general

[keen beacon]

Blue pill well forget i even mentioned anything

Hey, I gotta get back to work. I’ll talk to you guys in a little bit later. Good luck everyone.

I got a benchmark for anybody here who wants to pass the benchmark to see if they’re smarter than the AI currently

If anybody could solve this without using AI and tell me what it’s meaning is you 20 times smarter than the AI

It’s so easy it’s in front of you

🖐️

[sinful cedar]

Guys I might be making a mistake or something, but lowkey ever since I found out about lmarena, I've been just using direct chat, and using whichever model helps me the best for "x" task, instead of going all into Anthropic or OpenAI. Am I missing out on any features really by just using lmarena?

[fickle venture]

I should watch that, what's on the video tho?

[fickle venture]

Lmarena is good at leaderboard because user can shoose what model is the best and it will be ranked

[sinful cedar]

I find that it also gives me access to the models stuck behind an API and I can use the high or low models etc. Is this actually allowing me full access to the model through LMArena, or skimming some parts of it?

[queen veldt]

There is no computer

Is the answer

[native yarrow]

first is the screen of the monitor, second is the perception that the monitor is a container for something real within, and the third is a perception that whatever was in the monitor is now out and real?

idk lolz

[fickle venture]

There is limits tho and a cooldown before using the model Which is 50 minutes, so it's not unlimited

[latent crest]

How many free generations we have in nano banana pro?

[sinful cedar]

True but if I'm maxing out Claude, I could hop over to ChatGPT, or Gemini

[fickle venture]

What do you mean?

[fickle venture]

I think it might work idk

[latent crest]

It says to wait 50 minutes like a lot of time

[sinful cedar]

Honestly, I think the method could be to invest in a Anthropic Max plan, then when that gets full utilize LMArena. What are other people doing?

[fickle venture]

I mean you have to it's not unlimited

[fickle venture]

That'll cost alot, Claude is just expensive asf

[sinful cedar]

true, I mean I'm a student so I'm already getting gemini

[native yarrow]

dont wast emoney on claude

[fickle venture]

On LMarena or Gemini app?

[sinful cedar]

Gemini web

[fickle venture]

There is a free student plan 1 year free ai tools right?

[sinful cedar]

yeah that's what im using

[fickle venture]

I really need it bruh

[fickle venture]

Have they asked you for credit card

For verification

[sinful cedar]

from looking at livebench ai, it appears that gemini pro 3 and gemini pro 3 on lmarena utilize the same API so there's no difference unless you specifically dont want or dont need web search then use lmarena.

[fickle venture]

Idk stupid people on Twitter just waste their money on it, it can't even make a dam great looking Ui and alot of errors.

[sinful cedar]

I don't think so, I mean I could double check rq tbh

[latent crest]

I know. I was wondering how many “creations” before u hit the 50 mins wait

[native yarrow]

its just a dumb model too

gemini 3 is better at coding

[sinful cedar]

@latent crest depends I think with Claude it either uses a few tokens or every single token according to what i've read

[fickle venture]

They updated it check new changelog:
#announcements message

[sinful cedar]

and Claude is apparently only useful for the Claude Code part and or a Max plan the Pro basically helps start projects but rarely is available when you're at the finish line you're better off using Bard

[native yarrow]

thast the image upload limit

[fickle venture]

For Gemini 3 suck at coding, I ahve it a dam easy project but made alot of errors in files and UI so bad. But this new Claude Opus 4.5 literally the best one

[native yarrow]

eh not for me

[sinful cedar]

Anthropic is just king for everything unless I need a translation or UI fix tbh

[fickle venture]

I mean Claude is Only trained for coding and Gemini 3 is trained on everything

[sinful cedar]

I haven't touched OpenAI for ages

[fickle venture]

They died forget about them

Chinese ai are better than openai

After what they did to GPT 5

[sinful cedar]

I remember when Deepseek made an entrance to the ai game at halftime like lebron

[latent crest]

What’s Flux pro and Flux flex??

[native yarrow]

flux's sad attempt to topple nb pro

[sinful cedar]

@latent crest open source ai for images

[fickle venture]

New text-image model

[queen veldt]

Nice

[fickle venture]

DeepSeek literally changed everything

[sinful cedar]

They told investors quite literally, "put your money where your mouth is" and dunked on them.

[fickle venture]

Probably it bugged try again

[queen veldt]

Nah

It should work 100/100 times

No exceptions

[fickle venture]

The stock went all red that time lol

[sinful cedar]

I'm very excited to see how Deepseek implements their OCR technology, could be ground breaking for these other LLMs to interpret and process large files/PDFs.

[latent crest]

Pro is to create and flex is to edit ?

[fickle venture]

They were planned to release DeepSeek R2 on April but they delayed forever because they don't have better graphics card after what trump did to china

[sinful cedar]

Possible stock short opportunity? Honestly I may put in a few options trades and auto trade when they make all the other AIs go red again, then use that capital to reinvest into API tokens for Claude Code.

[fickle venture]

Flux is for text-image and image edit

[sinful cedar]

@latent crest Flux Pro -> Create Model, Flux Flex -> Control Model, Flux Fill -> Editing Model

[fickle venture]

I might convince you not to buy any of the ai right now they are weak, you need to wait for 2-3 years they will get improved real quick and they will be helpful for you and in your projects

[latent crest]

But there’s no flux fill bro

[sinful cedar]

are you looking at the lmarena options?

[latent crest]

Yessir

[sinful cedar]

Ah

Okay so I was looking at the official APIs from Flux AI, by looking at the actual website from LMarena, the Create model is Flux 2 Pro (best boundaries also for people who don't care just best quality), the control model is the Flux Flex (people whom are more particular about how long it does on certain aspects), then we have the Flux-1-Kontext-Pro, this is the edit model Lmarena has chosen aka "Flux Fill" which is to change objects, inpaint, fix hands; basically understands the "kontext" to make the seamless edits you want

i wont even blame you if you use ai to give you a summary my bad i ripped an addy and working on a project right now @latent crest

[sinful cedar]

Yeah, I guess that's where my initial "does it make sense to use LMArena, or should I use LMArena for backup small tasks then invest in Anthropic for projects (Claude Code)" question came into.

[latent crest]

Do not apologize , it’s not u , it’s me that is a newbie that knows nothing

[keen beacon]

Your missing the point why people pay for it

The people that are paying for it now and 2 or 3 years where do you think they’re gonna be?

[sinful cedar]

The only thing I wish LMarena could do, which I could understand completely, is the file upload of PDF's. I think they don't do this because they use it for information collection to then make money and keep it running

[fiery gull]

Is possible I use the gemini 3.0 agenticly in gemini app?

[keen beacon]

By the time you decide to start using it full-time?

You’re just gonna be paying to catch up.

What is the price of not knowing?

[sinful cedar]

Interesting, do you think now that Anthropic added conversations, I should invest in the Max plan, then use LMArena for tedious small things, to balance out the tokens? @keen beacon So then eventually when they produce models that use barely any tokens (nearing unlimited), I will have a whole archive behind me to reference?

[keen beacon]

From 2023 someone posted earlier

It’s almost 2026

[sinful cedar]

So I'm curious, what's your take on how or what I should be investing in at the moment. Anthropic or OpenAI. Gemini is already being paid for by education.

[keen beacon]

I think you’re asking the right question but in the wrong order

[sinful cedar]

If I go Anthropic, what plan, Max or Pro, or API based?

[keen beacon]

The question is, what are you willing to pay to learn?

[sinful cedar]

Could it be more beneficial if I told you my intended reasoning for the models and hear your feedback?

[keen beacon]

It would be irrelevant because at the end of the day the only person that knows what’s best for yourself is only you

And like everything in life, you get what you put in

You gotta pay to play

[sinful cedar]

I personally believe paying to learn isn't necessarily a 1:1. I think just by using LMarena is a biproduct of education on AI models along the way.

[keen beacon]

Let me rephrase this to a different way. What is the cost of not knowing?

At the end of the day, it cost the same because you still have to pay the price to learn

Either by mistake or by education

[sinful cedar]

@deep adder Investing as in investing my money into a plan

[keen beacon]

Have you ever had an expensive hobby you got into and you made a mistake and it cost you?

[sinful cedar]

Absolutely, but that can be tied to impulsively and blinded excitement of something new

[keen beacon]

Well, that price you paid that’s the price of not knowing

So when a person goes to college or university they paid so they could learn what they don’t know

The same when we make mistakes or errors and it cost us

That’s what I mean when I say that at the end of the day, you still pay the price

The real issue is the cost of not knowing, so the question I’d be asking in your position is: what price am I willing to pay now to learn what I don’t know, so I don’t end up paying even more for that ignorance later.

[keen beacon]

For me, I do simple math in the sense of let’s say entertainment I am willing to pay 1$ for an min of fun lol

[sinful cedar]

I'm investing my time (and subscription money) into learning/using a platform, betting that the company will continue producing quality models. Sorry I just didn't frame it correctly, my apologies.

[fiery gull]

why gpt plus?

[keen beacon]

there is no right or wrong. The whole point here is that this is something only you could answer.

You’re asking the right question, but nobody will have the answer except you

[fiery gull]

we money is 0,001% of money to AI, the rest is from shareholder

[fiery gull]

just talk to people the AI is the future, you is doing more that send money

[keen beacon]

If me personally, I was in your shoes and I used AI daily and I got a lot of value out of it just for messing around

[fiery gull]

is that I do

[keen beacon]

I’d get max plan

But if you’re just gonna be dabbling here and there, it may not be worth it

Yeah, I think it’s worth it also

And you get even more bang for your buck with Google

But if you want better performance and coding, then that’s gonna cost you what it’s gonna cost you

It’s a give-and-take

[sinful cedar]

You all have very extensive and great points. I guess I just don't know what to do with what model. Whenever I can I always use Anthropic, however I don't know if by using just LMArena, I'm doing myself a service or praying on my eventual downfall.

I'd love to use PDFs with Anthropic and extract more technical information for data analysis of the financal markets, or for research. Just don't know if my value will be extracted accordingly yk.

[fiery gull]

use yupp.ai so

the yupp accept pdf

I'm tired for ask pdf to lmarena

[sinful cedar]

I had no idea that even existed thank you! @fiery gull

[obsidian cargo]

@echo aurora the "multiturn editing" thing isn't disabled properly, there's context pollution going on.

[sinful cedar]

I guess I'm curious as to why LMArena doesn't include file uploads / PDFs for models only images?

[fiery gull]

IDK ;-;

[keen beacon]

Good luck 👍

[sinful cedar]

Oh gosh the crowd said good luck lol

[echo aurora]

Yeah so what's happening here is that model has a setting to take in old context, which is essentially what multi-turn image edit was doing prior. With this turned off, we recognize it doesn't make sense to have this setting on, and we're in the process of making corrections.

[obsidian cargo]

alright cool as long as its being addressed ^_^

[sinful cedar]

@echo aurora Whats your take on this, I feel like you have background knowledge

[echo aurora]

Yeah not just how the model handles this, but also the effects this potentially had on votes, and ensuring our leaderboards are accurate.

[sinful cedar]

Thank you for the comparisons I've already learned quite a bit just from talking to you and the value that I'm trying to extract out of these things

[fiery gull]

yeah.... I'm using more the opus 4.5 in the pratice... I don't fell a improve
just my first impression is good but, Idk, the opus 4.5 don't is the sonnet 4.5 with hallucination rate of haiku 4.5 and smart of gemini 3.0 pro

I'm just sad with opus 4.5

;-;

[echo aurora]

Hmm I think I'm misunderstanding the question so please help me out if I'm wrong, but the reason we allow for image upload for Text is for Vision capabilities. https://lmarena.ai/leaderboard/vision

[sinful cedar]

I understand image uploads test Vision capabilities (can the model understand/analyze an image). But many models also support PDF and document uploads for document understanding - analyzing text, tables, charts in PDFs, etc. So I'm wondering why LMArena doesn't have a similar arena/benchmark for document understanding - where users could upload PDFs and compare how different models handle document analysis, summarization, extraction, etc. Is that a planned feature, or is there a technical/practical reason it's not included?

I could be completely missing something please tell me if I'm wrong

[echo aurora]

Oh gotcha, thank you for elaborating. We absolutely want to add more file types for upload capabilities. If you'd like to share what file types you think we should prioritize using this thread would be helpful: #1432965582965440623

[native yarrow]

gemini 3 has successfully made a football simulator for me

its actually crazy ngl

[sinful cedar]

Absolutely, I will. One final question before you go: Is this process simply about gathering user feedback for consideration, or are there undisclosed technical limitations preventing this? @echo aurora

[echo aurora]

We take a lot into consideration when deciding what to build -> feedback, internal data we're seeing, what we think would make the most impact, etc. etc. At the end of the day it's all about how and where we allocate energy.

[sinful cedar]

Just sent my recommendation to the correct thread, thank you for sending me there!

Ah I see! When you say allocate energy, I have one last question. When we select the certain model for testing, are we actually using the specified model or a skimmed version of it to save on pricing? @echo aurora

[echo aurora]

It's the model that's offered by those model providers. We'd never say a model is X knowing it's Y. It's paramount for our business that people trust our leaderboards, so high integrity is a must. Now there are cases where mistakes can happen, in which case we'll be upfront about those and make corrections.

A good example of this is what was pointed out earlier in this message: #general message

A model had the ability to take in additional context, which is something we were doing for all (mutli-turn image edit), until earlier today when we turned this feature off. However, the model continued to take in the context when other weren't.

[fiery gull]

do you tested the nano banana 4k?

[echo aurora]

This doesn't make a lot of sense fairness wise, so we're working on making a change, while invalidating any votes we think we're unfairly obtained.

[fiery gull]

the people is talking is soo above that nano banana pro 1k

[obsidian cargo]

I dunno, after 1k I don't really notice a difference.

[sinful cedar]

I completely understand. I assume the data collected from user feedback and testing is being monetized, allowing LMArena to reinvest that revenue into maintaining operations? (without prying too much undisclosed information, sorry just curious it's an amazing website so I'm wondering)

[fiery gull]

I'm dont saying about resolution, the nano banana 4k (the people are saying) edit and create images better that nano banana pro 1k

[obsidian cargo]

oh, I've been flipping back and forth between using lmarena and gemini app for image generation and haven't noticed a difference

[native yarrow]

nah, lmarena is the same as gemini app

[echo aurora]

Short answer is yes. We are hiring with the intention of creating a better platform overall (including site reliability).

[sinful cedar]

Interesting, so apart of the maintaince of the website, I'm assuming some of that profit is for upkeep of models. I'm still surprised on how we can access such a diverse number of LLMs for free without a single sight of an advertisement for extra revenue. @echo aurora

[echo aurora]

The long-term plan balances public access with enterprise services. Enterprise and research partnerships generate revenue to sustain the platform, while all public leaderboards and evaluation data remain freely accessible. To be clear though we do not sell user data or individual voting information. Commercial activity is limited to evaluation services, not data monetization, which you can learn more about here - https://news.lmarena.ai/ai-evaluations/

[sinful cedar]

Woah! I didn’t know about that at all, thanks for the clarity, I assumed the website was collecting user information to then sell to data brokers for maintaining operations.

I appreciate your time and the feedback of others I’m logging off for the night, but thank you so much!

[echo aurora]

I'm glad you asked! I hope you have a nice evening.

[keen beacon]

See ur smart

[keen beacon]

😔 no. Unfortunately that’s not the right conclusion

Flux

[frank tulip]

Yo is there a way to solve the problem where I'm stuck at generating?

[keen beacon]

What’s stuck at generating the generation is not coming through and just looping?

[frank tulip]

Yea I input a message and it's stuck at generating and won't give me an answer

[keen beacon]

Have you tried refreshing? Was it giving you answers earlier and what model were you using? Because are you aware of rate limits on certain models?

[frank tulip]

Yea it was giving me an answer earlier, but it was lagging out so I refreshed the page, but then after I refreshed it stuck at generating, and no it won't give me a rate limit warning ;-;

[keen beacon]

You got a diversify

[frank tulip]

Wat's that

[keen beacon]

I think you could try it again in like 50 or 60 minutes or something like that whatever the rate limit is

It means try another models that are available in the meantime

People usually get hung up on a single model they like or perceive as the best model and so they don’t fully go exploring and seeing what else is out there

[sharp sorrel]

Excuse me, is it no longer possible to regenerate the image using the refresh button?

[keen beacon]

That feature as far as I could tell has been

https://tenor.com/view/terminator-arnold-im-back-gif-23975209

[keen beacon]

(Best thing that can happen to someone who’s trying to learn and get good)

[frank tulip]

I mean how can I try again when it won’t let me, I’m in a chat with a lot of previous message and don’t want to lose the memory, but now I got that problem and don’t really want to start a new chat ;-;

[keen beacon]

Ahh the classic dilemma

[frank tulip]

Yea

[lyric yacht]

Hey everyone
Excited to be here. I’m a full-stack & blockchain engineer working at the intersection of AI and Web3 — agent pipelines, trust layers, on-chain execution systems, AI wallets, and autonomous smart contract interactions.

If anyone is exploring AI agents, compute markets, verifiable inference, or on-chain automation, I’d love to exchange ideas. It’s one of my favorite areas to build in right now.

[keen beacon]

Unfortunately, that’s a very common situation. Most people find themselves in.

There really isn’t a easy answer. It’s either waiting or starting a new chat.

[frank tulip]

If it give me a rate limit warning or an error, I can switch model and move on with my life, but it’s just stuck at generating so idk ._.

[keen beacon]

If a stop sign is missing the letter S do you still stop?

Or do you go up?

[frank tulip]

Nah I’ll go on top of it

[keen beacon]

Well, then there we go

Just hang out until it stops spinning

[frank tulip]

Aight then I gave it like, 8 hours and it won’t stop spinning, so I guess I’ll give it some more

[keen beacon]

Oh wow

8 hours shouldn’t do that if it’s eight hours

[keen beacon]

Is it a coding model?

If it’s eight hours, bro, that’s not normal

[frank tulip]

No, it’s a regular chat model

Gpt 5.1 to be exact

[keen beacon]

Weird usually the rate limit is like 60 minutes or something

Oh

And you’ve tried refreshing?

Tried starting a new chat?

[frank tulip]

Yea I’ve tried it

[keen beacon]

Try to start a new chat see if you’ll have the same issue

[frank tulip]

Yea I started a new chat and it worked perfectly

[keen beacon]

Then you will know if you’re being limited or not for sure

Oh dang ya

That doesn’t sound like a rate limit then

Sounds like you got banned

[frank tulip]

Yea sounds more like a loop to me

[keen beacon]

Jk

[keen beacon]

But the thing is what would it be looping for for eight hours?

[frank tulip]

Bruh imagine being banned by an AI

[keen beacon]

lol

It’s happened

Plenty of times

Too many people I know

[frank tulip]

Yo what

[keen beacon]

Not on here

[frank tulip]

Yea .-.

[keen beacon]

But ya people burn through accounts

I went through 20 this week on Reddit alone lol

[frank tulip]

Nah people seriously got banned by just talking

[keen beacon]

Well, it depends what they’re talking about but yeah, it happens, bro

One of my buddies he got an email from open AI that he can’t ever use any of their services again lol

[frank tulip]

Damn

[keen beacon]

https://decrypt.co/349943/cameo-wins-temporary-restraining-order-against-openai

[keen beacon]

Kinda hard to get banned though to be honest

I’ve gotten an warnimng once when I first got ChatGPT but that was almost 2 years ago

[frank tulip]

Yea guess bro was just getting too freaky

[keen beacon]

Some of these people don’t care and they’re just brute, forcing and others are just idiots

They don’t know how to blend suspicious activity with normal activity and so they do the most obvious things in the most obvious ways

Like a sure way to get banned is pushing nsfw

[inner gate]

whats pro grounding 🤨

[keen beacon]

Just over and over

[still wolf]

Is grounding rated already?

[keen beacon]

Sorry, but I wish I could help you

I thought it was a regular limit for sure but sounds like that ain’t it. I’ve never experienced it so I don’t know exactly what’s causing it

=/

[echo aurora]

A new search model, check it out: https://lmarena.ai/?chat-modality=search.

[keen beacon]

dude 🤦🤦🤦

[echo aurora]

Not yet, same with 5.1-search. Stay up to date with our Change Log: https://news.lmarena.ai/leaderboard-changelog/.

[keen beacon]

Open ai really this dumb?

Baron App Inc., which operates Cameo, sued OpenAI last month, claiming trademark infringement, trademark dilution, and unfair competition tied to OpenAI’s text-to-video model Sora 2’s “Cameo” feature.

Lmao

I mean, out of all the words you could pick why would you pick the word cameo?

The restraining order prohibits OpenAI and its officers, directors, and employees from using "Cameo" or any confusingly similar marks, including "Cameos," "CameoVideo," or "Kameo,” for its Sora AI video generation products and related marketing in the United States.

Because the other alternative is calling it what it really is a deep fake lol

[still wolf]

Probably for instant recognition. They’ll try to say it’s a Verb like saying a Zoom call

[keen beacon]

A cameo was a nice way of saying deep fake without it sounding so bad

I’m curious the word they’re gonna pick next

Look brah this is so obvious I don’t get how anybody there thought this was a good idea

Actually, I take that back now that I think about it I’m not even sure what word would replace it

What’s in between deepfake and cameo like a single word ?

Avatar?

[fickle venture]

Better searching

[vivid coral]

Holy crap the search arena is lit 🔥🔥

[inner gate]

oh is it a research model

[native yarrow]

how so

[severe canopy]

Flux 2 plssssssssss

[torn mantle]

https://x.com/dwarkesh_sp/status/1993371363026125147

Dwarkesh Patel (@dwarkesh_sp)

The @ilyasut episode

0:00:00 – Explaining model jaggedness
0:09:39 - Emotions and value functions
0:18:49 – What are we scaling?
0:25:13 – Why humans generalize better than models
0:35:45 – Straight-shotting superintelligence
0:46:47 – SSI’s model will learn from deployment

X · Nov 25, 2025, 5:29 PM

ilya didnt say anything useful

hes been saying scaling is over for like 2 years now

but is it really over?

[austere sundial]

Seriously, these LMarena updates have been awful. A month ago it was the best possible version, and now they've simply removed everything from the website and replaced it with limited or buggy versions.

[keen beacon]

His bitter

[keen beacon]

Already here

Looks like he got some sun though so hopefully he’s doing better

[keen beacon]

It’s here

[torn mantle]

hows flux 2

[keen beacon]

I like it

All flux

[torn mantle]

is it flex or pro

which one is new

[keen beacon]

Pro

I’ll make a new one

[torn mantle]

whats flex then

like smaller model?

its not bad ig

but still not on nb pro level

• slower

i mean how can you even compete with something like that...

feels unfair tbh

[keen beacon]

Flex is nice

I would say all models have their pros and cons

[native yarrow]

nothing will ever top nb pro for the remainder of this year

[keen beacon]

Oh, for sure the remainder yet 100% I agree with you

[native yarrow]

NB pro is lowkey insane

[keen beacon]

What without a single doubt?

You’re totally right

It may be a while if anybody crowns it probably

[native yarrow]

proportionally in terms of blowing other models out of the water its the best image model in history

actually

it is just

the best image model in history period

[keen beacon]

Close to spring, I guess

[native yarrow]

flux is nice though

[keen beacon]

Maybe even later

[native yarrow]

honestly, only maybe next winter would i see a new king top nb pro

and jesus

with how good nb pro is now

imagine something BETTER

[keen beacon]

I use nano everyday for hours. It’s nice, but I’m not all that impressed with where I thought it would count.

I thought they would fix the minor issues that the old one had and I was looking forward to that

[torn mantle]

yea

[keen beacon]

It definitely improved a lot, not gonna lie

[native yarrow]

i havent had any of those issues yet like

with the no-edit issue

if your prompt is really unclear and hard to understand, yeah itll just

regurgitate the same image cause it doesnt know what youre saytingh

[torn mantle]

i already said that but nb pro is like an agi moment for me

[native yarrow]

yeah

imo

nb pro is no longer an image generation model

its like a pair of eyes

[torn mantle]

and it will only get better from now on

[austere sundial]

LMarena is worth nothing anymore, literally an image I generated a month ago vs an image I generated today, it's bizarre and horrible.

[torn mantle]

yea its crazy

[native yarrow]

like it UNDERSTANDS what you give it

[torn mantle]

exactly

[austere sundial]

What's going on with this site, it's generating more bizarre models than chatgpt and grok together

[torn mantle]

im also a visual learner

i love infographics and notes rather than like long text document

[keen beacon]

Try something complicated, multi angles consistency

[torn mantle]

so i find nb pro handy on that

[keen beacon]

One second I’ll give you a prompt

[austere sundial]

The photo I used with the same prompt as the image from a month ago

[keen beacon]

How do you say unborn? I don’t wanna traumatize myself again.

I keep looking at these stupid ships and In my images

[austere sundial]

Can anyone tell me if they nerfed the site?

[keen beacon]

And it just starts frustrating me

They didn’t

But I can understand why it feels that way

They took away a lot of the features that made this place truly special

But I think that just comes with the territory sense of the growth

Plus in order to upgrade you need to change things around

Otherwise, we wouldn’t have cool features like web development and all these other cool things we take for granted now

[native yarrow]

like?

[atomic moon]

how so?

[native yarrow]

its just insane man idk how to describe it

just look at the examples

[atomic moon]

it has better quality and understanding but it doesn't really blow me away

[austere sundial]

It only got worse I no longer see an advantage in using anything else on this site for example

[keen beacon]

I get where you’re coming from and you do show really sophisticated examples of your images

But try even more complicated things

[native yarrow]

do tell

[native yarrow]

idk how, it blows everything else out of the water

[atomic moon]

how?

[native yarrow]

because it does man

theres plenty of examples supporting that

:P

[atomic moon]

ah, just trolling

gotcha

nearly 3 month old model comes pretty close to it

[keen beacon]

Like city building and controlling your character fully within

Btw all these were made with nano ones

So yeah, I mean nano2 this far more superior

And no model kind of comes close so he does have a really strong point there

[native yarrow]

no, not trolling

[keen beacon]

I’m just being greedy

😝

Cause it would totally suck not having nano 1 or 2

Better then nano 1

[native yarrow]

the heck

[keen beacon]

Irish is right on many of his points

[native yarrow]

why does it randomly delete messages lol

[keen beacon]

Idk

I’m just being stubborn and greedy lol

[native yarrow]

i mean to not be impressed with nb 2 is just ragebait atp cus

like how?

[keen beacon]

I don’t know I think once you get to a certain amount of image generations under your belt

It’s hard to explain

I guess the world would be not impressed, but I guess I’m spoiled

I’m taking it for granted and I’m not appreciating what it is

[native yarrow]

nah i mena

you just get used to it

your standards increase

[keen beacon]

Yeah but my expectations are also not realistic

[native yarrow]

but if you take a step back and look at the MASSIVE jump from other models to nb pro

its insane

[keen beacon]

Because in my mind, the ideal model would just do anything without me even saying or thinking anything

But I get you and it’s good to be reminded and come down to my senses so thank you

That’s what I get for buying the hype lol

No, for sure more then a jump it definitely set the standard way higher

I do wonder what affect having Google search integrated into the model that it could search up images in reference to them. I wonder what extent this plays a role.

[native yarrow]

honestly, i havent seen that in works yet

that might just be a lie

but otherwise still like genuinely just an amazing model

it works more like a digital editing software more than teh other nano banana ever did

[keen beacon]

Oh yeah, I forgot I just remembered

The price lol

Now I’m back to my equilibrium lol

[cloud zinc]

yea the leap is insane.

[keen beacon]

Actually, it’s not too bad. It’s $.12 an image. 2k

I think I overreacted when I saw the api price 4k at .25

[native yarrow]

nah its not bad

i honestly dont use 4k for some reason

[keen beacon]

No it’s not

.12 cents chill

But u gotta drop 120$ lol

It sucks having constraints on creativity like this

If you’re not careful, you could really blow through it especially if you generate 10 hours a day non- lol

I think it might peak I had like 8 accounts or something

[native yarrow]

on what?

[keen beacon]

GPT back in Dalle days

So I never have to hit that rate limit lol

I would generate from the moment I wake up to the moment I go to sleep lol

Just running prompts heavy

I got some really rare images

Stuff you wouldn’t even believe that ChatGPT generated it

But all nsfw =/

No where 2 share

[cloud zinc]

get ai ultra, its unlimited

[keen beacon]

Well, thanks the AI video I slowed down on the img gen

And diverted my attention towards mastering video models

But it definitely helped having that experience from image generation

Translated really nicely into video

[cloud zinc]

hopefully nano banana moment happen for video early next year

[keen beacon]

🤲

[native yarrow]

its just gonna be censored to oblivion

[keen beacon]

I guess the only other thing that kind of really excites me and impresses me about AIS the guard rails

[native yarrow]

unless google does it and does the SYNTH id stuff

:P

[keen beacon]

That’s kind of the only thing I’m really interested in videos and the guard rails

[cloud zinc]

obv they have synth id for video

[native yarrow]

even then

[keen beacon]

Easy 2 break

[native yarrow]

i feel itd be censored to oblivion

[keen beacon]

There’s ways around that censorship that are so easy but very time-consuming to find

They do

But if you was the video twice and run through different media players in save it you get rid of it

So this the og

After washing

Removes meta data

[keen beacon]

Damn that Suno thing bummed me out

[quartz pike]

yall

how good is flux 2.0?

[prime mulch]

Rumours says on par with nanobanana pro

I didn't use that myself

[quartz pike]

how expensive is it?

[keen beacon]

It’s already in the arena

[limber crag]

its not

its on the site

[prime mulch]

How is it

[limber crag]

its a step up from flux 1 but its not as good as even nano 1

[prime mulch]

Is it worth or not

[keen beacon]

I see it right now

[quartz pike]

meh

[limber crag]

exactly

[quartz pike]

its def not on par.

but its kinda good.

[limber crag]

people just say it beat nano banana as a thumbnail

clickbait

[cloud zinc]

its slighly better than nano banana 1

[prime mulch]

Just tried nano banana beat it not even par with seedream 4

[quartz pike]

on the nanobanana scale from nanobanana 1. and nanobanana 2. i'd say its a nano banana 1.5

-# if it existed

-# but yk what i mean

[prime mulch]

Got it

[cloud zinc]

well thats they claim

[quartz pike]

oh lol

[prime mulch]

Still seedream 4 is best for facial details in img 2 img

[bleak lake]

Not at anything closer to nb2

[bleak lake]

Prompt understanding is low af

[prime mulch]

Seedream 4 is above flux 2 dev

[bleak lake]

But yeah flux strength is something diff always been diff

[limber crag]

have they fixed the error issue in nb2 btw?

[prime mulch]

No seedream 4 give me better results in facial features details but nb pro only give 60 to 70 percent of facial details sd4 gave me 90 percent

Reference image

[prime mulch]

Nb 2 pro

[prime mulch]

Seedream 4

See the face details

[limber crag]

nb is better right in what u sent

[cloud zinc]

nb better

[prime mulch]

I said preserving face details

[bleak lake]

Seedream hmm might be better

Nb 2

Flux 2 pro

[prime mulch]

Rather than that you can get celebrity faces better in nb pro

[modest prism]

Is Flux 2 better than NB pro on text rendering?

[native yarrow]

seedream is NOT better than nb pro man cmon

😭

[bleak lake]

Lmao seedream is not better than nb2

I read that as better than flux 2

[bleak lake]

nah

[cloud zinc]

not even they claim flux 2 is better than nb pro

[bleak lake]

Wtf

[cloud zinc]

nb pro is on another level

[bleak lake]

Nb2 is just phenomenal

[modest prism]

So what's flux good for?

[limber crag]

some people believe every new ai release is a sota

[native yarrow]

NB2 and flux is like comparing a grown adult man to a child

[limber crag]

some are allowed to be mid tier

[prime mulch]

I said nb pro better for me seedream preserve more face details than nb pro output wise nb pro is better

[limber crag]

hey whatever is good for your use case man not judging

[bleak lake]

Flux 2 pro

Nb2

It's not even close at text gen

[prime mulch]

No judging and im a big nb 2 pro user

[prime mulch]

Elon husk

[bleak lake]

😹

[modest prism]

Google just made AGI and we still don't understand.

[bleak lake]

Google got the best tpus

[cloud zinc]

nbp text gibberish?

[bleak lake]

nope

Nb2 text is perfect

[cloud zinc]

this feels like photoshopped in. so straight and robotic

[prime mulch]

Google has best resources ofc google is best

[bleak lake]

I mean with a simple prompt add a blackboard and write down trigonometry formulas

[prime mulch]

Bro need to add details in prompt

[obtuse heart]

Then ask Google to do smt with the font..? Google models require some details in ur prompt.

It's not like Claude where it adds like 900 other things

[bleak lake]

Essentially just need to type make it human like or similar

[keen beacon]

Gemini infinity loop

[prime mulch]

I think you just tell add some formulas if you add more details it gave more detailed results

[keen beacon]

Imma get it to 500k one out put

My buddy got it to 900k onnce

And it crashed

[modest prism]

We have smart boards nowadays which can be typed in them. So still correct

[bleak lake]

That's what I meant with a simple prompt.
If you want more human-like text you can prompt it

Seedream 4 max

Nb2

[prime mulch]

That's what i said

[bleak lake]

It's not even close

[bleak lake]

I agreed

[keen beacon]

But still

Really high res

[bleak lake]

yea

[keen beacon]

Make one of a everyday object

[bleak lake]

too vague give something detailed

[keen beacon]

1 sec

[prime mulch]

Brooooo

I was wrong

Nb pro was insane

[keen beacon]

It is

[keen beacon]

I was wrong too

[prime mulch]

Reference image

[keen beacon]

lol ya

[bleak lake]

It's even better when generating high res

That's what makes it good

It generates 2k and 4k like nothing

[keen beacon]

Gemini beat ChatGPT

Number 1 app

All cuz of a img gen lol

[bleak lake]

yeah in apple store

[bleak lake]

Also in google play store?

[bleak lake]

Actually image gen is of more use for avg audience than llms 😭

[keen beacon]

Not sure I would think so

[bleak lake]

I see

[keen beacon]

Have any u guys had convos disperse?

[bleak lake]

nope

[keen beacon]

Wtf

[native yarrow]

i agree with htat

[keen beacon]

Same

[native yarrow]

more people who dont even know how to use an LLM use the image model

[bleak lake]

Indeed, and for avg audience llm is almost what they want

Like we actually got the llm what they want

Good maths, problem solving , basic reasoning, basic coding, basic prompt writing

[keen beacon]

I herd same about video

When Sora came out

[bleak lake]

If you are using llm for more than that than you are not avg

[bleak lake]

It was same with Ghibli

[native yarrow]

ghibli?

[bleak lake]

When got launched Ghibli style

[native yarrow]

oh the ghibli style

[bleak lake]

Yeah

[native yarrow]

lmao that is so ass looking back on it

[bleak lake]

Real

😭

[native yarrow]

nb pro would CLEAR that

[keen beacon]

It feels good getting validated don’t

[bleak lake]

Surely does

[keen beacon]

Because I remember distinctly telling people that vision, models and image and video is extremely huge and it’s going to push that industry forward and the AI

But a year ago, this time people would laugh at me I remember that distinctly

For saying that, but I knew this moment would come

And probably because I was biased, and since I’ve always kind of been into Imogene ever since I pretty much started AI

[bleak lake]

I mean graphic designing is a huge market with about 50b market size

[native yarrow]

wasnt DALL-E the start of it all?

[keen beacon]

For me ya

[bleak lake]

yeah

[keen beacon]

My fav is Dalle 3

I just really love that model very comfortable with it

I wish they had something like that, but with like the capacity of a nano

[bleak lake]

image gen got huge potential like imagine the ability to disrupt a 50b market

[keen beacon]

It was really good at filling in the blanks and it really got your ideas down and would fill in the little details was awesome

[native yarrow]

how?

[keen beacon]

How what?

There was so much flexibility with Dalle

Erynpowerful model

Let me show u

[prime mulch]

Do you know any prompt template for generating images

[prime mulch]

I started with stable diffusion

[bleak lake]

I guess the least developed area of Gen AI is probably audio gen
@keen beacon

[bleak lake]

not really i just write them myself

I got some prompts saved in

My pc

But I am outside rn and on phone

[prime mulch]

Same here but I feel if someone sort prompts like this for nano bana this for seedream like that we can write better prompt

I usually tell my ideas to models to write prompt and i make some changes in that prompt

[keen beacon]

Nvm to many images

To go thru

II got no space in my iPad

U guys wanna c so,eyeing funny

Only 1 sec sneak peak

[prime mulch]

Use telegram to upload that images

Telegram is free unlimited storage

[bleak lake]

https://nano-banana2.com/blog/nano-banana-2-prompt-library

Check this ig

[keen beacon]

Nice

[prime mulch]

I saw that 😂😂😂

[bleak lake]

😭

cooked

[keen beacon]

I got super rare image no one ever seen

It would shock u

[bleak lake]

Show fr

[prime mulch]

Use telegram bruh

[keen beacon]

It’s nsfw but not the kind u think lol

[bleak lake]

Gore?

[keen beacon]

No

[bleak lake]

hmm

[keen beacon]

Nothing werid

[keen beacon]

U should see the other things I git

Whole things growing on people’s faces

Its halilrous

[prime mulch]

I bet you use local llms

[keen beacon]

No

[prime mulch]

For nsfw

[keen beacon]

ALL DALLE

NEVER

[bleak lake]

What

[keen beacon]

Ya

Really easily too

lol

[prime mulch]

Send me link of the model u use

[keen beacon]

Dalle

That one that was in gpt

[bleak lake]

DALL E 3

Right

[prime mulch]

Cuz gpt forgot dalle

[gusty saffron]

Hi everyone, does anyone know what kind of model 'raptor-1123' is?

[kind spade]

hi bro.
I dont know about it well, but when it comes to ai models, I am really interested in them.
and i have rich experience in several works related to ai models, like training, fine-tuning and deployment, etc.
can we discuss about it more detail?

[keen beacon]

Guys

Nano gunna get nerfed soon

https://youtu.be/RmmrVCUGYp8?si=Lm-kIxiwpfWyMBHd

Now Google will be in the hot seat lol

[atomic lagoon]

No its not lmao. Anyone who watches news for stuff like this I find brainless

[keen beacon]

😩

We probably got till after thanksgiving

[native yarrow]

FK

[keen beacon]

Ya

He said the golden words at the end of that

😔

But I did just realize that now that Google is in the number one spot and the number one seat I wonder how they’re gonna handle the pressure when things go wrong.

[limber crag]

its google

nothing will happen

they got synth id

[grizzled yew]

hey hey, I only came by to ask why they took the code rankings off the leaderboards.

[keen beacon]

You made my night thank you

[limber crag]

😂

[keen beacon]

I’m serious. I appreciate it.

I’m over here cheesing

[native yarrow]

lolz

google has rarely been in a genuine hot seat ngl

[grizzled yew]

or, did codestral just dominate it so hard, there was no reason to list it

[native yarrow]

synth ID saves them

if NB pro didnt have synth id itd be nerfed within DAYS

[keen beacon]

😞

Are any of you guys into like aviation or airplanes by any chance?

Or anybody like that show Mayday air disasters where they do the investigations of airplane crashes?

[native yarrow]

i like planes :3

i think synthid is down rn lmao

[keen beacon]

But that don’t stop the damage happening when it’s happening lol

Lmao

https://tenor.com/view/i-can-breathe-now-pamela-the-great-canadian-baking-show-i-can-rest-easy-now-i'm-no-longer-stressed-gif-2728018860866726529

I gotta go guys so I gotta work tomorrow if I don’t get to see you guys have a good Thanksgiving

[native yarrow]

holy wow gemini 3 is absolutely insane at narrative writing

finally did my first test

[atomic lagoon]

Awesomeee

[native yarrow]

bro u gotta look at my dayum dms 3:<

[atomic lagoon]

My bad I was at a thanksgiving family thing so I couldnt see

[native yarrow]

lmfao its good im playin

[flat heart]

wagwan people !? glad i found you

[native yarrow]

nb pro not workin on lmarena brah

[tribal bay]

is there any difference between flux-2-flex and flux-2-pro ?

[subtle lynx]

Close-up action shot of chefs adding tomatoes, green chilies, and spices into a sizzling karhai, rich textures, steam rising, golden restaurant lighting, background blurred but showing premium décor and glowing sign ‘M9 Cafe & Grill’, slow-motion feel, cinematic food commercial style.

[lofty frigate]

when i generate a new image with a new prompt it gives me a complete random photo, is it me or its worse i mean i get there is an edit but its not really as good

[rigid holly]

Are the buttons going to get fixed?

They Can't delete posts anymore. And the new chat button does not work either anymore unless i open a new tab with it

[wicked lotus]

Court

[ocean vortex]

new interesting benchmark, physics reasoning https://huggingface.co/datasets/CritPt-Benchmark/CritPt

Opus 4 to Opus 4.5 jump is crazy though. 0.3 to 5%. Looks like they actually improved a lot with that model, notable jumps in many other metrics as well

[eternal olive]

hi guys, what would u say is the most realistic photo generator?

[torn mantle]

it wasnt tested yet on the shared bench right

but overall they did improve the model yea

just the fact that its like much smaller but better is a big win

its also more efficient

uses way less tokens

could be served with a much cheaper price

but its not good for general usage, still lazy as always

• still behind on math and multimodality

[grizzled prairie]

New stealth model robin is better than opus 4.5

Crazy

[torn mantle]

what

https://x.com/btibor91/status/1993630817051201854

Tibor Blaho (@btibor91)

ChatGPT web app also has a new "ImageGenV2Banner"

"Make images with ChatGPT - Feature your friends, faster image generation, and more consistent editing."

X · Nov 26, 2025, 10:40 AM

[torn mantle]

where is this

cant find it

[grizzled prairie]

Lmarena

[torn mantle]

text only?

[grizzled prairie]

Must have code checked

[torn mantle]

oh

[grizzled prairie]

I get it every 2-3 chats.

[torn mantle]

interesting

let me test

[grizzled prairie]

I tried some tamagotchi and falling sand clones and it was better than any other model.

[torn mantle]

did you take screenshots

[grizzled prairie]

I have chat history.

[torn mantle]

or u have comparison

[grizzled prairie]

Here is Robin:: https://019abf95-548a-728a-900f-43e3cb0c6963.arena.site

[torn mantle]

yea i just got it

feels like oai model

for sure

[grizzled prairie]

Opus 4.5 https://019abfbe-9f0a-7d75-8b0b-cf0333dc38a1.arena.site

Yeah. It’s 100% OpenAI model

[torn mantle]

better

its slower too

[dim basin]

what did you use to make this?

[torn mantle]

robin is better than gpt 5.1 codex but not better than opus 4.5

[grizzled prairie]

In my use case, it followed the prompt better

[latent crest]

What does State of the Art ( SOTA) mean?

[dim basin]

like the best

or the newest

[latent crest]

Ohh

[dim basin]

for example, State of the art image generation with nano banana pro, means like the best and newest of its kind image generation with nano banana pro

[latent crest]

Who created the name sota ??

[dim basin]

im not sure

[steel torrent]

It's literally an acronym: State Of The Art

[latent crest]

Yeah I know, was wondering which one said from now on we call the newest model state of art

[torn mantle]

best of the best

[steel torrent]

It's simply because the best current model(s) is the state of the art. The term state of the art is decades old, and has specific meanings in relevant technical and legal fields.

[latent crest]

Do u think lmarena will have an app sooner or later?

[dim basin]

the website works fine

an app would be nice though

[latent crest]

If I understand correctly , some models have code names right? Like “autumn” ?

[native yarrow]

yess

those are stealth models

[latent crest]

stealth?

[steel torrent]

The developers don't want to disclose that they have/are preparing to launch a model, but want feedback; so they put it in arena under a codename, like "Nano Banana".

[latent crest]

Ohhh and u guys can actually find out the model? Like guessing it right

[native yarrow]

sorta yes

[latent crest]

Which do u think Autumn is for example ?

[lofty frigate]

Is there even a point in voting in image generation anymore

Comparison

[torn mantle]

oh robin is actually good

[fickle venture]

Yeah it's like the one you enable on chatgpt called Search so it's only job is find info on the internet

[fickle venture]

I mean it's very important when some models are out, right now Nano Banana Pro is the top 1 so I am pretty sure in a few weeks another open source ai will beat Nano Banana Pro

The heck is robin

[torn mantle]

oai codex

5.2

[fickle venture]

Openai?

[torn mantle]

or something like that

yea openai but not sure which ver

if its like codex high high high max max

or like 5.2

[fickle venture]

Oh it seems to be good at UI

[torn mantle]

yea they did improve on that for sure

but its still favoring dark colors

same color palette

i wish they change that

[latent crest]

Are DMs allowed here

[torn mantle]

gemini 3 pro attempt

when was this model nerfed................

wtf is this

[native yarrow]

lowkey it mighta been

[lofty frigate]

I know but the problem now is when voting and writing a prompt to generate the next image it just generates a random image and not based on the image i voted, in order for that i have to click on the edit

[hollow imp]

@prime mulch doesn't think so

[prime mulch]

Ofc like seedream

[prime mulch]

China will steal the algorithm and make it opensource

[torn mantle]

this battle is actually interesting

like between robin vs gemini 3 pro vs opus 4.5

[hollow imp]

Robin?

[torn mantle]

yea its a new stealth model

but its funny tho

[fickle venture]

Some models doesn't have edit mode so of course it will make random images

[torn mantle]

oai are just sitting on good models like that

this robin model is like their real hidden card imo

last codex update was just an appetizer

but it does take a lot of time tho

makes me wonder if its just actual codex + more thinking

they still need to work on UI/frontend and color choices tho

[lofty frigate]

Literally all the model that i got have the edit on top

[torn mantle]

maybe i gave it too much credit

[weary galleon]

What means "Hard Prompts" in Leaderboard?

[atomic lagoon]

Yo what

[atomic lagoon]

AND I JUST CANCELED MY GPT YESTERDAY

I KNEW THIS WOULD HAPPEN

[thin current]

After only 3 prompts, it gives me this message:
"You have reached your rate limit for claude-opus-4-5-20251101-thinking-32k. Please try again in 50 minutes."

Absurd

[atomic lagoon]

Yea its crazy lmao

[torn mantle]

this probably has to do with codex tho

[atomic lagoon]

I believe its because of the thinking

[torn mantle]

seems faster at editing too

[atomic lagoon]

Codex comes with pro plan, no matter though because I got until the 13th of december until it expires

[atomic lagoon]

Omg dude

[thin current]

Unfortunately it does it with the others too, at this point it's become impossible to use them

[atomic lagoon]

And I thought opus was really good and this is supposed to be better?

[torn mantle]

robin ( codex 5.2 )

did they improve or nah

[atomic lagoon]

Make a new account withj temp mail

[torn mantle]

codex 5.1

[atomic lagoon]

WHAT IN THE HELL

[thin current]

Yeah, I'll have to do it, but it annoys me because I had already prepared all the prompts

[atomic lagoon]

Asura how did people use this? Was it leaked?

[torn mantle]

they cooked with this model ngl

still has some cons but its waaay better than codex 5.1

[atomic lagoon]

Wait is that an app or?

[torn mantle]

spare me with the svg test bs pls

[torn mantle]

i told it to generate windows webapp

[atomic lagoon]

Good lord

[native yarrow]

one prompt?

[torn mantle]

no two

[native yarrow]

ooo

still good

[atomic lagoon]

TWO?

LMAOO

[torn mantle]

2nd prompt i told it to improve color palette

but the 2nd was just to improve the colors

[atomic lagoon]

2 is insane for that level

[torn mantle]

also the editing is like the fastest ive seen

[native yarrow]

how fast it makencode

code

[atomic lagoon]

How did you use it?

[native yarrow]

[atomic lagoon]

Is it just codex 5.2?

[torn mantle]

im guessing its 5.2

its an updated version of codex

[native yarrow]

did the gpt model get an update or just the codex

[prime mulch]

Can we download code from lm arena code mode use that in vs code?

[torn mantle]

but idk what they will call it

[torn mantle]

yea

top right

[atomic lagoon]

Yea but how is it used, is this a model not everyone has?

Must be, because its on 5.1 currently

[torn mantle]

you can try it on lmarena battle mode

[atomic lagoon]

Ohhhhh

[torn mantle]

its a stealth model not available yet

[native yarrow]

man i swear ai is prob the most competitive market rn

[atomic lagoon]

LMAO

Probably

[native yarrow]

it hasn't even been like 3 weeks since gemini 3

:P

[atomic lagoon]

Its been like 1 week since gemini 3 LOL

[torn mantle]

they have to tbh

with gemini 3 and opus 4.5

[native yarrow]

yes lolz

[torn mantle]

oai needed that

[native yarrow]

this is open ais prob

last

gasp for air

[torn mantle]

yea 200% oai

[native yarrow]

i wonder if they're tryna catch up with image model too

[atomic lagoon]

Doubt it

Well

They might honestly

[weary galleon]

Rodin is Gemini 4

[atomic lagoon]

How do we know this?

[native yarrow]

lolz nah

[atomic lagoon]

Or is it speculation lol

[weary galleon]

It's easy

[weary galleon]

Not speculation, just fact

[atomic lagoon]

Right but how would you know this?

I just want to know

[native yarrow]

he's joking prob

[brisk turret]

When opus 4.5 data? I really need it

[atomic lagoon]

Wdym

[weary galleon]

I don't see any robins.

[atomic lagoon]

Asura said its in battle mode and stealth model

[weary galleon]

What means "Hard Prompts" in Leaderboard?

[atomic lagoon]

Found it I guess lol

[weary galleon]

Minimax is better

[torn mantle]

its not bad right

[atomic lagoon]

Not at all, its actually really good

I've never seen an AI off one prompt add screen shake effects on something like this either LOL

[limber crag]

why is claude sonnet rate limit so high?

only 3???

[atomic lagoon]

You're just rage baiting

[atomic lagoon]

Its the same with opus

Kinda lame

[fiery gull]

imagine minimax m2 agentic mode (focus of minimax m2)

[limber crag]

weird man this website has become unusable over the past few weeks

[whole swallow]

Yeah bro I did.. a couple of times..

It was because they used some filter to make it kinda blurry.. and it was super short

[latent crest]

Is @echo aurora The owner of this site ?

[weary galleon]

What means "Hard Prompts" in Leaderboard?

[rocky mauve]

I can’t keep up with all these new models, new better ones everyday at this rate

[atomic lagoon]

I know right? I honestly shouldn't buy a model right now

Because there will just be another one

[fleet lintel]

is OAI planning to do 12 days of christmas this year?

[dim basin]

guys what the hell is up with nano banana 2.5 flash

never asked for any text to anything, it was doing this before and i reloaded the page and tried again and it still says this

[round island]

just specify the text

[dim basin]

i did

i said after, no text, and it still put some random text

[high ginkgo]

what model is this

codex 5.1?

@torn mantle

[marsh violet]

Hellow...

[fiery gull]

hi :).

[robust dome]

why does this keep happening with me in LM arena

[lofty frigate]

Not just you

[ocean vortex]

"much smaller"?

I don't think it's smaller at all

But, obviously, they had big margins to play with

when it comes to pricing

They are gonna make much more money in total by lowering the price

as opposed to keeping high price and having much less active users

Also, Sonnet 4.5 released barely 2 months ago

[latent crest]

This happens to me too

[ocean vortex]

it would have been near impossible to improve that same size model to this level by now to name it Opus. It isn't really smaller than the earlier Opus I do not think...

[nimble tartan]

Yo chat

Can we upscale here or nah

[neat apex]

i doubth a lot that Opus 4.5 is actually smaller, its likely an discount due to the new 64k reasoning upper limit

like, it looks like how it should be, its at very least a great optimization

and they dont increased the limits in the plans yet by what i know

[hasty sky]

Hello

[hollow imp]

@echo aurora put some words in automod so that these bots cannot send their message

[hollow imp]

Hey dom please new system prompt

[echo aurora]

Not a bad idea, I'll consider it.

[modest prism]

Opus 4.5 costs the same as opus 4.1 for anthropic. They're just losing money to compete with Gemini 3.

[stray aspen]

There will be a point where we won't be able to talk at all lmao

[atomic lagoon]

Its robin, a stealth model in battle on lmarena. Apparently its codex 5.2 but I've also heard its gemini 4 which I find hard to believe

[queen veldt]

Gemimi 4 is crazyy

[atomic lagoon]

The guy who said it kept rage baiting so it was probably fake lmao

[queen veldt]

We won't be getting that until next year

OpenAI is panicking hard

They addes shop research and voice mode in chats

That voice mode sucks a lot

[leaden sun]

a few people are speculating opus 4.5 to be originally sonnet 5/6, anthropcíc made this into the opus class same like gpt-5 happened to save costs apparently, when i heard that it resonated with me for some strange reason cause opus 4.5 indeed sounds more like sonnet than the previous iterations

[upbeat dune]

This happens every 30 minutes, why?

[torn mantle]

one thing we are sure about is the tokens efficiency, it uses like way less tokens compared to 4.1, but im also guessing they've done some other optimizations (advanced caching) + they had new investements deals

[torn mantle]

guess im right

robin = current codex + more thinking time

which is sad tbh

oai are so desperate

their KPIs probably went crazy on gemini 3 + nb pro + opus 4.5 release

wasnt it like the 1st just yesterday

what happened

are people confusing the viral image generated by nb pro for chatgpt image model?

[ancient locust]

hey!

https://tenor.com/view/dead-chat-dead-chat-skeleton-gif-25954239

[torn mantle]

it did improve tho

this was on 17th