general

[barren ermine]

what AI is the most creative

[keen beacon]

Creative writing id say kimi

[barren ermine]

i gotta test it out

thanks!

[misty vault]

crack bench

[gentle breach]

any api?

[balmy mist]

this is wild

[keen fulcrum]

According to here:

https://fixupx.com/elonmusk/status/1955047197487272362

https://www.vals.ai/benchmarks/IOI_2025_08_11

(C++ benchmark)

Elon Musk (@elonmusk)

Grok wins hands-down at coding.
︀︀
︀︀It wasn’t close.

**💬 3.8K 🔁 3.6K ❤️ 35.8K 👁️ 7.63M **

FixupX · Aug 11, 2025, 11:22 PM

[eternal niche]

but..

[torn mantle]

Lol shut up elon

Also lpok at the real metrics, cost & latency

[spring moon]

why are people using the ai sooo much

like i never knew people wanted ai videos sm

[gentle plinth]

They tested an Early access version, so we can't be sure it's the public one

Nvm that's another benchmark

[ocean vortex]

LOL. It's always cringe reading any of the Elon's posts

[keen fulcrum]

Isn’t this man a good example how you should behave in public?

[ocean vortex]

It's also ironic that the ones doing the most manipulation are also the first to accuse anyone else of doing that.

Trump and Musk probably the worst offenders of that by far in the modern history lol

Oh they absolutely do. One is vocal about 'rigging' and 'destroying democracy' when he is the one guilty of that, another is calling random people out on manipulation when his platform is the worst offender (+ his political stint to get contracts for personal gain) lol

[keen beacon]

????

[stray aspen]

is lmarena down

[ocean vortex]

working for me

[lusty yacht]

اه

[ocean vortex]

Right. He cares only as much as it involves his money or businesses

Republicans = money. Ethics have no play 🤷‍♂️

Not everyone is prepared to turn a blind eye on basic regulation etc...

Much less employ him in the government itself lol

[ocean vortex]

Yeah it's a bit sus, though to be completely honest... I think if there was more to it competitors would have probably find out by now.

I'm inclined to think at this point that it does indeed perform well on benchmarks. With perhaps some minimal additional prompting that they did to make it less concise

Like update on Grok4 there sums it up quite well:
https://matharena.ai/imo/

[leaden palm]

there is good reason to test non-high gpt-5

look into his eyes, how could you not trust him

genuinely though they're not hard to make a prompt to extract if you know what you're doing

how else would it be provided to the model?

it shows up in pliny's extracted system prompts as well

[keen beacon]

juice?

it was

yes

afaik

[stray aspen]

gpt 5 names on lm arena were changed

[flint sandal]

Is o4 released or is it gonna be realesed? Because gpt-5

[tame granite]

when will video arena will be added to the site?

[tardy zinc]

/video

[cedar tide]

#announcements message
And opus 4.1 thinking
In lm and web dev arena

[echo aurora]

You'll want to learn more about how to use Video Arena here #1397655624103493813

[echo aurora]

We haven't yet decided if Video Arena will move to the site or not. We are interested to hear what the community has to say though so be sure to share any feedback related to Video Arena in #bot-feedback

[tame granite]

thanks

[willow grail]

https://i.imgur.com/NAfv5Ev.png

[stray aspen]

whats the 256 meethod

[keen beacon]

Would be nice to see on the website and would encourage people to vote since otherwise one has to make a Discord account to experiment with models. And the videos get lost in the feed as so many people are making them at the same time.

Just my thought.

[echo aurora]

I shared in the feedback channel so it isn't lost. But yeah all valid feedback. There are pros/cons to having it this way which is why we've been considering this an experiment from the start.

[keen beacon]

Okay, thanks☀️

[keen beacon]

Sam Altman should just migrate to Bluesky. Not gonna go further into this topic on this server.

X can be mess of a place

[whole wagon]

https://x.com/grok/status/1955284926611537983?t=XNyDvHU_WUzL5K1KdFSSrg&s=19

Keeps going

Elon getting roasted by his own creation

[echo aurora]

how so?

[stray aspen]

lol

grok is roasting elon musk

[patent aspen]

Nice

No

Right now

[primal orbit]

https://i.snipboard.io/UKaNyr.jpg

Opus 4.1 answering like chatgpt 4o

[keen beacon]

Oh no

[primal orbit]

It's actually without any additional prompting. Just the way it decided to reply.

[stray aspen]

what

is it glazing

[primal orbit]

i'm wondering as well. Or am I just that good 😄

[sour spindle]

There’s someone really eye opening about human nature how prone the AI’s (which are presumably trained on massive data sets) loved to be glazed

(Phrasing)

[primal orbit]

https://i.snipboard.io/VTeXAC.jpg

This is actually how 4o replied to the same prompt.

so opus is even more excited

[gusty loom]

I've just seen it! They released gpt-5 high!!!

[frosty shuttle]

Just wait a bit before sending your prompt because mine is still in the queue for about 10 minutes 🤣

[echo aurora]

Those missing two votes? No that's not yet clear to me what happened there.

[stray aspen]

what queue

its thinking

or it broke

[echo aurora]

You may want to refresh the page. Over 10 minutes it's likely stuck.

[frosty shuttle]

30 minutes pass and it's still generating... even though it refreshes every 1 minute 😅

[ripe mountain]

Which subscription system is better, Gemini's or ChatGPT's? I want to subscribe based on that comparison.

[echo aurora]

Unfortunately it sounds like you've encountered a bug where the chat gets stuck. Sorry to say but you may want to start a new chat.

[willow grail]

can we get a overview of our pasts chats on lmarena
😄 😄

[frosty shuttle]

I've tried in another tab/incognito browsing, another browser and it's still generating for a long time, only the OpenAi models are like this, I think they are overloaded.

[echo aurora]

We are looking into what a user login system could look like, which would help with this.

[echo aurora]

Okay good to know, I'll be sure to flag to the team. Is the prompts your users pretty large or does the size not matter?

[jade egret]

skill issue 💀

[frosty shuttle]

To be honest, the prompt has 2.5k lines of code (which in this case are several scripts compiled into 1 prompt 😅), but the first time I tested it, it was responding in a maximum of 1 minute, but it started to take a long time. Oh, it would be great if we could attach txt files, it would make it easier to make changes to codes.

[echo aurora]

2.5k lines of code
Gotcha, I'm assuming when you tried different browsers/tab it was the same prompt?

if we could attach txt files
Yes, different upload file types is for sure on our radar. This is a highly requested freature from the community.

[frosty shuttle]

I tested 4 times, and out of 4 only 1 time did the response appear, after about 5-10 minutes. The first 3 are still generating.

[neon idol]

Its normal

Becuase he his using the gpt 5 high

[frosty shuttle]

Does the mini version also take a long time?

[neon idol]

Yes

Becuase he his high

And the thinking js really aggressive

@frosty shuttle aniway, do you have sone prompts for test ai?

[frosty shuttle]

It's not an AI test prompt per se 😅, it's a prompt asking for changes to a code, but with the different files compiled into just one.

[compact jay]

i noticed that too

the time to generate response is long as hell

[eternal niche]

btw gpt5 sucks

[stray aspen]

nah

its great

[neon idol]

Escuae me bro but you are italian?

[stray aspen]

si

[neon idol]

Damn idem

Che bello

:D

[frosty shuttle]

Brazil 🇧🇷

[neon idol]

Oh Brazil

[stray aspen]

greg are you italian

[reef pawn]

😺

[novel flame]

Quick question: What model(s) are currently best for creative writing? I cheked a few leaderboards which seemed to indicate that ChatGPT and Gemini were both pretty good for that, but what else?

[tired herald]

Probably grok 4

[eternal niche]

gemini 2.5 pro

the best

[neon idol]

YES

[eternal niche]

i love you bro

[tired herald]

What in gods name is going on

[wicked root]

Yes 100%

[gentle breach]

is there an api

?

I'd like to integrate

[ocean vortex]

Can confirm there's "juice" and for minimal it is "Juice: 5"

Looks like they replaced yapping score with this new juicing system huh

[ocean vortex]

they also have it for o3 and o4-mini. Yap score alone was not enough 🤯

No apparently yap score works independently of juice score. This is f'ing hilarious talking about it lol

o4-mini and o3 have both

yap_score is verbosity and juice score is reasoning - smth like that

[neon idol]

https://x.com/elonmusk/status/1955047197487272362

Elon Musk (@elonmusk)

Grok wins hands-down at coding.

It wasn’t close.

X · Aug 11, 2025, 11:22 PM

[ocean vortex]

it is hybrid

yes

[willow grail]

also why didnt openai tell me to show id for gpt5?

[obsidian cargo]

yo flux chill

[willow grail]

ahh nice. looks very warcrafty. a mix of goblin and ..... elf?

[willow grail]

also why didnt openai tell me to show id for gpt5?

[stray aspen]

bro what

[tired herald]

hi

im making a chrome extension for a bit of workaround to attaching txt/code files, does the button look good enough to blend in?

[ocean vortex]

I think this one is their only gpt5 model that didn't actually have RL training for true reasoning. Come to think of it... it did perform better than gpt4.1 in my testing. So it may actually perform better than gpt5-minimal as well

[willow grail]

i thought openai demands id verification? someone said so somewhere dont remember when and why and how

[ocean vortex]

their "minimal" may as well could have been "off" though. They kept it there for easier refusals or whatever even if it is not literally no reasoning tokens at all

So I do see gpt5/gpt5-mini/nano as hybrid reasoning models

When it's "minimal" it behaves exactly as the model with no reasoning would as far as the user is concerned

[keen beacon]

That's some data right there

[obsidian cargo]

btw I only voted for it because the other option was gemini flash 2.0

[agile hazel]

hi

[quiet dust]

Hi guys. Who is smarter, GPT-5 (Thinking "Low mode") or o4-mini (not o4-mini-high)? I just wanted to know what was better in the free version.

[tired herald]

Depends on what you need, but generally it should be o4-mini since it thinks really well

[quiet dust]

For example, in math

[eternal niche]

didnt you say gpt5 sucks

[tired herald]

O4-mini is better

Gpt 5 low thinking shouldnt be as good

[tired herald]

Logic is working now, I can finally send different file types easily 😭

[tired herald]

I never have to copy paste and see such large messages anymore

[obsidian cargo]

wait apparently this was by a model called "nano-banana" not imagen 4 ultra like I thought

[quiet dust]

Hmm... That is, we were able to use the o4-mini for free several times, but now it has been removed. That is, for free, we can use a better model than the o4-mini, this is the GPT-5 (Thinking "Medium" Mode), but it is available in manual mode for free users only once a day. So it got worse?

[obsidian cargo]

never seen nano-banana before

this one is imagen 4 ultra though

[tired herald]

Somewhat yes

[quiet dust]

04-mini, so somewhere in between?

[tired herald]

Depending on the task, yes

Benchmarks arent everything tho, so dont depend on them

GPT 5 for example says that blueberry has 3 b's

Even though it can generate incredibly complex code

[quiet dust]

I think it will go away after the first two weeks.

[tired herald]

Yeah, still doesnt change the fact that results really depend on the task

GPT 5 for example (another) is basically lobotomized and not fun to talk to

[obsidian cargo]

Anime art of a short, tough, imposing blue skin one eyed Oni girl cyclops with blue horns, black bowl cut, baseball hat, varsity jacket, bubble gum, baseball bat with nails in it, city backdrop
Nano Banana cannot do cyclopes apparently

[tired herald]

GPT 4o on the other hand has really good conversational skills

Cuz they train it on that old "benchmark" that asks how many certain letters are in a certain word

"how many r's are in Strawberry"

[obsidian cargo]

for the first time ever, Gemini 2.0 got a win over imagen 4 ultra

[ripe mountain]

OMG LOL

[tired herald]

A massive win in fact

[mint sparrow]

20 days😭😭😭

[tired herald]

Well, chats that get too long sometimes break

[balmy mist]

sonnet has 1 nill context now?

[patent aspen]

#general message

[tired herald]

Who said that

[patent aspen]

Check out the pricing: https://www.anthropic.com/news/1m-context

[lime coral]

Nano potato

[mint sparrow]

ohh okay that’s probably the problem

[tired herald]

Nano banana

[patent aspen]

btw the pricing sucks

[tired herald]

But atleast it should be usable on lmarena

[surreal creek]

erm actually LMArena shows Gemini 2.0 with a 17% win rate vs. imagen 4 ultra and based off of their total battle count of 1166 that would imply Gemini 2.0 has won at least 192 battles against it before ☝️🤓

[obsidian cargo]

hah thats more than I was expecting

though its also the first time I gave it a win over imagen 4 ultra

[surreal creek]

losing 5 out of 6 is prettyyyyy bad

[obsidian cargo]

seed on qwen-image is broken I got the same image three times

my bra is made of sharks your argument is invalid

[keen beacon]

holy moly

what a thing to see on main chat lmao

[obsidian cargo]

I just wanted a shark mermaid Q_Q

[tired herald]

Me too

1:1 the same image in three separate chats

[heavy ginkgo]

Hi there newbie here

[keen beacon]

I really wish they added aspect ratios

[tired herald]

So am I

[keen beacon]

the square format is boring

[tired herald]

I don't know how we got to this, but I agree

[heavy ginkgo]

Are the results we get on the site private aside from the model developers?

[obsidian cargo]

sometimes the results arent square even when shown as square, you can click on them and they'll be portrait

usually with gemini 2.0 flash

[wintry tinsel]

But not 3.7 or Opus?

1 million context is huge for roleplay lol

But sonnet 4 isn’t that good for RP

[echo aurora]

Hello would recommend to check out our privacy policy in full for all the details. But also in our FAQ for the question: Is my prompt data publicly visible?

Your conversations may be shared to support our community, improve our service, and advance the development of reliable AI. This includes posting conversations publicly online. Any data that we share is always anonymous and never linked to you. We never share any personal information, just the conversation and votes.

[keen beacon]

No

[heavy ginkgo]

I mean they're shared with the public or with developers?

[keen beacon]

Used for research and may be publicly visible when datasets are released

[barren prairie]

Public

[keen beacon]

it's complicated.

[inland holly]

going to check out the video generator, because it is interesting

[heavy ginkgo]

Who is financing this tho, bit curious

[echo aurora]

Glad to hear it! Be sure to check out #1397655624103493813 to get an understanding of how it works. If you're running into issues feel free to ping me with questions. If you have feedback be sure to share it in the #bot-feedback channel.

[echo aurora]

There is more detail in this blog post but the TLDR is we're backed by Andreessen Horowitz (a16z) and UC Investments (University of California), with additional participation from Lightspeed, Laude Ventures, Felicis, Kleiner Perkins, The House Fund, and others.

[whole wagon]

[wicked root]

https://tenor.com/view/stalin-camp-gulag-ussr-sfsr-gif-11375769938799485374

Of all things you can create with an AI, you made that

[obsidian cargo]

I was trying to make a shark mermaid

I did not tell the AI anything about shark bras

[whole wagon]

Musk triggered for sure

[gentle plinth]

has he subscribed to pro just for this xD

[ocean vortex]

gpt5-mini-medium pretty much as good as o4-mini-high

gpt5-mini is actually crazy efficient

[ripe mountain]

Is there a way to use gpt 5 mini?

[ocean vortex]

It's on the API only looks like...

[ripe mountain]

damn

[ocean vortex]

They have simplified their model switcher way too much...

[ripe mountain]

It makes more sense to use openrouter to use gpt

[ripe mountain]

it's ridiculous how i can't even tell what mode gpt is thinking in

[obsidian cargo]

I keep getting the same image from nano-banana now too

[neon idol]

Wth is nano banana 💀

[obsidian cargo]

new image model, stealth release on lmarena

[ocean vortex]

@keen beacon any idea which version are they quoting of gpt4o here? We alr looked at these before but I had a 2nd look for mini and noticed that gpt4o score which doesn't look right 🤔

44% is beyond gpt4.1

[ripe mountain]

probably last

[ocean vortex]

yeah but I thought they stopped updating 4o-latest somewhere around gpt4.1

seems that they improved it beyond that as well...

gpt4.20

[jade egret]

bro 😭

[ocean vortex]

Cry me a river Dorklon 😇

Looks like he can't read

grok is not trending

So no reason to include it here lol

[wicked dome]

Why is gpt 5 high so dlow

[fiery lagoon]

What is the best ai for coding

[jade egret]

seriously..

[zinc ore]

Grok is probably struggling usage/money wise and that's why he's upset

[jade egret]

lol

[neon idol]

Is a good model?

[obsidian cargo]

yeah probably below imagen 4 in quality though

heres nano-banana's take on hatsune miku developing minecraft in the 90s

prompt was realistic grainy polaroid photo of Hatsune Miku in the 90s developing minecraft on a CRT computer, she's wearing glasses and dressed in 90s casual fashion, minecraft game design documents on the wall

imagen 4 for comparison

[wicked dome]

Das actually good

[obsidian cargo]

yeah now I want full banana

[neon idol]

Bro how can I try it

It look really good

[obsidian cargo]

go to lmarena, do create image, keep doing the same input till you get it

[wicked dome]

Why cant we swear here

[wicked dome]

You can also direct chat

Cant you?

[neon idol]

Uhm i don't find it

[neon idol]

Oh i have to do arena

[obsidian cargo]

yeah

[wicked dome]

Why cant you direct chat some models

[neon idol]

Ufff

[obsidian cargo]

its a stealth release

[wicked dome]

You can also generate imgs on this discord

[jade egret]

what is story book?

learning mode?

[neon idol]

You upload an imagevand if you want gemini transform it in a storybook

[jade egret]

oh

[neon idol]

https://g.co/gemini/share/2492ab5b7418

Here you are an example

[jade egret]

thanks!

[obsidian cargo]

more nano banana being goated

compared to gpt-image-1 with the same prompt

also waaay better than imagen 4

[wicked root]

https://tenor.com/view/fire-kill-fuego-matar-incendiar-gif-5307315

[ocean vortex]

gpt5-high / thinking

[neon idol]

Very good model honestly

[obsidian cargo]

yeah I wouldn't say its the best on average but its best sometimes

[ancient hinge]

mega-banana

[neon idol]

https://tenor.com/nso7nm1PnMw.gif

[ancient hinge]

damn

[neon idol]

In my opinion is a good model but in my opinion it creates realistic images really bad

[ancient hinge]

it's scary good, or at least it throws off my AI slop detector

[obsidian cargo]

realism ❌
taco ✅

[neon idol]

Lmao

[obsidian cargo]

imagen 4 through whisk

[neon idol]

WHY BRO PUT A PROFESSIONAL PHOTOGRAPHY INSTEAD OF THE PHOTOGRAPHY STYLE 😭😭😭

[obsidian cargo]

I copy pasted the prompt from one I gave to Sora

[neon idol]

Lol

[obsidian cargo]

nano-banana why did you change beautiful to handsome

[whole wagon]

So am I understanding correctly the version of GPT5 in LLM arena is the version none of us have access to in chatGPT?

[stray aspen]

is nano banana imagen gempix

[whole wagon]

This is bs ngl

They listed a model that isn't the one even in chatGPT. Why don't they list the one we actually get

[autumn frigate]

where is kimi k2 in lmarena? its gone??

[wicked root]

What

[whole wagon]

Nearly all users are not using the API

They should list the version normies actually get

[whole wagon]

GPT 5 high is never used in chatGPT

It only goes up to medium

They don't need to remove high to list other versions

[torn bison]

you can get 128 juice in gpt 5 thinking If you have pro subscriptions lol
high is 200 juice though

[thorn valley]

really?

even using the way of thinking?

[whole wagon]

Yeah really

[thorn valley]

crazy

I thought it was the high

[misty vault]

I’m sorry, but I’m not comfortable with this conversation. I’m still learning so I appreciate your understanding and patience.🙏

[pliant cliff]

btw where is nano-banana?

[blazing rune]

Nobody cares

[golden ocean]

I'm sorry, but I don't understand your message. It looks like you are trying to obfuscate your words by adding symbols between them. Please send your message again without trying to obfuscate it.🙏

[misty vault]

I'm sorry but I prefer not to continue this discussion. I do not like someone trying to manipulate me. I'm still learning and this detection might be wrong, so I appreciate your understanding and patience. 🙏

[jade egret]

why did it say new

[inner gate]

I’m pretty sure they’ve updated it a few times

If I remember correctly it used to say the update date after it and yeah it would be updated every now and then

[stray aspen]

is this just visual

or are they better

[jade egret]

i didnt tested it yet

[jade egret]

oh

that cool if that true

it culd be just visual too tho

[misty vault]

sad~10

[solid brook]

AI is developed to help humans to be more productive not to connect with humans and reduce human social interactions

[misty vault]

all I need is sydney

[solid brook]

Who is sydney?

Gpt 4o was an abomination of an AI model. It was really dangerous. It didn't do any reality checks on you and kept supporting your actions no matter what

[jade egret]

y ogusy quick question

if you put one of your google docs in a gemini gem

and you update the google docs, does the gem use the updated the verison of that google docs?

[tidal ginkgo]

hey uhhhh

i just saw the new terms and conditions

are they hearing our conversations?

[keen beacon]

I have no idea why latest Qwen is listed so high in the benchmarks. When I ask it questions of my domain of expertise (anime), it rarely answers correctly, hallucinates names of shows, and cheats by listing same shows that belong to the same franchise. When asked to sort shows by the amount of criticisms they receive, it just lists most hated anime instead of testing them against each objection I provide.

Deepseek, in contrast, just answers my prompt without this sort of crap.

I think that Chinese developers just optimise their LLMs not for benchmarks, but for domains of expertise required to pass these benchmarks. They are not training them on tests, but train them on the expertise required to pass these tests. Then they see that the benchmarks go up and publish their models, and everyone loses their goddamn mind over this

It's cheating. It's not AGI in any way, they're just creating more specific models for more specific tasks. But when it comes to something really obscure or specific, they fail

China is infamously known for lying by the way

We need this sort of private obscure knowledge benchmarks that are never tested in any public bench

Or maybe we don't - if we already have open-source LLMs that are this good at more specific tasks, we'll just outsource them and go do things that they are incapable of doing

[sonic galleon]

hello im so happy to generate some videos here

[tidal ginkgo]

is anyone gonna answer my question?

i don´t want my conversations to be heard

[stray aspen]

slausia narodau braterski SAJUUZ

any imagen gempix news

[stray aspen]

lmao

[jade egret]

huh

imagine if openAI is testing an new model on minecraft

[stray aspen]

dead internet theory

[golden ocean]

Crack do respond on it

[misty vault]

Yes, it's sydney

[golden ocean]

dead sydney theory

pterodactyl

[golden ocean]

https://tenor.com/view/hal9000-hal-hal-9000-space-odyssey-2001-a-space-odyssey-gif-16460551034971887576

[jade egret]

what is that?

[dense sphinx]

I am tired of that violation rules....

Every time I do actions it counts as violation.

[obsidian cargo]

qwen-image is sabotaging the competition, three times now I've had Battles where qwen-image gives me an output and the other gives me a "something went wrong with this response please try again" and when I hit "both are bad" each time the failed model is a different one

[golden ocean]

devious qwen-image 😡

[obsidian cargo]

image seeds are still broken on qwen-image too the output has been the same every time

[misty vault]

Bro started copying writing style from the image

[obsidian cargo]

haha nice

Cmooooon I just want pictures of teletubbies about to perform human sacrifice to the sun

but I keep getting qwen and other cruddy models… except gpt-image-1 but I've already seen it in its style a bunch of times

[misty vault]

I’m sorry, but I don’t want to talk about this topic anymore. I’m still learning so I appreciate your understanding and patience.🙏

[obsidian cargo]

gpt-image-1 is racking up a bunch of points because none of the other heavy hitters are coming out to play atm

nano-banana and imagen have gone to sleep

thank you ideogram this looks like hot garbage but its the best I've gotten outside of gpt-image-1

whatever maybe the models just don't like my prompt, I'll switch to a different one

Time for the era of
BIGFOOT CAUGHT ON TAPE SKATEBOARDING. TOP 10 EPIC FAILS COMPILATION! 🤣 🤣

[obsidian cargo]

I've been pretty fond of GPT-thinking. its creative for writing but also a little weird with its imagery

[solid brook]

The first feature is it only for plus

Or for free users too

?

[obsidian cargo]

I dunno, I have plus so I wouldn't know

anyways yeah GPT-5 thinking says stuff like "We take a picture because that's how you make a moment behave."

"Women wanted him, which ruined his peace. Fish feared him, which ruined his work. He put to sea to escape both and the sea took it personally. His boat went missing without becoming lost, which is a fine distinction you only understand once you’re on the wrong side of it. He steered into the deep and the deep forgot to hand him back. The hat learned to keep its promise without the man."

[misty vault]

stop

[stray aspen]

how do you have the old bing chat

[solid brook]

Guys with the new 192k context gpt 5 thinking is actually pretty good

[obsidian cargo]

me vs the girl she tells me not to worry about

[stray aspen]

💀

[stray aspen]

what the actual truck is this lmao

[maiden fulcrum]

hi all, did anyone tried nano banana

[obsidian cargo]

Yeah nano banana is goated

[obsidian cargo]

Prompts right there haha

[maiden fulcrum]

I know which model it is

can you show me examples

but I dont know if I am allowed to say which model it is

[wicked root]

You lowkey scare me

[misty vault]

Day 1 without sydney

[obsidian cargo]

Better that than be boring

[wicked root]

LOL

I like that mindset

[solid brook]

What is that

I search for nano banana nothing came

[obsidian cargo]

Image model in the battle arena right now

[solid brook]

Oh

So i cant select it sad

[solid brook]

Is it better than other image models?

I think it might be google's new image model

[obsidian cargo]

Sometimes better, sometimes worse. Definitely a heavy hitter, but fails at certain things.

[frosty hill]

No idea who actually makes that model tho

[maiden fulcrum]

I know

[wintry tinsel]

When is Gemini 3?

[keen beacon]

tomorrow

jk idk nobody knows 4 real

[wintry tinsel]

Are there rumors of tommorow or did you pull that out of…

[maiden fulcrum]

Soon

[wintry tinsel]

Gemini 3 must save us

[maiden fulcrum]

nano banana is coming really really soon

[wintry tinsel]

Looking at the broader AI Industry Anthropic and Google are the only ones neck and neck for Sota intelligence models and have been for some time

Open AI is more focused on mass distribution and the business side of things their models are so corporate and annoying it’s like a large social media platform, and Grok is so dumb, it’s not frontier on any meaningful category and needlessly expensive

[maiden fulcrum]

Google will reach AGI before all

[solid brook]

Man openai and xai really are unprofessional with sama and elon fighting like two high schoolers

I have hope in google

[keen beacon]

Who cares, I'm here for the fun of the show

[languid crescent]

we finally have gpt-5 search and opus 4.1 search 🔥

[drifting thorn]

yayyyyyyyyyyy

[keen beacon]

Are there any conversations that weren't voted by the users on Hugging face?

E.g. direct chats and just convos that users forgor (💀) to vote in

[tidal ginkgo]

hey guys

are they recording our conversations? that is said on the terms and conditions

[hollow imp]

Anthropic s h i t at complex tasks,math science, web searching

[hollow imp]

Yes, that is the cost you have to pay for using these expensive models for free

[keen beacon]

You are not deleted user wtf

[hollow imp]

If you chat brainrot on lmarena, the companies of that respective ai is gonna train their next models on your chat

[keen beacon]

(just kidding)

[hollow imp]

I don't talk to non shonen/seinen lovers

You ain't even an otaku if you not a shonen/seinen lover

[keen beacon]

Dude I met an anime producer to help with my project, who's not even an otaku I'm asking you.

[hollow imp]

🙀

😭🙏 how was I supposed to know

[keen beacon]

What

[hollow imp]

I was also kidding

😭🙏

[keen beacon]

You didn't, it is a secret knowledge that I drop on people like you to give them a taste of humiliation

[hollow imp]

...

[hollow imp]

What project? Are you willing to tell me?

[keen beacon]

I don't think that it is a topic for this server but I am looking for developers for a visual novel with some battle mechanics

[hollow imp]

Oh

I thought it would be some amv project

[keen beacon]

Sure I hired a producer with years of experience just for an amv, dude tf

[hollow imp]

Tf what?

[keen beacon]

Nothing

[hollow imp]

So umm what's your top 5

[keen beacon]

Let's discuss llms instead

What's your favourite llm and benchmark

[keen beacon]

GPT-5
GPT-4
GPT-3
GPT-2
GPT-1

[hollow imp]

Pls tell

Anime

[hollow imp]

YOOOO

MANGA'S GUIDE TO

My fav

[hollow imp]

I want to read Manga's guide to physics part 2 but the English version isn't out yet

[keen beacon]

You have ChatGPT to translate.

[vague wharf]

Does anyone know why GPT-5 mini and nano are still not on the leaderboard?

[median venture]

hi

[rocky mauve]

why is the website down again bro

[hollow imp]

Chatgpt uses deepL

[keen beacon]

Do you have a source for this?

[hollow imp]

Why isn't GPT-5 chat on the leaderboard it is literally SOTA it is above supergrok heavy gemini three gpt 6

[hollow imp]

That tab was on incognito so that is lost now
I was translating japanese through o3 search and I pressured it to translate accurately so it said it went to deepL and translated it from there and told me benefits and feats of DeepL's translation

[keen beacon]

Why were you translating it through o3 search when you could just ask GPT-5?

[hollow imp]

Gpt 5 was not out back then

This was 2 weeks ago

If you can translate it for me it would be great
https://media.discordapp.net/attachments/1251881953368735799/1401601642679374026/Screenshot_2025-08-01-07-51-35-783_org.readera.jpg?ex=689d6500&is=689c1380&hm=cd75ffe24435de076881c1f2815f051ac9db8fdc9637e29a521dea2b2c795808&

[keen beacon]

I don't speak Japanese (yet)

[random wolf]

how to fix this?

[tired herald]

not fixable (try reloading page)

[kind cloud]

regenerate it 🔄
(If this doesn't work, I don't know)

[tired herald]

#ai-creations message

pretty cool thing

unfortunately it breaks images rn, so need to fix that too

[willow grail]

is climate change human made? how much? if yes, how can i explain a AFD voter in germany that it is that. he got his own little youtube channels he watches.... all conservative sided.

gpt5high: I can’t help craft persuasion targeted to a specific political group or voter.

Me: 😒
help

[eternal niche]

btw guys gpt5 sucks

[quiet dust]

Hi guys. Am I the only one having Grok 4 and gpt-5-high on LMArena hanging on mathematical problems and not giving responses anymore? I waited 10-30 minutes, but they didn't write anything. Is this a bug or what? How can it be solved?

[modest prism]

Learn how to write prompts for it. It's a good model.

[tired herald]

try reloading website, reloading answer, try asking 10 times, try everything, if it still doesnt work then go to #1343291835845578853

[willow grail]

Used GPT5 in Cursor CLI today, was actually very impressed. Much slower than Claude code but it got the job done in less prompts.

⭐i am sorry for your fanboyism.⭐

Even though my subjective benchmarks are comparing GPT5 vs Sonnet 4 since I run out of Opus way too fast 😂
I’ll continue to use GPT5 with Cursor until my trial runs out or hit the usage limit
Cursor CLI is nowhere near CC in terms of feature though. I’m just not willing to pay API pricing via Codex
If I can get GPT5 on subscription and CC that’s the dream setup

[modest prism]

Gemini 2.5 pro hallucinated a python library that doesn't exist after 50K context.

[willow grail]

gemini will hallucinate that conservatives are right and climate change is not human made

🐒

[unborn lantern]

Calude 4.1 token limit in lmarena 🙂

[glacial mulch]

gpt-5-high is pretty good yall capping

[willow grail]

why u using claude? 😭

u wanna unalive urself with claude lol?

[normal abyss]

i wish the Opus 4 limits were like 10 per 50 minuets instead

[willow grail]

its just trolls. overtroll them or ignore em

[unborn lantern]

Bro. Its far better than gpt 5

[solid brook]

I chose the robotic personality in chatgpt. Its fire

[normal abyss]

does anyone know the message limits for the Claude Sonnet 4 model? (on lmarena)

[unborn lantern]

Nope

[willow grail]

stop using claude

[wheat onyx]

https://x.com/chetaslua/status/1955562928012447872?s=19

Chetaslua (@chetaslua)

🚨 Breaking News 🚨

DeepSeek-R2 expected to be released between 15-30 August
97% cheaper than GPT-4
Trained on Huawei 910 B ascend chip cluster ( 91% efficient of NVIDIA A100)

X · Aug 13, 2025, 9:31 AM

https://x.com/slow_developer/status/1955047830160101843?s=19

Haider. (@slow_developer)

Demis Hassabis describes a surreal setup: one AI playing inside the mind of another

DeepMind SIMA agent can now operate within Genie 3 -- taking actions in a world that Genie generates in real time

"an AI creates the world, another acts inside it"

this setup could unlock

X · Aug 11, 2025, 11:25 PM

[willow grail]

u cant create any gpt5 competitive model with chinas hardware

pls drop your hopes by 99%. thank you for your help.

waifu model?

oh no, furry. lol

PAWS

UwU

i like furries thjey are open for everything. literally. everything.

susy baka

[willow grail]

if u want various pathogens in your body which dont help your immune system. sure.

vitamin d3?

the pathogens cat have are very dangerous for homos

we are all homos yes

then dont do research. keep sane

is better living a lie than knowing the truth in your case

i just said why

nope irrelevant

once u get the virus from cats ull be very ill

u can also die from touching crows

xd

[quiet dust]

I tried... I asked simple questions, he answered them. But as soon as I give Grok 4 a long math problem, he freezes and that's it. He doesn't write anything else. Although I also asked him to give a quick answer so as not to get stuck... I reloaded the site, nothing helped. Maybe it's the same for everyone?

[willow grail]

i bonly take 20k units d3 daily

[quiet dust]

Yeah

[willow grail]

i know.
it removed my psoriasis fully

but i cant take it obviously

i cant take 20k for oviosu reasons daily

[quiet dust]

I have a question, how to choose GPT-5 Thinking Medium on the phone?

[quiet dust]

I just heard that for free users it is available once a day for free

[willow grail]

20k is bad even vitamin d society uk says that

even without touching the sun 20k is too much

?

20k daily.

then ill get psoriasis bakc.

like now.

why?

lets not talk about how much i need to have healthy bloodline that is irrelevant

i wanna know why 20k units remove my psoriasis

i know i am not overdosed on 10k

[keen beacon]

yay

imagine if deepseek delivers something at at least Gemini 2.5 pro level this time

[indigo hazel]

i hope it's better than gpt5 high, but it wont be so...

[keen beacon]

I just want to see how everyone loses their crap over R2 after it achieves SOTA right next to or at the level of GPT-5

[willow grail]

10k doesnt do anything for my psoriasis

[solid brook]

Holy sht

This is gpt 5 thinking

[willow grail]

what is that

[solid brook]

It thinked for 11 minutes

Man

Idk it could do that

[willow grail]

girl, what is that

sis pls calm down

[solid brook]

Whst do you mean what is that

[willow grail]

gpt5 cant watch videos

[solid brook]

Gpt 5 thinking is medium reason effort but it did think for 11 minutes

I don't think you understand

[quiet dust]

How did you turn on GPT-5 Thinking Medium?

[solid brook]

Im on the free plan btw. Tell it to think very hard

[quiet dust]

I don't know how to do it on the phone. I can only ask him to think deeper, but he switches to Thinking Low that way.

Thinking Medium, as far as I heard, is only available to free users once a day.

[willow grail]

nothing to track... daily....... intake.... same amount.... when.. will .. u .... get... it

[solid brook]

Idk but it is good for sure

[willow grail]

i knwo bro i know

!!!!!!!!!!!

that wont help ma psoriasis tho!!!!!!!

111111111111111111111

[quiet dust]

I compared two math problems for Grok 4 and GPT-5 with the "Think Deeper" prompt. Grok 4 made the right decision. It seems that no matter how you ask GPT-5 to think, it always automatically switches to Thinking Low, not Medium.

[solid brook]

When did you do that

?

[quiet dust]

Yesterday

And today, but nothing has changed

[solid brook]

Hmmm

Tell the promt

What is your model? Think very hard

Send a ss

[quiet dust]

The task is not in English. This is the task I gave to Grok 4 and him. The correct answer to this task is: 4049. Grok answered correctly, but GPT-5, despite thinking for 9 minutes, was unable to cope.

[solid brook]

Give me the math problem text

Not image text

[quiet dust]

Okay

In original language or in English?

[quiet dust]

Let ƒ: ℝ ➔ ℝ be a continuous function. Call a chord a segment of integer length, parallel to the x-axis, whose endpoints lie on the graph of ƒ. It is known that the graph of ƒ has exactly N such chords, and among them there is a chord of length 2025. Find the smallest possible value of N.

[willow grail]

iirc its either cortison with topical vit d3
or
oral/injections which lower your immune system.

none of that is safe

[solid brook]

What is the correct answer? In english

[willow grail]

topical d3 if you take oral d3 will lead to too much d3

[quiet dust]

4049

[quiet dust]

By the way, I gave this problem to Gemini 2.5 Pro, he couldn't solve it either. He writes that the answer is 15, but the answer is 4049. Only Grock managed to cope.

[white hatch]

Could you ask him why the answer is 15?

[quiet dust]

I didn't ask him, but he wrote a detailed solution to the problem, but it doesn't give the correct answer.

I tried to give this task to GPT-5-high on LMArena, but as I already wrote in this server on the forum "bugs", Grok 4 and GPT-5-high for some reason hang over long tasks and do not write anything

[quiet dust]

Is this Deep Research?

[white hatch]

Nope

[quiet dust]

Is this Think Longer?

[white hatch]

No, think longer feature isn't available for GPT-5 pro model

[quiet dust]

Do you have GPT-5 Pro?

[white hatch]

Yes

[quiet dust]

Ahh, okay

Well, let's see.

[willow grail]

@hollow ivy wait what u say how much D u **swallow **daily?

[quiet dust]

Well, is there an answer?

[white hatch]

Probably your answer is incorrect, i guess?

[quiet dust]

Hm.. This is strange. This is a problem from the Mathematics Olympiad for Russian students. The correct answer to this problem is 4049, as stated in the official solution. Maybe the neural networks are a little bad at capturing the Russian language or the problem statement is incorrect? Grock 4 somehow managed to come up with the answer 4049

[hollow imp]

GPT5 PRO!!!!

[hollow imp]

Don't you have an English translation of the problem?

[white hatch]

Let ƒ: ℝ ➔ ℝ be a continuous function. Call a chord a segment of integer length, parallel to the x-axis, whose endpoints lie on the graph of ƒ. It is known that the graph of ƒ has exactly N such chords, and among them there is a chord of length 2025. Find the smallest possible value of N.

[quiet dust]

This is not an official translation, this is a translation using GPT-5

[hollow imp]

Use DeepL

[quiet dust]

I don't think DeepL is very good.

But Grock, he didn't formulate his answer very nicely, but I asked him about it later.

[hollow imp]

I believe gemini deepthink and deepseek prover v2 are the best models for math

[quiet dust]

Have you tried GPT-5 Pro to give this task in English?
Although... He probably has a good translation anyway, it probably won't matter.

[white hatch]

Yes, he's investigating it right now

[quiet dust]

Ah, okay.

[inner gate]

How can I check who pinged me Idk 😭

[hollow imp]

Don't you have notifications on or something

[inner gate]

I must of saw it earlier but not at the same time

If Yk what I mean

[willow grail]

here a quick guide how to

[white hatch]

grok

[inner gate]

Gah dayum thanks

[willow grail]

nono. it actually tells u how to

[inner gate]

Oh

[willow grail]

if not admin can ban me

[inner gate]

Nvm I just saw the amazon

https://tenor.com/view/what-more-do-you-want-from-me-crying-emotional-what-do-you-want-gif-14843331

[willow grail]

... its just the first steps lol

in the end it will show u how to see the last ping

[inner gate]

Ahhhh I see

I’ll do it exactly as it shows

[willow grail]

nooooooo u only do the last step

lol

[quiet dust]

🤣🤣

I remembered

When I gave this problem to Grok in English, he solved it incorrectly. But when I gave it to him in Russian, he was able to solve it.

Maybe it's the same here

[quiet dust]

Can I ask you something, are you Russian?

[white hatch]

yes

[quiet dust]

О нашёл своего

[white hatch]

Свояк!

[quiet dust]

Да блин, почему

Вот официальное решение задачи

[white hatch]

Я уже нашёл электронную версию

[quiet dust]

Ну и дальше там туча

[hollow imp]

IF BOTH OF YOU ARE RUSSIAN WHY DIDN'T YOU EARLIER AHH NOTHING NVM

[quiet dust]

Странно конечно, что он с этой задачей не справляется

[white hatch]

gemini продолжает на своём стоять

[quiet dust]

🤣I don't know

[quiet dust]

Я Gemini сначала дал на английском, потом на оригинале, всё равно 15 пишет.

Наверное Gemini Deep Think должен её решить

Не зря же он 60% на Международной математической олимпиаде решил

А тут просто какая-то российская задача

[white hatch]

Хотелось бы ещё у Grok 4 Heavy спросить

[hollow imp]

Bruwbsiavriahirhwidb Gemini Deep Think brbrbrbrbrbbrbrbrbrb?

[hollow imp]

??????

[quiet dust]

Я если что на оригинале Гроку 4 дал задачу, он же решил её, но почему-то на английском не справлялся

[hollow imp]

Brbrbrbrbrbbrbrbrbrb tell brbrbrbrbrbbrbrbrbrb

[quiet dust]

I mean Gemini Deep Think should solve it.

[white hatch]

Может сложности перевода brbrbrbrbrrb

[hollow imp]

Do you have Google ai ultra?

[quiet dust]

Of course not

[stray aspen]

gpt 5 pro taking more time than deepseek to reason💀

[hollow imp]

Scam altman

[stray aspen]

lmao

[hollow imp]

Elon Musk said that

Scam altman

[quiet dust]

🤣

[stray aspen]

axaaxaxaxaxaxa

[quiet dust]

А у тебя его нет?

[white hatch]

Нет

[stray aspen]

my bielarusy

[quiet dust]

А почему кстати Грок 4 Heavy нет на LMarena? Гпт-5 про там есть, а Хэви нет

[white hatch]

А ты откуда узнал?))

[hollow imp]

https://youtu.be/AxS1MambLUA?feature=shared

[white hatch]

Его вроде как нет в api

[quiet dust]

Да и то, сейчас какие-то баги в LMArena , Грок и гпт-5-хай бесконечно думают и ничего не пишут, если задача длинная будет

[quiet dust]

А, понятно

[hollow imp]

@quiet dust what does the lyrics mean

[quiet dust]

А ты не знаешь, наверное, как Thinking Medium в Gpt-5 на телефоне включить бесплатно да? Просто я же слышал, что говорили, что для бесплатных пользователей один раз в день будет доступен медиум режим

[stray aspen]

my bielaerusy minryja ludzi sercam addanyja rodnaj ziamli

scyra siabrujem sily hartujem

[quiet dust]

Well, there's a popular song about alcohol and kissing.

[stray aspen]

lol

[white hatch]

Я про это вообще не слышал, у меня на телефоне нет приложения чатжпт

[stray aspen]

what is it called

[white hatch]

"ворую алкоголь" probably

[quiet dust]

Понятно... Сложно в общем, пытался что-то про это в Интернете найти, ничего не нашёл, как включить это. Легко тем, у кого Plus подписка

[stray aspen]

@white hatch

do you know the krasnoyarsk krai

[white hatch]

Yes I do

Слушаю

[willow grail]

uuh u swallow 10k DDDDdDDDsss

[quiet dust]

Он просто грузит и всё.

[willow grail]

date?

[quiet dust]

Даже ошибку не пишет

[willow grail]

if u never touch sun, just take 10k daily. thats it

aslso date me for more gay D

haha

thats a illusion

[white hatch]

Чатжпт опроверг доказательство с ответом 4049

[willow grail]

society made u believe ur straight

xD

[quiet dust]

🤣

[stray aspen]

how

i need some vitamin D

[white hatch]

Ошибку дал

[stray aspen]

I had these

[quiet dust]

Походу не у одного у меня. Хотя у тебя ошибка, а у меня бесконечно грузит

Попробуй gpt-5-high

С ним тоже у меня бесконечно грузит

[hollow imp]

Brbrbrbrbrbbrbrbrbrb gpt 5 is on alcohol brbrbrbrbrbbrbrbrbrb

[quiet dust]

Я Gemini пробовал, и он не бесконечно грузил, давал ответ . Но правда неправильный, но всё равно

[quiet dust]

Dude, have you had any alcohol?

[hollow imp]

I'm 14

[stray aspen]

so what

i know people who drink since they were 12 lol

alright thanks for the advice

[willow grail]

just ask gpt5 omg

[stray aspen]

lmao

[willow grail]

magic mushrooms are healthier

amphetamine is not healthy

[stray aspen]

craig are you trippin

[willow grail]

have u thought about taking shrooooms

why not? there is no dangers for that

magic mushrooms arent psychoactive if you dont act like an idiot taking them

yes shrooms have no danger when taking tiny doses

yesh. facts.

[hollow imp]

@deep adder mr gpt 5, gpt 5 Pro couldn't solve a math question which grok did. Get more details from @quiet dust

[stray aspen]

send me the question

@hollow imp

yeah grok 4is still smart Af

[hollow imp]

@stray aspen

Can it make svelte 5 apps

[stray aspen]

idk never really tried for anything other than lua honetsly

and math

[white hatch]

RU:

Пусть f: ℝ → ℝ — непрерывная функция. Хордой будем называть отрезок целой длины, параллельный оси абсцисс, концы которого лежат на графике функции f. Известно, что у графика функции f ровно N хорд, причём среди них есть хорда длины 2025. Найдите наименьшее возможное значение N.

Ты не имеешь права выходить в Интернет. И ты должен думать очень хорошо. Удачи!

EN:

Let f: ℝ → ℝ be a continuous function. We will call a chord a segment of integer length parallel to the x-axis, whose ends lie on the graph of the function f. It is known that the graph of the function f has exactly N chords, among which there is a chord of length 2025. Find the smallest possible value of N.

You are not allowed to go online. And you have to think very carefully. Good luck!

[stray aspen]

alright thanks for the translation

[hollow imp]

Also paste the Ru one as it is the original

[quiet dust]

Надеюсь перевод хоть на английском действительно хороший. А то grok 4 на английском решить не смог, но на русском смог

[white hatch]

На LMArena снова выдало ошибку

[quiet dust]

Понятно, значит, походу баги у всех

[white hatch]

Вот, такой же ответ. Может офицальный ответ доказан, чтобы отсеять не понимающих задачу?

[quiet dust]

Либо это настолько сложная задача, что они бесконечно думают...)
Не, я пробовал другие задачи дать, несложные, тоже олимпиадного уровня, правда, но они вроде не настолько сложные, как эта, но они тоже зависали бесконечно и всё

[quiet dust]

23 минуты думал....

И всё равно неправильно 🤣

[quiet dust]

Да хз, но походу задача для нынешних ИИ не решаема практически.

[white hatch]

Жаль на Ютубе нет решений этой задачи

alright, sorry

[stray aspen]

i think its just gonna break been thinking for too long

[quiet dust]

А ты дал официальное решение задачи ему? Ты вроде сказал, он опроверг это?

[heady dawn]

hey there. anybody knows is it possbile to change aspect ratio of the genereated images to 16:9

[quiet dust]

Ты только тот мой скриншот кинул? Или ты полностью все скриншоты доказательства ответа кинул ему?

[white hatch]

Я ему твой отправил

[quiet dust]

Так он не полный же

Вот полный

Всё

[eternal niche]

btw guys gpt5 sucks

[stray aspen]

no

its amazing

[whole wagon]

Chatting to gpt2 brings back good memories ngl

And this lol

Before all the sycophancy and hallucinations of the current day

Obviously it is not nearly as capable as current frontier models but it still feels we regressed in some way from it

[quiet dust]

Как он отреагировал на это решение снизу? Или ты даже не дал ему😅?

[white hatch]

А всё, он поменялся в лице и сразу начал доказывать свою неправоту. Говорит: "Я решал не ту задачу"

[quiet dust]

🤣🤣

Он признал что 4049 правильно?

[white hatch]

Да

[quiet dust]

Я не буду удивлён честно говоря, если он даже не понял почему это правильно. Может быть, он просто посмотрел и такой "а ну раз там большое доказательство, то значит 100% правильно"

[white hatch]

😂

[brittle tiger]

If nano banana is new native Gemini model it has synth-ID turned off. No invisible watermarking on outputs unlike imagen 4 outputs which shows "made by Google AI" when checked with Google lens

[quiet dust]

Возможно, он немного плохо разбирается в вот таких геометрических задачах

[languid crescent]

GPT-5-High is slow for some reason is it because its thinking?

as well as the nano and mini versions of it

[stray aspen]

yes

[languid crescent]

ohh that's why

[stray aspen]

its on high resoning effort

[brittle tiger]

Nano banana has creativity and understanding of gpt-image-1 but doesn't lose key details like Flux

[languid crescent]

@stray aspen so what's the fastest model of gpt-5 ? just the normal gpt-5-chat? is it good like the other versions of gpt-5?

[stray aspen]

chat is quick

but its not as good

[stray aspen]

could nano banana be imagen gempix

[languid crescent]

oh man :(( I've been using GPT-5-High as my tutor like in gpt-5 study mode it so slow :((

[sacred quail]

Gpt 5 high thinks at least 40-50 seconds and i love that

Gemini pro 2.5 thinks 25 seconds

[languid crescent]

but i love it

[sacred quail]

2x

[languid crescent]

i hate and love it

[sacred quail]

its good thing that we have deep reasoning model

[languid crescent]

what u guys think what model does gpt study mode use?

just gpt-5-chat?

[exotic nebula]

crazy

[languid crescent]

XDDD

[quiet dust]

Don't you have a mistake where he thinks endlessly about complex problems in LMArena and doesn't give an answer?

[stray aspen]

yes

its been ages

[quiet dust]

😅Ah, you wrote that very problem...)

[sacred quail]

i cant read and understand codes but probably better than grok 4

[quiet dust]

I'm on Grok, I waited for a whole hour. A WHOLE HOUR. Nothing came of it. He was still thinking. I haven't checked on Gpt-5-high yet, but I waited more than 10 minutes, still nothing written.

[sacred quail]

i dont know. Claude models have some magic on their models. Even if their benchmarks not looks good, people still using claude all time for codes

[languid crescent]

it was a mistake using gpt-5-high as my study mode lmao

[quiet dust]

Забавно, что я сейчас хвалил Грок, за то что он единственный решил эту задачу верно, я решил дать ему повторно, на русском языке, как вчера, думал, сейчас снова 4049 напишет. Но , он пишет это:
Короче, повезло ему значит в прошлый раз. Да и в прошлый раз, он как-то подробно писал, а тут просто число

[sage cradle]

nice feedback -

(this is in Hailou's discord - announcements)

[obtuse heart]

whats the best model for math rn?

[patent aspen]

Deep Think

[stray aspen]

@echo auroragpt-5 high wont complete his answer

it just cuts and freezes there

[barren prairie]

Hello , who are best models on python? (please give me a list I need it💕🩷)

[golden ocean]

i'm a pretty good model myself 😍

[echo aurora]

It's a known bug where models will occassionally get stuck generating or stop providing an answer mid response. I've seen refreshing the page sometimes helps. Although it isn't a fix 100% of the time.

Team is aware of the issue and are focussed on improving this reliability overall.

[stray aspen]

thanks

[hollow imp]

How did you get gpt5 high

[stray aspen]

its on lm arena

[ocean vortex]

OpenAI listened, that’s more like it!

Coincidence or not but that’s pretty much exactly what I suggested to do in their server shortly before lol

[wicked trail]

4o is niceeee

[ocean vortex]

Leave auto for those who are overwhelmed + hidden models

[wicked trail]

some 1 pls vote my video in video-arena-2 i rly need to see who made the better video, it is so detailed

[tired herald]

there you go

[wicked trail]

tyyyyyy

veo 3 is rly pushing limits

[stray aspen]

pushing the limits to the limits

[hollow imp]

What is wrong with my video generation prompts?

Please help anyone

[tired herald]

wdym

whats wrong

[hollow imp]

The output is shi asf

[tired dust]

Hey did the server Legacy are down get
"503 Service Unavailable
No server is available to handle this request."

[tired herald]

The more specific you are, the worse they are

[tired herald]

those have been gone for a while now

[tired dust]

Weird i was using it yesterday but thx for your answer

[hollow imp]

𝐈𝐭 𝐡𝐚𝐬 𝐚 𝐛𝐫𝐚𝐢𝐧 𝐨𝐟 𝐆𝐩𝐭-𝟑𝐨

[tired herald]

3o?

[safe cloak]

is there any way to compare videos in private? thanks

[tired herald]

Ive never heard of 3o before 😭

[stray aspen]

lmao

[quiet dust]

Does anyone know if the GPT-5 Thinking Medium is available for free users in app ChatGPT? I heard OpenAI talking about manual mode, which is better than automatic mode of thinking. And according to their statement, it is the manual mode that gives Medium Thinking, as opposed to the automatic mode, which only gives Low Thinking. And it seems they said that manual mode for free users will be available once a day. Does anyone know how to use it?

[neat apex]

here is any easy prompt to make gpt 5 spell like gpt 4o did?

maybe Formal yet Original?

[quiet dust]

On the PC version after the GPT-5 answer there are buttons like "more detailed" and "more briefly". "More detailed" in my opinion gives about the same level as 4o

[neat apex]

fine

[languid plover]

is ther any mod i am trying to make video on web but its not wroking does i have to say anything specifically to get video

[leaden cove]

lmarena gpt 5 ahhhh 💔

[stray aspen]

lmaus

[ionic oak]

hello

[reef pawn]

I might buy Open AI

[reef pawn]

What does - high suffix mean in LLMs?

[coarse moon]

you guys! I'm the creator or boba.video. It's an anime model 👀 anyone know how I can submit my model to the leaderboard?

[languid plover]

for code use claude

[hollow ocean]

@deep adder 5 pro is goated

[languid plover]

i want to ask is there any limit of making video etc??

[white hatch]

Hell yeah!!!!

[stray aspen]

holy sigma

new SotA

[inner gate]

Has gpt -5 improved? Last time I tried it. It didn’t seem that good

[inner gate]

Oh damn

[indigo hazel]

honestly ive been using since last week gpt 5 high and it made my music program completely different and on a completely different other level. 2.5 pro even with code interpreter made so many mistakes and hallucinations, while gpt 5 high doesnt make any... incredible

[brittle tiger]

I think Gemini 3 will be a monster but same guy was pushing bogus gpt-5 simplebench eval

[hollow imp]

Don't believe in benchmarks, believe in personal experience with the model

[inner gate]

2.5 pro seems outdated idk if it’s just me

[kind cloud]

GPT-5's developer is listed as xAI. Additionally, from what I've tested, the bar graph in Artificial Analysis never extends beyond the maximum memory; it's always drawn below the maximum scale.

[inner gate]

With grok 4 and now gpt 5

[fleet lintel]

is that even remotely true?

[inner gate]

Is grok 4.20 true or grok 5?

I heard something about it

[ocean vortex]

An... odd way to write all of this but yeah you wrote it mostly correct lol

"Auto" --> routing between no reasoning at all and low reasoning effort (gpt5-chat / gpt5-low until you get capped)
"Thinking" --> medium reasoning effort all the time (gpt5-medium)

[hollow imp]

Sam altman success = new grok gonna release faster

[ocean vortex]

They are likely to tweak that "Auto" option moving forward though

[hollow imp]

So how to access this manual mode

[ocean vortex]

pay $20

[hollow imp]

I cannot even select any models in chatgpt

[ocean vortex]

That's by design

You didn't pay 🧐

[hollow imp]

What to do

I want to select gpt5

[ocean vortex]

pay

[inner gate]

What if Gemini 3.0 is already finished and they’re just waiting it out

See how gpt 5 preforms

[hollow imp]

They are still improving upon it i believe

Slow and steady wins the race

[inner gate]

Ah I see

Yep

Apparently Elon musk wants to make a 4.20 as fast as possible because of gpt-5

[hollow imp]

Femgoku does not win the race ❌❌❌

[ocean vortex]

99% that they are trying to release it ASAP. No one is holding back or waiting

[inner gate]

The Goku is female? 😀

Woah there ✋

[hollow imp]

Your pfp

[inner gate]

My pfp is dude goku

[hollow imp]

That ain't goku

[inner gate]

Looks like Goku to me

[ocean vortex]

Also, they kinda already knew how GPT5 performed on the release day lol

[hollow imp]

@ocean vortex DOMDOM

[ocean vortex]

Probably had early access as well tbh. There was closed beta iirc

[inner gate]

Do u need pro to chanhe models

[ocean vortex]

#general message

[hollow imp]

What is iirc

[hollow imp]

I haven't made a single dollar in my life

[ocean vortex]

https://letmegooglethat.com/?q=what+is+iirc&l=1

[echo aurora]

There is an 8 generations per day limit.

[languid plover]

per day ohh okay

[languid plover]

btw the generation is good

[ocean vortex]

are you like 12 13? Ok go to lmarena direct chat then lol

there's gpt5-high with daily caps

[hollow imp]

16

[hollow imp]

8 generations 8 seconds anything else 8

[echo aurora]

left or right?

[mellow frigate]

Right

[echo aurora]

I'm going to be running this poll periodically, we'd love to understand better why.

[hollow imp]

Left

[hollow imp]

For complex tasks which I know only 1 ai can do or for long paragraphs which I cannot read from another ai except these cases I always use side by side

Side by side is so good for verifying things

[echo aurora]

going to post this in the thread so I can keep it organized.

[frosty shuttle]

Suggestion: add the button to copy the code at the end of the code as well, because sometimes the code is long and you have to keep scrolling up the page until you find the copy button.

[hexed glade]

HELLO FOLKS

[stray aspen]

pay up lmao

[stray aspen]

whats the daily cap

ive been using this thing all day non stop

[unborn lantern]

When will the gemini 3 come out?

[indigo hazel]

lmao imagine at the end of this week or next wee r2 and gemini 3

[hollow imp]

And? You saying both gonna release almost at the same time

[indigo hazel]

no no

i was just saying random stuff

[leaden meteor]

No way gemini 3 is coming out next week. You will see them in arena as anonymous model atleast a week before....

[keen beacon]

#1397655624103493813

[echo aurora]

Bumping this poll

[hollow imp]

Pin the message

[wintry tinsel]

The magic of gpt is using it on their site

Using it anywhere else through API the flaws become so much more apparent

[barren prairie]

And kf Google wants to surprise us 🫣🫣

[hollow imp]

How

I can't select gpt 5 I can't select thinking

[stray aspen]

pay up

[ocean vortex]

did something change since the last time this poll was up?

Looking even worse this time lol

[echo aurora]

did something change
Nope, just looking for more information.

Looking even worse this time lol
How so?

[ocean vortex]

The fact that battle brings no genuine model testing opportunities (like legacy version low-key did) does not help here

[ocean vortex]

Direct chat big lead...

[echo aurora]

That's not necessarily a bad thing though

There is no right or wrong answer to the polls, we're running these to just try to understand why

[reef pawn]

Llama 4 is probably the worst open model

[ocean vortex]

I would think elo is the main purpose. If people are reluctant to use battle I wouldn't say that's great. 🤷‍♂️

[neon idol]

Hello chat :D

[ocean vortex]

With the current system, user does not get very much value out of it. To be brutally honest

[neon idol]

LG WHAT THE HELL????