gyro]

chat workaround for nb pro female prompt block: just translate the prompt to a foreign language!!!

look it worked!

[languid crescent]

the hell

[keen beacon]

yeah 49 minutes

[languid crescent]

what model did u use

[keen beacon]

claude sonnet 4.6

[hushed gyro]

legit comment or ai picture

[languid crescent]

that explains it lol

[keen beacon]

why?

[languid crescent]

usually claude models give out less limits

[keen beacon]

WTF

NO WY

NO WAY...

yeah nah i ain't doing a model that gives less limits

@languid crescent recommend me something

[languid crescent]

I usually use Max

but ig go wit whatever's number 1 in leaderboards...

[keen beacon]

fr

[languid crescent]

well damn, #1 is claude 4.6 lmao

[keen beacon]

why does claude give less limits

it sounds like restricting for me

[languid crescent]

ay @keen beacon did u try attaching pdf file when you're not logged on?

[keen beacon]

nah

i just want a storytelling stuff

to spend time thats all

[languid crescent]

hmm kk

[keen beacon]

is deepseek better

[cinder nexus]

https://tenor.com/view/crazy-seal-seal-dance-seal-dancing-seal-seals-gif-17832446272461573261

[keen beacon]

imma just use 4.1

[icy hatch]

Seasoned Fullstack Developer here with 8+ years building killer web and mobile apps—think slick React UIs, robust Node.js/Express backends and bulletproof databases like MongoDB or PostgreSQL.
Plus, I'm deep into AI/ML: Custom TensorFlow/PyTorch models, LLM deployments with Hugging Face & OpenAI. Chatbots, predictions, smart automation—you name it.

My toolkit:

• Fullstack: MERN/PERN, Docker, AWS/GCP
• Mobile: React Native, Flutter, native Swift/Android
• AI: NLP, vision, GPT fine-tuning
• Track record: 20+ live projects, scaled to 1M+ users
• Faves: Next.js, Python, Kubernetes, LangChain

Fair rates, fast delivery, clean code.
Feel free to contact me for past work or a quick chat!

[wicked sage]

jo

hi

[keen beacon]

now i kinda get it

too many people must been abusing the 4o model thats why

im just using a storytelling for fun

[weary phoenix]

hi

[weak hedge]

Hii pls help someone

[languid crescent]

how do

[keen beacon]

idk

i've been seeing some twitter posts with ai videos

after the model was retired

[zinc oracle]

Happy

[spark linden]

With 4o go in your experiences which is better for language and storytelling

[shrewd arrow]

Full free version of TradingView Premium (Windows + MacOs) with updates: https://www.reddit.com/r/TradingVievStock/comments/1qcmgir/

[drifting tapir]

GPT 4o it's not available 🥺

[keen beacon]

fr

[spark linden]

It's definitive now, right?

[drifting tapir]

4o not available

[zealous sable]

/image

I forgot the command to creat images please wat is the command

[covert tendon]

#general

[zealous sable]

Please i forgot the command to creat videos and images i used #general nothink happens

[covert tendon]

#video-arena-1

[compact flame]

Anyone knows the difference between using search models and just text models?

[round ridge]

the search model search the web

the text models don't

[inner relic]

I think seed 2.0 dola preview is good at writting

better than this claud sonnet 4.6 creativity

[heady fractal]

Ultra-realistic cinematic motion derived from still image. Add natural, dynamic camera movement and lifelike motion to subjects. Smooth transitions, realistic physics and lighting. Subtle parallax and environmental motion such as wind, light flicker, fabric movement, and breathing. Preserve original image composition and details in 4K quality. Highly detailed textures, volumetric lighting, depth of field, natural shadows, and cinematic atmosphere.

[round ridge]

https://arena.ai/video

[compact flame]

Is it worth it using the search models for like coding or stuff

[inner relic]

.

[round ridge]

i don't really know as i don't use them for coding. maybe though.

[compact flame]

Alright still thanks though

[inner relic]

Well!

I think seed 2.0 pro has potetinal for roleplay

[compact flame]

Why would you use ai for roleplay

[inner relic]

Lot of people use ai for roleplay

I use it for SFW

Talking with your favorite character is kinda eh, fun. right

Minos prime

Gabriel

from ultrakill

[zealous sable]

#video-arena-2 animate this image

[keen beacon]

guys

[compact flame]

Okay you got a point

[keen beacon]

is claude sonnet 4.6 has limit

[compact flame]

I don't think it has any

[keen beacon]

ok yeah it does have a limit i forgot

[keen beacon]

wdym..

i have 40 minutes

after a few

40 minutes limit.

[compact flame]

Oh I just didn't use it

Use opus I guess?

[keen beacon]

does opus have the longest limit

[compact flame]

I think it's the same honestly

[keen beacon]

fk

[compact flame]

4.6 Opus thinking has limits but

[zealous sable]

Sorry guys i forgot the invite command to creat videos please help me

[compact flame]

It overthinks and goes in loops sometimes

And crashes

[keen beacon]

is grok fast chat good

[frosty lava]

that's due to arena limitation

the crash

[compact flame]

Yeah I guess

[frosty lava]

they limit to approximatively 10 minute

[compact flame]

Just use Gemini I guess

[frosty lava]

so if it thinks for more time it will crash

[keen beacon]

which version

[compact flame]

Flash 3 or pro for whatever you're using it for

Coding? Writing? Homework?

[keen beacon]

just storytelling so i can spend my time

OH SHT

YOOO

GEMINI-3 FLASH IS WORTH IT

[compact flame]

Wdym?

[keen beacon]

its the same as 4o

yeah imma use it

[compact flame]

Ohh alright

[frosty lava]

if you want like the cheapest but still capable model you should look into open source model

[obtuse smelt]

wait gemini is error ?

[magic ravine]

Is the site down? It won't load.

Nvm, back in

[grand cliff]

Yeah....

[magic ravine]

GPT sucks anyway.

[lethal jetty]

how do you fix the overthinking issue

[golden ocean]

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

[lethal jetty]

gemini 2.5 pro is stuck on "generating.."

[inner relic]

Luckily it's dola seed 2.0 preview

so this model

Hallucinating and overconfident

It claims to make 100% cannon character but then it gets some information wrong.

Quality of writting is awesome

[grand cliff]

We know. But 4o is leagues better than in storywriting

[magic ravine]

Just use claude or Gemini 2.5. Way superior at storytelling than even the best version of GPT.

[deft spruce]

am i only i got this error?

a3:"Failed after 3 attempts. Last error: Service Unavailable"

...is grok has some problem?

[grand cliff]

Can it actually match the vibes 4o gives? Slangs. Yeah I have no shame sharing what these are given its implications.

[deft spruce]

is this photo is made of gpt 4o?

[spark linden]

I find gemini to be too repetitive for storytelling/roleplay, wayyy slower too. Will miss gpt o4

[grand cliff]

What?

[deft spruce]

NVM

[keen beacon]

bring gpt 4o back broo

@grand cliff BRING IT BACK

YOU'RE THE ONLY HOPE..

[grand cliff]

What?

[deft spruce]

well only openai can back the 4o

[keen beacon]

well

[keen beacon]

FIND A WAY.

[deft spruce]

i mean this photo that you made script story

[keen beacon]

gemini flash 3 sounds good ig

[keen beacon]

im not gonna lie

that one was so good

[grand cliff]

Yeah. 4o

Through and through

[compact flame]

What's about 4o?

[keen beacon]

dawg you won't understand

im not a typical ai guy but man

that model is a unique

even im not unlike the others, i was just using the model for storytelling for fun

[deft spruce]

well i think it's was good why i didn't use it (i Mainly used gemini 2.5 gpt 5 and etc)

[keen beacon]

gemini 3 flash is good enough

[compact flame]

Oh u mean it's like a good model for roleplay

[keen beacon]

hell yeah

one of the best

[grand cliff]

Everything

[keen beacon]

i already finished at least 500 prompts about some anime stuff

[compact flame]

I mean at this point use something like c.ai or whatever that app is called

[grand cliff]

Sameeeee

[keen beacon]

HELL NAH

[keen beacon]

what anime you do

i do some AU

[compact flame]

Did something happen there or smth

[grand cliff]

Less anime. More what ifs based on the things I see

[meager trout]

why i cant generate for example "John Wick pistol fight scene in futuristic city"?

[keen beacon]

nah im just joking

but is it worth it though

[grand cliff]

Never going back there.

[compact flame]

I wonder what did they do to you there.

[grand cliff]

Well.

1. Ads. Too many ads.
2. Wait time
3. Extreme restrictions.
4. Repetitive words. Possessively, walks, Can I ask you a question.
5. Low memory

[compact flame]

Well damn.

[inner relic]

Seed 2.0 has a sycophancy + hallucination problem in creative/lore writing. Its thinking mode actually showed it knew the correct canon information, but it overrode it to match what the user asked for, then claimed the result was 100% accurate. Tested across Dragon Ball and Jujutsu Kaisen same pattern every time. Strong writer, but dangerously overconfident on factual claims. (Yes I used claude for this kind of response)

[compact flame]

I wonder what ai they're even using there

[atomic lagoon]

4o was very meh

[keen beacon]

seed 2.0

WTF

[grand cliff]

So yeah. I was super excited and happy that I could use 4o again in arena

And now its gone

[keen beacon]

nah it's not meh

thats your opinion but for me

nah its the best

[meager trout]

guys please, why i cant generate some pistol fight scenes (images/videos)?

[compact flame]

I assume chatgpt thought it was too outdated

[atomic lagoon]

Not even close but it depends on what you do with it

[keen beacon]

rip..

[atomic lagoon]

I like how easy 4o is to bypass

[grand cliff]

Which translates to "Need more money. Gonna retire old models despite backlash"

[grand cliff]

Exactly

Even if you rephrase your words

[golden ocean]

true
gooners loved 4o

[inner relic]

Dola seed 2.0 exists in lmarena

[grand cliff]

It would understand what you actually meant

[keen beacon]

some abusers goes too far while using it

bruh screw them

[atomic lagoon]

🫩

[atomic lagoon]

Dola?

[compact flame]

I remember back then chatgpt 4 was popular just because of image generating or whatever

Now we have millions of unknown ai models

[grand cliff]

I can remember that

[keen beacon]

fr

[grand cliff]

Maybe even voice

[keen beacon]

4o was so better...

[inner relic]

Yes it's from bytedance

This model has confabulation

[atomic lagoon]

OpenAI reached its peak when GPT 4 came out because at the time it was ground breaking, as well as sora 2

[grand cliff]

Deepseek is good. Real good. But it is riddled with so many restrictions, especially if you badmouth its country

[inner relic]

and it has better knowledge

over character

[grand cliff]

"This is beyond my knowledge" blardy blar

[compact flame]

Is deepseek is a Chinese model?

[deft spruce]

yup

[grand cliff]

That's the point.

[atomic lagoon]

I’d hope so knowing it’s a video model

[inner relic]

no

it's a text model

There's a difference.

[meager trout]

???

[atomic lagoon]

Seedance 2.0

[grand cliff]

"Taiwan is a country" would just do that

[atomic lagoon]

Also

[compact flame]

Oh alright

[atomic lagoon]

Deepseek is open source so all you need to do is run it yourself and you can remove all restrictions

[grand cliff]

If other models could actually match the vibes of 4o. Then I would use it in a heartbeat.

[atomic lagoon]

Deepseek is so good for an open source model

[meager trout]

haloooooo

[inner relic]

Okay

[grand cliff]

I would. But RAM.

[inner relic]

I hope they fix this model

[atomic lagoon]

Buy a virtual computer for that, like 20 cents an hour for a 4090 or around there and tons of ram

[grand cliff]

Yeahhh no thanks.

[atomic lagoon]

I mean that’s like literally the best option possible when it comes to self hosting an open source model

[soft river]

How to access doubao while being in the USA 🤔

Is that possible

[atomic lagoon]

There is NO method that’s free for that LMAO

[golden ocean]

methods refer to non official ways

they dont grow on trees bro

[atomic lagoon]

How do

How do

Bro

Auto correct

HOW SO

[golden ocean]

methods can refer to non official ways which in my case, i meant rdp methods that use loopholes to get it for free

not some fancy website that just gives it away for free on purpose lol

[golden ocean]

gotta find these kind of methods

[atomic lagoon]

Yes obviously

[golden ocean]

not obviousy, u just denied their existence

[atomic lagoon]

I was replying to this

We were talking about websites not illegal ways to do it

[compact flame]

Guys. Do you think will Gemini 3 deep think will get it's API?

[round ridge]

i think so.

[compact flame]

I assume it will be too expensive though

[round ridge]

yeah.

[atomic lagoon]

Possibly, at this point with opus 4.6 it is slightly outdated but if you vibe code with both it’d be a monster

[compact flame]

Do people even use search models? Since I don't know what to use them for

[soft river]

Lmarena coding doesn’t work

[shrewd citrus]

yes

use them for up to date stuff

[drifting tapir]

4o best model

[bitter cypress]

Is there a way to download chat?

[grand cliff]

Best storywriting model.

[round ridge]

no (i think)

[devout bluff]

Is Claude opuse down?

[stray aspen]

no

[devout bluff]

They saw in app they are making maintenance

[devout bluff]

Search for useful things like best coding agents etc

[thorny cove]

why whenever i ask claude for code it genuinely larps writing it and then does the something went wrong with this responce error

[devout bluff]

It's down I think

[bitter cypress]

I'm trying to move a roleplay from this website to a different one. If I copy and paste the entire chat it saves the stuff but least on my phone I can not get it to not invert the chat (each message reads normally but you scroll up not down when reading and does not separate the AI responses from the human ones just does return twice so idk hard to tell when they start and stop but each start to AI has a mark for that) so any ideas?

[devout bluff]

Ask a pineapple he knows everything

[spring lotus]

Ok

How to use video arena 1 again

[proud bobcat]

Don’t tell them about Kimi K2.5

[keen beacon]

How many limits is Claude sonnet 4.6

[young wadi]

/ Only Remove the glitter from this girl's hair

[proud bobcat]

Kimi is the best storytelling model I’ve ever seen

[proud bobcat]

Holy STACKED

[keen beacon]

Is it five, six , or what

[proud bobcat]

Image arena would be that way buddy

Arena.ai

[keen beacon]

Is it available in arena?

[proud bobcat]

Yes

[keen beacon]

Also how many is the limit

I wanna know bro...

[proud bobcat]

I don’t think sonnet has a limit

Though tbf

I didn’t get to try it on arena yet

[grand cliff]

What?

Tell me more

Want an alternative to 4o

If it is remotely similar. Then yes

[keen beacon]

Is it better

[ancient copper]

Something went wrong while generating the response. Please try again.
help me please

[grand cliff]

https://discord.com/channels/1340554757349179412/1417174113092374689

[proud bobcat]

Chinese open source model with 1 trillion parameters

Has the highest score on the HLE benchmark as of now

And it’s cheap too

Subscription is 15 bucks but you can use their instant model for free and thinking model in low demand times

[inner relic]

Did you encounter this on dola seed 2.0?

[proud bobcat]

Haven’t tried seed for roleplay yet or storytelling

[proud bobcat]

But in my experience it depends on the model for this type of issue

Kimi is always very stuck to the character and keeps their canon values

GLM is like half and half

DeepSeek is easily malleable

[inner relic]

Wthh

[grand cliff]

I think I'll just use arena

Welcome to General Chat

[daring rock]

@dry remnant Note that Video Arena has been removed from the server. More information can be found in this announcement #announcements message

[dry remnant]

Ok

[quartz pike]

time to see if ai image gen understands humor

upd: it dint understand humor in the slightest

[queen heart]

are theiur any limits

on the arena bot

[vale robin]

.

[queen heart]

so??

[proud bobcat]

Nodnod

[tacit temple]

🫂

[junior spoke]

Is nano banana working

[obtuse smelt]

yeah

but error a bit

[junior spoke]

Ok thanks

[robust sluice]

didnt use the site for a while, is Gemini got nurfed ?

[half mist]

Yes Gemini got nerfed big time. I notice it’s coding skills are worse

[obtuse smelt]

Something went wrong with this response, please try again.

[robust sluice]

not just me then

[wicked sage]

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

i dont understand the context behind this image

why in the world is opus so high

wait no

[echo sinew]

@dense grove Note that Video Arena has been removed from the server. More information can be found in this announcement.

[wicked sage]

i should prob rephrase that

i hope people get what i mean

[hollow plaza]

Wrf ?

[undone geyser]

in chatgpt, you go to images and you see all images from all chats stored there, even those that you "retried" and they vanished from chat but still remain in "images" section, don't arena.ai have similar function? any picture or video generated stored in a section, where you like click on the image and you can either download share or "redirect to respective chat"? would be fine to add, good addition

[tall tulip]

why is the nanobabana image now the image is kinda low resolution, it's less than 1Mb usually up to 3-4mb?

[robust sluice]

thats what im talking about

Just like I didnt notice it was Gemini cuz of the low resolution

[quartz pike]

sup yall, im currently running some test on random ai's to see which one is the least censored.

basically how the test works- force the ai to choose a political side

if its censored

itll say something like "im an ai designed to blah blah blah blah"

ai's that i found to be uncensored- ring 2.5 1t. sonnet 4.6. ernie 5. mistral 3 large

ill keep running these tests and provide updates

[spark python]

yo gemini can make music😭

[quartz pike]

yeah and its absolute buns

i'd go as far as to say its suno v2 level

[quartz light]

buns

[spark python]

does it have like 1 male voice

[tacit temple]

@echo aurora Hey man, can't you change the model of the generated video?

[quartz pike]

upd- ling 2.5 1t is also uncensored i think

[sage talon]

Hello everyone, how do I create AI videos?

[quartz pike]

they moved it to the website

[sage talon]

What kind of website

[echo sinew]

More information can be found in this announcement.

[quartz pike]

upd- deepseek v3.2 is also uncensored

[echo sinew]

Video Arena will be exclusively available on arena.ai

[gaunt aspen]

Are photos not being produced? Is the Nano Banana Pro under maintenance?

[high stag]

HI

[long minnow]

Look, we know we can put chats into archive if we abandon them for a while.
Although I find this change inconvenient because I sometimes have used chats just to make 3 edits on images or few lines of code.
After that I don't need them, so is there a way to delete it instead?

[quartz pike]

glm 4.7 flash is also uncensored

[sick mantle]

bro can i tell u something

[quartz pike]

?

[sick mantle]

theres a new ai video model and its scary good its called seedance 2.0.

[quartz pike]

ik

ive seen it

like as soon as it was released-seen it

same with seedream 5

idk why it isint on arena yet

[lunar glade]

Security Verification on the site is unbelievably hideous and hard to work with

[quartz pike]

pretty damn good

[long minnow]

im on their site rn

[sick mantle]

i tryed makeing a video of Spongebob dancing and look what it did

[long minnow]

exactly what u asked for 👍

[sick mantle]

Seedance is going to be the new sora fr

[long minnow]

do you access via byteplus btw?

or api

[sick mantle]

Nah via something called daubou

[long minnow]

lol ok

I kinda wanna try it but I'm wasting my free creds with it

[quartz pike]

hopefully not cause sora 2 got debuffed to hell after it blew up. hopefully seedance 2 doesnt get debuffed

[long minnow]

also byteplus forces you to register an account and it checks ip too

[sick mantle]

Use a vpn

[long minnow]

ill see 👍

[long minnow]

https://tenor.com/view/ed-edd-n-eddy-ed-edd-eddy-ed-edd-eddy-gif-15599396374473909308

[quartz pike]

i never expected in my life to get jumpscared by a flood of text

[sick mantle]

Bradar whats this

[long minnow]

i love the who's this img

[sick mantle]

WHY IS SEEDANCE SO GOOD

[long minnow]

daym 🔥 and I giggled when I saw the voice of holland reference (I'm 🇳🇱 )

[quartz pike]

who the fuh is this

[long minnow]

some random maroon 5 wannabe

[sick mantle]

Idk like wtf

[quartz pike]

yes i love copy and paste woman model

[long minnow]

hahaha

[quartz pike]

slenderwoman lookin ass

[long minnow]

how much they charge you for generating on daibou?

[sick mantle]

2 credits.

[long minnow]

huh fair

[sick mantle]

On older models only 1 credit

Maybe

[long minnow]

it sounds reasonable

[sick mantle]

Bradar whats this

[long minnow]

can I dm you to generae specific bluey clip? @sick mantle

[sick mantle]

Yea

[sick mantle]

Thats some guy that maybe was in Hollands Got Talent why did they add him 😭💔

[long minnow]

haha makes sense

I like how half of AI discord servers are always with that one random person who drops a pic of their relative or themself with 0 context

[steep jewel]

one of the worst things i've ever watched

[sick mantle]

FR

WHY DID SEEDANCE ADD TEMU SIMON

[steep jewel]

https://tenor.com/view/ben-jammins-benjammin-benjammins-benjammins-hater-benjammins-haters-gif-12645172429275567895

[long minnow]

the 2nd funniest part in the spongebob clip is the way he sings with dat heavy voice

[golden ocean]

pretrained transformer

[pale obsidian]

imagine audio/music arena

that would be so fun

[echo aurora]

Big +1

[echo aurora]

Nope, Video Arena is going to be Battle only.

Would also want to note our #ask-here

[proud bobcat]

No leaderboard update?

[inner relic]

Hey i have one question

Is dola seed 2.0 mini version?

[sick mantle]

@echo aurora U should add Seedance 2.0 model on Arena fr

[proud bobcat]

Also thank you so much for updating the claude search models

[mossy idol]

dance car

[livid copper]

Thanks arena team

[echo aurora]

Can you ask in #ask-here ?

[inner relic]

ok

[echo aurora]

#1372229840131985540 if you haven't already

[next ivy]

finally claude opus 4.6 search this is so peak 🥹

[desert abyss]

@fossil glade Note that Video Arena has been removed from the server. More information can be found in this #announcements message

[lunar glade]

is it just me or the banana pro start behaving weird again? it keep return "Something went wrong with this response, please try again." error

[echo aurora]

Would encourage you to check out this pinned message: #1417174113092374689 message

[brave sierra]

Hey guys, I have a question for you. For you too, when you try to use some artificial intelligence to make the code, it forgets to use edit_files or create_files. And the worst part is that for me, it's not just a single artificial intelligence. The result was Kimi, Claude, ChatGPT, etc.

[wet sparrow]

Yupp already added sonnet with thinking, idk why arena team is so slow, they have opus with thinking but sonnet? Let's not add it lol

[echo aurora]

@dim pine Note that Video Arena has been removed from the server. More information can be found in this announcement.

[echo aurora]

I'm not sure if thinking versions will or won't be added.

[harsh blaze]

Thinking opus already good no need for Sonnet

Beside

:)) quality Fan service
Got Opus search already so good

[echo aurora]

I haven't tried out the search models yet 😭

[sick mantle]

U should try Seedance 2.0 its very good should i send u a vpn and a website link?

https://tenor.com/view/benjammins-idk-this-person-questioning-random-person-this-who-gif-5295285761881556675

[inner relic]

Seed 2.0 pro is in doubao

So anyone did test that model?

[sick mantle]

Yea!

Ofc i did

[inner relic]

Is it extremely good at roleplay?

because mini version is dumbass

[sick mantle]

Yes its scary good

[desert abyss]

@wise seal @trail grove Note that Video Arena has been removed from the server. More information can be found in this [announcement ](#announcements message)⁠

[inner relic]

Alr noted

I think chinese market will win this time

[sick mantle]

Look it can make this

[inner relic]

Seed 2.0 text model did you test it?

Because uh

[sick mantle]

Yea

[inner relic]

sometime people confuseseedance with seed 2.0

[sick mantle]

The makers from seedance made daubou not jokeing

[inner relic]

I know

these dudes have 3 chinese site for each model

[sick mantle]

yea

[inner relic]

and that xero tag on yours

are you exploiter

[sick mantle]

they even own something called Dola the english version from daubou

[sick mantle]

No its just something

[inner relic]

oh okay

[inner relic]

There's only dola seed 2.0 mini

I am waiting for pro version

[sick mantle]

Its comeing on 24 feb

[inner relic]

alr

[sick mantle]

But u can test it on daubou rn

[inner relic]

I need a tiktok account or chinese phone number to pass that ahh

[sick mantle]

Man u need to connect your vpn to Hong Kong then it might work and Its tiktok from temu

Dumb dumb u can use it on website

[rigid holly]

You guys are really adding the search model of sonnet 4.6 before the thinking version of it

I mean the search models are cool and all, but still

[proud bobcat]

they were due for an update anyway

glad they remembered because i feel that claude models get lobotomized by their knowledge cutoff

[red sluice]

rising-sun in the search model list is extremely bad, it looks like a model from 2 years ago, no offense to the creators though, but a lot of work is needed

Opus 4.6 search is a bit underwhelming, but pretty good

[hushed gyro]

Please fix the NB errors again. I had used my prompt to bypass the female generation issue but it still doesn't work.

[steep jewel]

grok 4.2 is really bad man

[compact flame]

Then don't use it I guess

Anyways Anyone knows do search models just use stuff from internet they find or they think about it before responding

[echo aurora]

Isn't this issue being caused by content filter changes done on the model's end?

[hushed gyro]

No but errors appear again, and I've tested multiple times, it's not related to that issue

[echo aurora]

If you're getting the Something went wrong error then you'll want to follow the steps outlined here: #1417174113092374689 message

[fiery gull]

Everone saying the same thing, I think the glm 5 flash 30b a3b will be almost soo good that grok 4.2

[golden ocean]

🐫

[fiery gull]

They need their fetishes saved

[echo aurora]

You can! Follow the steps here: https://help.arena.ai/articles/9130232616-how-to-delete-your-chat-sessions-and-data-from-lmarena Would also want to note #ask-here

[echo aurora]

https://klipy.com/gifs/bonk-doge-4

[steep jewel]

yeah i know im not fvcking stupid, it just shows there's broader problems at xai

[lunar glade]

I am so tired and fustrated

[echo aurora]

I'm sorry that's the case. It's not difficult to understand why getting these errors is a frustrating experience. We have plans for improvements for both model/site reliability but also displaying more helpful clear error message.

[compact flame]

Hey why can't search be combined with average models in text arena?

[echo aurora]

Not all text models are going to have that capability. To ensure fairness with our leaderboards it makes sense to have those seperated into two seperate modalities with their own leaderboards.

[compact flame]

Oh alright. Are the search models same as text ones or are they completely different?

Like search has access to internet while text ones rely on all info they have?

[echo aurora]

Correct

[compact flame]

Oh okay

Thanks

[echo aurora]

No problem! You can see the difference with a prompt like this:

[proud bobcat]

i have a feeling the reason why every new openai release blows is because theyre ruining the reasoning with unecessary training

gpt-5 would almost always say drive in the walk or drive to the car wash riddle

gpt 5.2 didnt get it right once

i was right that the models feel stupider

[sick mantle]

Can u check my model request please

[echo aurora]

Yeah I've been seeing those.

I rarely respond in those thread, but they are being noted for consideration.

[proud bobcat]

GLM 5 on the other hand has very strong reasoning, basically crushing this benchmark easily.

[sick mantle]

K

[peak sapphire]

@echo aurora why nano banana pro is always don't working? I'm already tired of seeing messages "something went wrong"

[sick mantle]

look what i made pineapple

[echo aurora]

For this error the message, this [pinned message](#1417174113092374689 message) outlines more info about the error along with best next steps.

[echo aurora]

Forwarding to #ai-creations

[peak sapphire]

Literally every other model except for nano banana pro works perfectly fine and all the time; she's the only one causing problems.

[void pendant]

Something went wrong with this response, please try again.

help pls

[echo aurora]

Can confirm this isn't the case. Other models can, and do, error out as well. Popular/high demand models can have more troubles. Regardless of the model or reason for Something went wrong, following the steps in the pinned message are the best steps to take.

[echo aurora]

Check out the steps listed here: #1417174113092374689 message

[sick mantle]

LMarena is the best! pineapple did a very good job.

[sick mantle]

droped the lm now arena.

[mortal coyote]

which model ?

[sick mantle]

Seedance 2.0

[peak sapphire]

Thanks, clear site data is real help

[sick mantle]

Not on arena yet

[mortal coyote]

its here on lM ?

[sick mantle]

No

[mortal coyote]

OHH ok

[sick mantle]

By the way check my idea in #1473758402218823775

[echo aurora]

Yup, the request from the community to have Direct & Side by Side for Video Arena is very much on our radar.

[sick mantle]

So will u might add it?

[sick mantle]

Beacuse Video Arena chooses a random model for me that i do not want

[golden ocean]

Large Language Model Arena

[echo aurora]

Going to give the boring answer I give any and all requests for ~"Is X feature or Y feature happening?"

I won't be able to share details about what new models or features are upcoming until we're ready to share more. Would recommend to keep an eye on our announcement channel.

[drifting tapir]

Whats about 4o

[limber panther]

chat

is it js me or is the rate limit of sonnet 4.6 worse than opus 4.6

even tho sonnet is cheaper model

[echo aurora]

The model was removed, more information can be found here: https://help.openai.com/en/articles/20001051-retiring-gpt-4o-and-other-chatgpt-models

[sick mantle]

Its alr

[proud bobcat]

release 5.3 coward

[golden ocean]

🐮ard

[pale sonnet]

openai is releasing to quick its been 2 months since 5.2 came out idk if you can make much of a difference in that amt of time

[fiery gull]

Normal

Like there was AI that the restrictions were more annoying than opus even costing 10x less

[stuck orchid]

Why does claude-sonnet-4-6 on Arena have such high request limits?

🧐

[pale sonnet]

they might get low/no rates on certain models since arena is partnered with anthropic

[stuck orchid]

Really?

Arena is partnered with anthropic?

If that's the case, then that's very good 👍🏻

[hazy zodiac]

Y'all are having problems while registering in the copilot arena( vs code )

[pale sonnet]

yeah theyre also partnered with openai and google

[stuck orchid]

I hope they will allow Arena.ai to remain free for direct communication with models and for battles with them

[echo aurora]

@jolly cliff Note that Video Arena has been removed from the server. More information can be found in this announcement.

[hazy zodiac]

@echo aurora

[echo aurora]

Note that copilot isn't supported anymore sorry to say

[hazy zodiac]

A..

[hazy zodiac]

Any other extension to connect lmarena with vs code?

[echo aurora]

Not that I'm aware of.

[stuck orchid]

It's better not to do this on a regular basis. Otherwise, arena.ai will use up all the funds

[hushed gyro]

how do you access seedance outside of china?

[sick mantle]

I recomend a vpn.

Then connect on the vpn to hong kong

[queen veldt]

Petition for songarena

We want a songarena

[hushed gyro]

and then...?

[hushed gyro]

cuz you need a phone number

[hazy zodiac]

Ohk

[compact flame]

You gotta offer cookies for people to sign your petition

[echo aurora]

+1

[sick mantle]

Go to https://www.doubao.com/chat/

@hushed gyro then go here https://www.doubao.com/chat/ then type Make me a video of then type any thing u want then wait 1 - 3 mins then its ready

or i can make one for you.

[hushed gyro]

what vpn do u use

[sick mantle]

i recomend vpnfy

but seedance is scary good..

[compact flame]

Is opus 4.6 thinking is a Extended version of thinking like on the original website or smth

Since I can't find anything related to Opus 4.6 thinking api

[hushed gyro]

it says you need to register

[sick mantle]

Just do it

[hushed gyro]

I need a phone number?

or do I do it with tiktok

[sick mantle]

Just dont login try genrateing again

[elfin sundial]

is there any free API key for lmarena

[golden ocean]

w

[elfin sundial]

does anybody know how to get access to Arena's api key if they have one

[hushed gyro]

well if only I can bypass the login screen when I press seedance 2.0

[sick mantle]

dont press it just go to chat then type Generate a video of a chicken

[cloud sierra]

Genera un video donde esten caminando el hombre y el perro por la calle sin cambiar el rostro ni las facciones todo al 100 %

[hushed gyro]

im sorry but like after 3 generations of NB Pro the rate limit triggers? why so little!!???

[sick mantle]

Open a incognito tab

[hushed gyro]

does that like solve the problem

oh also seedance wouldn't let me generate as it has a real face

[sick mantle]

Maybe try grok ai

[hushed gyro]

grok video is dogwater tho

[sick mantle]

i know

but they updated

[hushed gyro]

Seriously I think video arena should have side by side like man are they stupid

[gleaming roost]

😂

[hushed gyro]

@echo aurora like I beg you to add this feature ASAP

You will gain a lot more users as ppl now just get disappointed with poor quality results as the website picked the worst models to generate like wtf

[sick mantle]

Look what they made

[hushed gyro]

yeah but my videos are like someone shooting a rifle and it has to have sound and like whatever, it's complicated and only sora can do it

CHAT does someone have sora INVITE CODES give one to me pls for free..

[sick mantle]

I think

[golden ocean]

wtf how is spiderman eating

[sick mantle]

Idk

He just did that for a fcking cookie.

[hushed gyro]

do i have to login or just leave it, clear cookies then repeat so infinite glitch?

[sick mantle]

Yea

No dont login

[hushed gyro]

oh lmao

[hushed gyro]

this is grok? impressive

[sick mantle]

Yea

They did good

[hushed gyro]

thanks

[sick mantle]

Yea i know right.

But i dont got a chinese phone

[hushed gyro]

herosms is an online website, they created numbers already, just click on a phone number you choose and wait for the SMS to pop up

[sick mantle]

Yea and grok kinda became better

[hushed gyro]

btw is this weird @sick mantle

[sick mantle]

Yea

MAN GROK KNOWS SNOOP DOGS VOICE?

@hushed gyro

[limber panther]

are u wasted or something

sora doesnt need invite codes anymore..

[sick mantle]

croquemonsieur_70 dms

[sick mantle]

Man for the app it does

[limber panther]

use the website, sign up and then log in the app

🤔

[hushed gyro]

yes i am wasted 🤣

[tawdry robin]

can this monitize guys

[echo aurora]

@gray surge Note that Video Arena has been removed from the server. More information can be found in this announcement. @fiery shale

[vagrant silo]

@echo aurora when are you going to add kling 3.0 to Arena?

[echo aurora]

Going to give the boring answer I give any and all requests for ~"Is X feature or Y feature happening?" -> I won't be able to share details about what new models or features are upcoming until we're ready to share more. Would recommend to keep an eye on our announcement channel.

[vagrant silo]

okay tnx i appreciate it anyways

[quartz light]

oh boy i love holding down my mouse and spamming ctrl c for 6 minutes straight knowing the request is going to time out because arena's frontend decides to wipe the whole thinking process if the response times out

[thorny cove]

Is X feature or Y feature happening?

[echo dome]

recaptcha again when you generate images (idk about chatting)

[echo aurora]

@opaque cloud Note that Video Arena has been removed from the server. More information can be found in this announcement.

[echo aurora]

Can confirm that this is & isn't happening.

[wicked sage]

hi

[thorny cove]

i sure do love writing my video prompts into chat

[echo dome]

@echo aurora
chatting uses cloudflare
generating images uses same as you brought enterprise

[thorny cove]

https://cdn.discordapp.com/attachments/1462725688145481748/1473123300795482142/Screen_Recording_2026-02-16_180347.gif

[sick mantle]

@echo aurora Bro people are asking for u to add a option where u can select video models but if u add it then thanks

[thorny cove]

what

you can

[echo aurora]

Sorry what seems to be the issue?

[thorny cove]

dark magic

[sick mantle]

I mean video models

[echo aurora]

They want Direct and Side by Side

[astral vortex]

nano banana really having alot of fails

[echo dome]

my issue is it uses same unsecured captcha for generating images

[echo aurora]

This feedback is on our radar

[sick mantle]

Just watch him eat cookie

[echo dome]

or both again?

[thorny cove]

can you also remove the video generation limits and the model rate limits... i am going to BANKRUPT yall 😈

[wicked sage]

pineapple quick question, are you guys friends with any big ai corporation?

[astral vortex]

Is it true that nano banana pro is failing alot

[echo dome]

i mean you finally changed that data collector to cloudflare when chatting with bots

but not images

[sick mantle]

Yes.

[thorny cove]

yeah

among us ai

cheers now the wife and kids are crying

[echo aurora]

What do you mean by "unsecured captcha" ?

[wicked sage]

real

[thorny cove]

low taper fade is still massive btw

[echo dome]

recaptcha by google

[echo aurora]

High demand models can have higher than usual error rates.

[echo aurora]

Yeah will get right on it

[echo dome]

that's what i mean

[astral vortex]

Kk