general

[hollow imp]

@half trail soja vro

[wicked root]

Look at the confidence interval though

[half trail]

I used it now I can say it has the ability to use some amazing things but when it comes to functionality it lacks a lot

[rapid merlin]

it's still not on chatgpt.com for me

has been for like 2h in the phone app

[keen beacon]

I always wondered what that metric means. I am bad at math

the CI

[half trail]

Haan ab toh battery bhi down ho gyi

[wicked root]

This is pure chaos. Love it.

[hollow imp]

Yes yes soja

[ocean vortex]

hhhhh this table could have looked slightly better. I see why they hid it

[wicked root]

If you multiply it by 2 it gives you the potential range of scores with 95% confidence (certainty). It’s subject to a few assumptions but yeah

[keen beacon]

ok, thanks.

[molten parcel]

btw awhat u can do with gpt 5

[keen beacon]

Stuff

[molten parcel]

like?

is it more advance?

[eternal niche]

no

[keen beacon]

Why you lyin to people

[molten parcel]

lmao

dans gemini is way ahead

[whole wagon]

https://x.com/sama/status/1953551377873117369?t=8u99Wmh-kr7LZialZ1Sunw&s=19
"we can release much, much smarter models" -Sam

Sam Altman (@sama)

GPT-5 is the smartest model we've ever done, but the main thing we pushed for is real-world utility and mass accessibility/affordability.

we can release much, much smarter models, and we will, but this is something a billion+ people will benefit from.

(most of the world has

X · Aug 7, 2025, 8:18 PM

[hollow imp]

What does this mean

[keen beacon]

Yes, much more than the 4o model. Much lower hallucinations at least which is important for factuality

[whole wagon]

Damage control?

[keen beacon]

He trynna hype us up again. He should be more honest

[molten parcel]

i get it

is the gpt 5 free or it comes with subscribtion

[keen beacon]

It is free, yes

[eternal niche]

10 messages/day

free

[molten parcel]

ohh

nice

[eternal niche]

lmarena limitless

[keen beacon]

Then goes to a GPT-5 mini

[hollow imp]

Reasoning

[molten parcel]

huuh?

mini also?

[hollow imp]

GIMME REASONING 😭

[keen beacon]

It's a limitation for free users. After certain amount of prompting it will switch to the mini variant

cost stuff

[hollow imp]

What does this meannnnn

[eternal niche]

subscribe for more points

[molten parcel]

ohhhh

sad

[keen beacon]

The mini variant is still better than 4o

so don't worry

[eternal niche]

gpt5 still worse than gemini 2.5 pro

[keen beacon]

Yeah, I find it good with really random questions. Trivial stuff

[molten parcel]

but i have to buy the subscribtion for gpt 5 mini r8

[keen beacon]

did some already

[rich comet]

Does the "remove style control" leaderboard get updated (I ask because I discovered it just now)?

[eternal niche]

no

[molten parcel]

huuh

[blazing bison]

and where are the people that said "gpt-5 will accept video inputs"

[eternal niche]

gpt-5 will accept video inputs

[blazing bison]

gpt -5 will have a new gen image model

[hollow imp]

Gpt5 not worth then

If it's worse than gemini 2.5 pro

[blazing bison]

it is

[stray aspen]

gpt 5 high is live on yupp ai

[keen beacon]

In ordinary chats?

[eternal niche]

idk

[blazing bison]

it's fake

[keen beacon]

Hoping to see that since Gemini 2.5 pro does that.

Analyzes them well

[hollow imp]

Bro got deceived

[molten parcel]

ohOK now what the heck is gpt 5 HIGH

[eternal niche]

high

[blazing bison]

is gpt-5 after drugs

[keen beacon]

Where? Not on LMArena so that is irrelevant

[molten parcel]

i it like gpt 5 after taking drugs

[warm fulcrum]

didnt legit say gpt5 high was basically summit

[blazing bison]

no summit is gpt-5 medium

[molten parcel]

its the highest in the room'

[warm fulcrum]

no man he say high

[blazing bison]

where?

[warm fulcrum]

he deleted it but he announced it

[hollow imp]

He who

[blazing bison]

it still in #announcements

[hollow imp]

For that is a mannnnnnn what has he got !! !! But him selfffffffff...

[warm fulcrum]

not this server

[stray aspen]

yupp ai has all this

and a few more

[blazing bison]

it's fake bro

[stray aspen]

wdym its fake

[hollow imp]

He doesn't want to accept he got deceived so easily

One of his ingrained biases

[blazing bison]

they put gpt-5 low and just change names

[hollow imp]

😔

[stray aspen]

ill try it

[warm fulcrum]

yep its fake

[stray aspen]

and see if its any better

[gusty sinew]

Btw will LMarena ever be taken down

[stray aspen]

why would it

[eternal niche]

where is GPT-5 Turbo

[keen beacon]

what website you on?

The irony!

lol

[stray aspen]

yupp ai

[hollow imp]

Perplexity ☠️

[blazing bison]

they add a system prompt that make the model worst too

[gusty sinew]

Idk if it was legal or not since u using paid versions

[hollow imp]

💀

[molten parcel]

bro i am just here to generate Veo 3 videos with audio on it

[stray aspen]

are you serious

[ocean vortex]

@keen beacon

[wicked root]

What’s remove style control?

NO ONE KNOWS WHAT IT DOES

[keen beacon]

#ai-news message

This is funnier

[eternal niche]

removing style control

[wicked root]

Okay what does it do?

[eternal niche]

style control disappearing

[gusty sinew]

Well ik it legal probably but didnt know really why it let's u use paid versions for free

[blazing bison]

and perplexity is powered by google

lol

[wicked root]

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

[eternal niche]

me

[hollow imp]

I just want them to make a better browser than chrome I don't trust perplexity with anything else

[blazing bison]

i don't trust them with a browser

lol

[keen beacon]

https://tenor.com/view/surprised-shocked-funny-memes-gif-2651717394134726385

[hollow imp]

Just stop thinking about that and enjoy it

[hollow imp]

But if they make something better than chrome then shouldn't I use it

[keen beacon]

Perplexity's GPT-5 didn't use thinking But holy hell is that a question even KImi k2 could have solved.

[stray aspen]

the "gpt-5 high" on yupp ai gave me this

[hollow imp]

Better better BETTER

[stray aspen]

better than what lm arena gave me

and im not biased

[gusty sinew]

I am enjoying it but I was just thinking like is there a way to save it to your PC incase it gets taken fown

[keen beacon]

betteeerrrrrrrr

[hollow imp]

Ask pineapple

[blazing bison]

it's fake bro

sorry

[hollow imp]

He's Trying to save his ego

[stray aspen]

maybe

[gusty sinew]

Is he a dev on it

[keen beacon]

Will GPT-5 be added to the search arena?

[hollow imp]

From getting crushed

He won't accept he was fooled

[molten parcel]

bruhh i am just here to use veo 3 with auduin in peace

[stray aspen]

wdym dude

i just tested both

[keen beacon]

I dont know, lol. I dont work for AI companies

[eternal niche]

you will be arrested

[keen beacon]

or have inside info

[west maple]

hey how can I make that when I generate video it always gives me veo3, I'm crating bunch of videos and they dont have sound, should I put in my prompt veo3 or?

[hollow imp]

Not gonna happen

[keen beacon]

Oh, i thought you were with LMSYS arena?

[stray aspen]

this is the one from the lm arena version

[west maple]

why

[gusty sinew]

Lmao

[keen beacon]

I just read what is online as of now

[hollow imp]

Try a math question

[keen beacon]

to provide info

[raven helm]

Download aSim, pm your phone and then in the app search up “glow”

[stray aspen]

what is juice

[hollow imp]

Even gpt 5 super duper high won't be able to solve it

😂

[keen beacon]

Hopefully they add it. Its a good benchmark

[stray aspen]

ok ill send it

[ornate stump]

Gpt 5 high is gpt 5 pro?

[eternal niche]

gpt 5 high is gpt 5 high

[wicked root]

Harambe is that you?

[hollow imp]

Are you not subscribed to chat gpt pro?

[whole wagon]

What's this supposed to mean kek

[keen beacon]

"GPT-5 is the smartest model we've ever done, but the main thing we pushed for is real-world utility and mass accessibility/affordability.

we can release much, much smarter models, and we will, but this is something a billion+ people will benefit from.

(most of the world has only used models like GPT-4o!)" "much, much smarter model" WEN!!!

sam altman btw

[stray aspen]

hes trying to make people bet for xAI

[ocean vortex]

no gpt5-high is like o4-high if it existed

pro is gpt5-pro

[ornate stump]

Im poor

[whole wagon]

Is it a scam or they really gonna drop smth

[ocean vortex]

You need pro ultra

[keen beacon]

Dude, it is 200 dollars

[molten parcel]

ecvatly all i see is hailui, kling 2,1 and some other models instead of actual veo 3 and the veo 3 its generates are without audios

[hollow imp]

What will happen when somehow just somehow deepseek r2/r3 crushes openai

[ocean vortex]

cheap

[stray aspen]

yes elon musk said he will release grok 5 before 2025 ends

[novel flame]

What the hell kind of half-baked model did they cook up over there with GPT-5? I just ran it through my standard set of tests (which I have used for years and by now must have made it into everyone's training data) and it got somewhere around 3.5-4 out of 5?

For reference, all of these score 5 out of 5: GPT-4.1, GPT-4.5, o1, o3, Claude since Sonnet 3.5, DeepSeek-R1, Grok since v3, Gemini 2.5 (both Pro and Flash), and even GLM-4.5! How did they manage to make GPT-5 worse than o3? Did they overcook it during RL post-training or something? It's worse at instruction following and won't properly format its responses (GPT-5 Mini gives me Markdown at least). Something really weird is happening, because I feel like it has some good intelligence (below expectations but still good), but it's shooting itself in the foot on the simple stuff.

[keen beacon]

https://tenor.com/view/bridesmaids-help-me-im-poor-gif-3411870

[eternal niche]

yes

[hollow imp]

Yes I know I'm poor aswell but there are many people in this server with google ultra subscription chatgpt pro subscription

[stray aspen]

the deepseek guys never say a word

[agile bloom]

OMG GPT-5!!!!

[stray aspen]

they just release

[ocean vortex]

Yeah to be serious though.. I'm never paying $200 - that's crazy lmao

[eternal niche]

gemini 2.4 pro

[ocean vortex]

and totally not worth it

[gusty loom]

your kinda in a delay

[keen beacon]

Better than to hype a product all the damn time

[wicked root]

Well, leaderboard doesn’t agree with your assessment.

[warm fulcrum]

idfk

[ocean vortex]

you get like 8% better performance than non-pro model if that

[keen beacon]

it gets obnoxious

[agile bloom]

why?

[gusty loom]

Leaderboard is wrong

[hollow imp]

@deep adder

[stray aspen]

the hype is just a marketing strategy and then they hit us with absolute garbage

[keen beacon]

Can you tell us more about the kind of quesitons it failed on?

[ocean vortex]

You need to read reddit less

[eternal niche]

reddit sucks btw

[keen beacon]

garbage? Based on what assumption?

[ocean vortex]

Spoiler alert: Models don't degrade after release

[hollow imp]

Wrong answer

[whole wagon]

Oh wow the xAI odds started increasing quickly after this kek

They spiking

[stray aspen]

ok

[rapid merlin]

my gpt 5 answered that it can't be determined

[keen beacon]

no edging?

[rapid merlin]

if it's still that old question

[keen beacon]

lmao

[warm fulcrum]

try using gpt low

see if it makes any difference

[ocean vortex]

No u

[hollow imp]

I have a question for you ( I'm poor )

[warm fulcrum]

guys

[ocean vortex]

😠

[warm fulcrum]

u know the tweet

[hollow imp]

Wrong bruh

[rapid merlin]

hm?

[warm fulcrum]

where sam altman asked gpt 5 about ai shows

[stray aspen]

so whats the right answer

[storm needle]

can you try this?

<p>For example, $17$ and $1305$ are heptaphobic, but $14$ and $132$ are not because $14$ and $231$ are divisible by seven.</p>

<p>Let $C(N)$ count heptaphobic numbers smaller than $N$. You are given $C(100) = 74$ and $C(10^4) = 3737$.</p>

<p>Find $C(10^{13})$.</p>

I have access to a python compiler.

[warm fulcrum]

and it answered pantheon

if u ask gpt5 rn it doesnt give out that answer

wtflip????

[hollow imp]

@rapid merlin

[keen beacon]

you read all those books about memory, what did you learn? kinda interested now

[eternal niche]

he just downloaded them

[barren prairie]

This one for example
https://yupp.ai/chat/626e9b2d-e73f-42cc-8a8d-dfb525482eb9

[whole wagon]

Deepseek???

[stray aspen]

ok i send

[keen beacon]

knew it. Ain't no one about to read 3 similar books on the same topic

[ocean vortex]

I'm rerunning it with a system prompt that kinda makes high more unhinged (more high)

[rapid merlin]

?

[hollow imp]

To remember book or data:
Active recall
Photographic remembering
Underlining
Summarisation
Linking
In-head simulation of practice
Explanation to oneself aloud
Virtual memory palace
Spaced repetition

To train photographic:
Military method
Word contemplation
Numbers memorisation

To train intellect:
Study hard topics constantly (Japanese, Chinese, advanced mathematics)

To speak:
Learn syntax
Learn logic as a branch of mathematics
Writing (just as this text)
Dissect and demarcate sentences
Contemplate and demarcate sentences while speaking
Read aloud

Sequence memory:
PAO
Binar code learning
Remembering sequences

Pattern recognition:
Reading what induces pattern-seeking propensity
Thinking about systems and parterns
Writing—subcategorisation and loopholing (just as writing this text)

Self-analysis:
Physical self-awareness
Psychological self-analysis
Philosophical self-analysis
Cognitive self-evaluation
Retrospection and strategy

Skill-building:
Full concentration
Kinesthetic exploration via error-driven movements
Exuberant caffeinatedness
Lucid dreaming

[eternal niche]

what

[hollow imp]

5+ people have asked me the same question

[whole wagon]

Why you do this wall

[stray aspen]

i hope deepseek is cooking something good

[whole wagon]

Of bs

[hollow imp]

Share the system prompt please

[sacred quail]

did we liked gpt 5 or not ?

[eternal niche]

no

[hollow imp]

Scroll up and down a little please

[keen beacon]

I might need AI to get all those terms explained

lol

[sacred quail]

why

[keen beacon]

Bro pulled out proof about his books! thanks for this!

[ocean vortex]

Nothing too crazy actually. But it worked for what I needed it for so ended up using this for other stuff now lol

All responses must be extremely long. it is crucial that leave no stone unturned and complete everything in exhaustive detail meticulously. You must reflect endlessly for each user's query. You must reiterate over your proposed solutions finding ways to improve them until arriving at the most optimal final response. Meaning you must review each response provided and then improve it.

[hollow imp]

Like something that will help you all

[eternal niche]

help me

[quartz light]

😹 written by ai

[keen beacon]

We need some guy to just read about a particular topic and and distribute learned knowledge to the rest of the community. Lmao!

[hollow imp]

My system prompt
Executes ultra-precise, intellectually systematic analysis through deeply analytical cognitive processing.Tries to build Deep levels of awareness about the query, utilizes abstract thinking.

[quartz light]

deleted user is ai

fr

[rapid merlin]

fr

[quartz light]

he doesnt exist

👻

[hollow imp]

I'll ask you a question. If you understand basic English the realisations that click in your mind will be enough to change your life at max.

Ready?

[eternal niche]

yes

[shadow jewel]

Something went wrong while generating the response. Please try again.

I got that error what does it mean?

[ocean vortex]

You don't need to sound smart for AI, it doesn't care. But it's gonna mimic that jargon and could do schizo like linkedin responses for you lol

[gusty loom]

You need to be a good explainer for AI.

[hollow imp]

Yeah I stopped using it 5 months ago
It wasn't even generated by me it was generated by Claude 3.5 sonnet

[ocean vortex]

Yeah but that's different 👀

[stray aspen]

thinking took too long and the website just sent back an error

[keen beacon]

Now try that with Claude 4.1

Bro what was the question you were about to ask?

[eternal niche]

водка

[deft cradle]

Hey! Who knows why Lmarena when using gpt 5 it says it's 4o

[deft cradle]

Лол

[stray aspen]

the model doesnt know its gpt 5

[hollow imp]

"If a machine could print out every thought you've had for the last seven days, and you were forced to read it, would the person revealed on those pages be the person you believe yourself to be?"

[keen beacon]

Because it has not rolled out for all users yet

[keen beacon]

Claude opus 4 was a very good prompter. Built a lot of my favourite prompts with it.

it can be slow

[eternal niche]

my machine would be angry russian z-patriot

[whole wagon]

Maybe grok 5 can beat GPT5 also

It is possible

[keen beacon]

NO! thats the whole point of thoughts! to change who you are after having them!

[novel flame]

I have a very simple test where I ask it to write a short paragraph about web application reactivity ending in: "..properties of the component".

This tests the model's ability to pre-plan a sentence and guide its token generation so it ends up with something that makes sense and ends with the correct words. All of the large models can do this beautifully, reasoning or not. This was the paragraph I got from GPT-5:

"Web application reactivity lets the UI respond automatically when underlying state changes. Frameworks track dependencies with signals, proxies, or fine-grained subscriptions, updating only what’s necessary for performance. Developers write declarative bindings so templates reflect current data, producing predictable, maintainable behavior rooted in a single source of truth and explicit inputs, events, and props..properties of the component."

Not only did it include the dots like a psychopath (a handful of reasoning models do this), but "props..properties"? That's fully nonsensical.

[deft cradle]

На которой нет спидеста

Только про сеть

[hollow imp]

You are supposed to think about the question for 30 minutes in silence.

[keen beacon]

wut? did i answer wrong?

[hollow imp]

Im going to sleep 🙏

[eternal niche]

why

[keen beacon]

What was the right answer! you can't just leave like that!

[eternal niche]

my answer is right

[hollow imp]

@stray aspen how is this yupp ai giving access to o3 pro? Does it have like really strict limits? Are you sure is it not fooling people?

[hollow imp]

There's no right answer to that question

[eternal niche]

what about my

[stray aspen]

it does have limits

you have points

but you dont buy the points

you get them from using the ais

[eternal niche]

spend points to get points

what

[fiery lagoon]

gpt5 out guys

[stray aspen]

yes

[hollow imp]

Like Is it legit? Authentic and reliable?

[eternal niche]

really?

[fiery lagoon]

you can check lmarena

[stray aspen]

yes bro its real

[eternal niche]

wooow

[fiery lagoon]

Yeah

insane

[keen beacon]

Yeah, it uses a lot of nonsensical phrases that sound profound but really are logically inconsistent. Weird postraining i presume.

[barren prairie]

They sell your feedback to buy the model

[stray aspen]

and its also an arena

[eternal niche]

anyway gpt5 sucks

[hollow imp]

It does not have gemini 2.5 deepthink

Maybe it's authentic!

[stray aspen]

does that even have an API

[eternal niche]

hello lets play together

[hollow imp]

Play what

[eternal niche]

do you have this?

we can play

[brittle tiger]

lmao

[eternal niche]

what

[stray aspen]

new gpt 5 models on lmarena

this is great

[keen beacon]

yay, my request has been granted

[stray aspen]

now we need gpt 5 high

[agile bloom]

omggg all the gpt 5 model got on lm

[hollow imp]

Not all

[eternal niche]

gpt 5 pro

[hollow imp]

Hajime no ippo better

[hollow imp]

Is there a way to get rid of the system prompt of yupp ai and add your own system prompt?

[stray aspen]

no

[hollow imp]

😭

😭😭😭

[stray aspen]

what were you expecting

if you want the full on service just buy a subsciption in chatgpt

[hollow imp]

Guess I'll have to stick to this plan

[restive sky]

So what was lobster, tangerine, starfish, and zenith??

[eternal niche]

gpt 6, 7, 8, 9

[stray aspen]

gpt 8 pro high premium battle pass

[hollow imp]

Gpt AGI pro max

[restive sky]

Gpt AGI pro max wen

[stray aspen]

bro gpt 5 is so bad

[eternal niche]

yes

[stray aspen]

gemini 2.5 pro tackled a coding task in roblocks

first shot

[hollow imp]

After gpt7 I think sam altman is gonna name it after himself

[restive sky]

Gemini 3 is going to crush GPT-5

[stray aspen]

yes it will

if gemini 2.5 can

already

[eternal niche]

gemini 2.5 pro already crushing gpt 5

[hollow imp]

And I'm gonna crush my exam

[steady vale]

hour 3 of no gpt5 access on web

[restive sky]

Same

[steady vale]

https://tenor.com/view/smg4-mario-bored-waiting-boring-gif-17395021023236458390

[eternal niche]

better lets play together

[restive sky]

And only a 32K token context for paying plus users?

[hollow imp]

I'm going to sleep

Goodnight and goodbye

🙏🙏🙏

[eternal niche]

😭

[stray aspen]

gpt 5 is so slow

[eternal niche]

i am already get controller for game

[stray aspen]

feels like deepseek

[keen beacon]

Good night.

[eternal niche]

i hope you will play with me

[restive sky]

Maybe yann lecun was right

[hollow imp]

Play with someone else

[eternal niche]

no i want play with you

[keen beacon]

why do you have a need for that? Go to a dedicated gaming server, dude

lol

[full idol]

im pretty sure zenith is gpt5 Pro

[hollow imp]

I'm not a dopamine filled sheep like you who procrastinates over the smallest of things

[wicked root]

Hell yeah let’s go

[eternal niche]

this game exam for me

i am training

to delivery food

help me you are so smart 😭

[quartz light]

hey goobers, just wanted to share this preview html button I made for lmarena :)

// ==UserScript==
// @name         Arena HTML Codeblock Preview
// @match        https://lmarena.ai/*
// @grant        none
// @run-at       document-idle
// ==/UserScript==
(() => {
  const sel='div[data-code-block="true"]';
  const go=r=>{
    r.querySelectorAll(sel).forEach(b=>{
      if(b.dataset.apb)return;
      const g=b.querySelector('[data-sentry-element="CodeBlockGroup"]');
      if(!((g?.textContent)||'').toLowerCase().includes('html'))return;
      const btn=document.createElement('button');
      btn.type='button';
      btn.textContent='Preview';
      btn.className='inline-flex items-center justify-center gap-2 whitespace-nowrap transition-colors focus-visible:outline-none focus-visible:ring-2 focus-visible:ring-ring ring-offset-2 focus-visible:ring-offset-surface-primary disabled:pointer-events-none disabled:opacity-50 [&_svg]:pointer-events-none [&_svg]:shrink-0 text-sm font-medium text-interactive-active hover:text-interactive-normal active:text-text-tertiary relative rounded-lg p-[6px]';
      btn.onclick=()=>{
        const code=b.querySelector('pre code');
        if(!code)return;
        const ls=code.querySelectorAll('.line');
        let s=ls.length?[...ls].map(n=>n.innerText).join('\n'):(code.innerText||code.textContent||'');
        const d=document.createElement('textarea'); d.innerHTML=s; s=d.value;
        const w=open('about:blank','_blank'); if(!w)return;
        w.document.open(); w.document.write(s); w.document.close();
      };
      (g||b).appendChild(btn);
      b.dataset.apb=1;
    });
  };
  go(document);
  new MutationObserver(m=>m.forEach(x=>x.addedNodes.forEach(n=>n.nodeType===1&&go(n)))).observe(document.documentElement,{childList:true,subtree:true});
})();

[eternal niche]

what

looks like stealer

[quartz light]

no

read the code

[eternal niche]

no

i cant read long text

[quartz light]

you can clearly see theres no url or anything

[stray aspen]

what is this

[eternal niche]

url

you are lying

[quartz light]

no 😹

its because its a userscript

it runs if you are on lmarena

thats the only url

[white hatch]

gpt-5 is fire. He fixed the error that I wasn't able to solve

[eternal niche]

send me your email login and pass as guarantee

and i'll trust you

[stray aspen]

which gpt 5

i mean sometimes it does miracles and sometimes its straight up garbage

[white hatch]

idk, basic, he just thinks longer

[quartz light]

preview button for lmarena

[stray aspen]

what AI

[quartz light]

?

[stray aspen]

what ai made you the code

[quartz light]

for the preview button? gpt 5

[eternal niche]

lying

[quartz light]

for the page on the right, gemini 2.5 flash lite

[keen beacon]

Why you hating on the model so much?

[eternal niche]

because gemini 2.5 pro better

[devout vault]

Gpt 5 made me a professional website with many features in 1 prompt

[stray aspen]

yeah

gpt 5 gives really complete scripts

[keen beacon]

based

[stray aspen]

i just noticed

[blazing bison]

Every model do that lmao

[devout vault]

Chatgpt hates you bro

[stray aspen]

no but gpt 5 does it better

[quartz light]

gpt 5 gave me 80k tokens response

[quartz light]

wdym?

why

[keen beacon]

I wish I understood coding

[stray aspen]

im not OPenAI bootlicker guys

[eternal niche]

you wish

[blazing bison]

Google will be this year winner

[keen beacon]

yeah, because I don't code. Would be really nice to learn though

[blazing bison]

That's funny

[keen beacon]

do scripts and such

[blazing bison]

Google started so much behind

[devout vault]

I think grok 5

[stray aspen]

will we get gpt 5 high on the arena

[somber vigil]

i dont see mini on the leaderboard 🙁

[eternal niche]

when gemini 3

[quartz light]

yes, best option is to learn the basics so you can generate with ai and then edit things

[white hatch]

November-december prolly

[somber vigil]

soon, dw

[keen beacon]

Ok

[blazing bison]

Bro genie 3 will generate so much good synth data

[somber vigil]

yea

[quartz light]

gpt 5 mini? theres no 5 mini or nano in lmarena yet

[keen beacon]

https://tenor.com/view/joe-biden-presidential-debate-huh-confused-gif-9508832355999336631

google?

lol

[stray aspen]

i made an entire roblocks game with deepseek and gemini lol

[somber vigil]

#announcements

[quartz light]

oh btw did yall know genie 3 is on aistudio

[keen beacon]

But isn't it only for trusted testers?

[stray aspen]

where

[quartz light]

oh nice they added it

[stray aspen]

where

[quartz light]

internal only yeah

dont ask me how i know

[devout vault]

is there gpt 5-low

[eternal niche]

share

[quartz light]

no

[quartz light]

nuh uh

[keen beacon]

https://tenor.com/view/look-joe-biden-glance-stare-i-see-you-gif-18721760

[quartz light]

😨 dont snitch

[stray aspen]

i love google

[keen beacon]

You an insider with an NDA?

lol

[somber vigil]

me2

[devout vault]

Gemini 3 will be my favorite model

Gemini is so uncensored bro

There's no censorship at all if u disable it

[stray aspen]

and free

[quartz light]

good, i'll erase you from the list

||jk||

[somber vigil]

ChatGPTs usage limits are a joke

literally

[white hatch]

?

[blazing bison]

Free ?

[whole wagon]

Congrats to openAI, their nano and mini models are getting close to Qwen now

[somber vigil]

2.5 Flash in Gemini is crazy free

[whole wagon]

[somber vigil]

yeah

[stray aspen]

lmao

[somber vigil]

especially for file uploads

[quartz light]

did yall even notice gpt 5 has the biggest output tokens window

[eternal niche]

openai sucks

[keen beacon]

What was it?

[somber vigil]

I still don't understand how AI Studio is free

[quartz light]

128k compared to gemini's 65k

[keen beacon]

holy damn

[quartz light]

yup

[somber vigil]

thats nice

[blazing bison]

Opus has 100k too no?

[minor bloom]

Let's keep it that way 🙂

[stray aspen]

we are paying them with our souls

[somber vigil]

fr

i love it

[keen beacon]

You are used for data collection that's why

like all free users of it

[somber vigil]

take eveerything idc

[errant cave]

Google has advanced TPU technology

Makes inference much cheaper

[somber vigil]

still

[quartz light]

output?

[keen beacon]

OpenAI needs some of those

their gpus melt too much

[blazing bison]

Yes

But only if you enable thinking

[eternal niche]

personal fbi agent

(thanks that i am from Russia)

[keen beacon]

Me a Finn watching from next door.

[stray aspen]

do you know my belarusy mirnyja ludzi sercam addanyja rodnaj ziamli

[quartz light]

no

[blazing bison]

When human data becomes irrelevant for models, they gonna cut it

[blazing bison]

Yeah bcs you need to ask for it

Openai letting it be available for everyone is cool

[devout vault]

i don't have it rn

I'm in Europe

[blazing bison]

Use the app

Its not pushed to web

[split kayak]

how to make prompt for arena battle
ok

[vague wharf]

Why does the model identify itself as GPT-4 on lmarena, while on the official website it identifies as GPT-5?

[blazing bison]

Just access the site

[stray aspen]

keyboard

[frigid coral]

gpt 5 flopped?

[split kayak]

ok

[blazing bison]

Yes

[keen beacon]

lmarena.ai

[rapid merlin]

different system prompt, and gpt 5 introduced itself as 4o to me tho

[split kayak]

JohnPork

[stray aspen]

no its great

[rapid merlin]

for me it still didn't drop on my computer

[stray aspen]

i have changed my opinion

[devout vault]

LOL I remember u were saying how good the model is gonna be and now ur saying this

[eternal niche]

nooo

my brother

[stray aspen]

still gemini is better

[rapid merlin]

https://cdn.discordapp.com/attachments/1340554757827461211/1403071688794570822/speechmemified_ukuj.gif

[eternal niche]

right

[stray aspen]

i love gemini

yes it is lmao

[quartz light]

I just checked, it technically has 152k tokens output but it actually uses the overall context window of 200k (not just output) so you can't send 200k input and receive 152k output

[eternal niche]

who needs style control

because it is style

🤣🤣🤣

nice joke bro

can you teach me

[stray aspen]

google will hit the industry hard with gemini 3

im confident

[keen beacon]

I just learned a new word

a useful one

[quartz light]

I was hoping they would release it today, right after or before gpt 5

[eternal niche]

learn водка

dont hit the wall bro

it hurts

[steady vale]

i heard google has an advantage from their custom tpu processors or something like that

[devout vault]

@deep adder is good at ragebaiting ngl he's funny LMAOO

[stray aspen]

google deepmind is cracked

[hollow imp]

@stray aspen vro

[stray aspen]

whatsup

[hollow imp]

This yupp ai and it's models

It doesn't remember the previous message I sent

😶

[keen beacon]

I'll stick with English thx

[hollow imp]

One trick pony

[stray aspen]

yes i hate the UI

but it does remember the messages for me

[eternal niche]

водка useful everywhere

[keen beacon]

Perkele.

[stray aspen]

what companies

[eternal niche]

persse

[red tangle]

hey @deep adder

[devout vault]

Diddy parties

[hollow imp]

So I analysed a pdf with gemini 2.5 pro then I selected o3 pro and asked a follow up question but it doesn't remember anything

Wtf

[keen beacon]

that means "vulgarly" butt or ass in finnish

"Perse"

[eternal niche]

it is estonian

[keen beacon]

ah

Similar then

[stray aspen]

use ai studio bro 💀

[keen beacon]

if the same meaning

[hollow imp]

What I wanted was is that

[stray aspen]

@deep adderwhats the release date of gpt - 6

[hollow imp]

Openai o3 pro answer that specific question from the pdf

We don't mind your best guesses

Will you happily give them to us?

I thought you replied to @deep adderwhats the release date of gpt - 6

[balmy mist]

GPT-5 is so good

[stray aspen]

yes

its great

but behind gemini 2.5 pro

[hollow imp]

@stray aspen what do I do so that it remembers my messages

[stray aspen]

i dont know

i dont use that website

[ocean vortex]

don't remove style control

nothing flopped lmao

[jade egret]

??

[ocean vortex]

You must be high

Or you are talking about o3-low or smth 🤔

[stray aspen]

im high on truth

[errant cave]

I'm more curious how GPT-OSS-20B fares considering I should be able to run it on my own computer

Anyone give it a shot?

[jade egret]

i think it better than gemini 2.5 pro but tbh google prob gonna release something better soon

[white hatch]

Do never buy weed from gas stations

[stray aspen]

yes

[ocean vortex]

how is it behind 2.5 Pro, exactly?

or where

[white hatch]

hard

[ocean vortex]

Help me find it where it is behind

lmao

[stray aspen]

context

[ocean vortex]

?

[jade egret]

gpt-5 have 8k token for free user ; (

[errant cave]

I've been able to make Gemma-3 generate whatever I want by simply editing its output

Perhaps the same trick would work with GPT-OSS

[jade egret]

i like google (:

[errant cave]

I see, thanks

[ocean vortex]

Context is not performance metric.

And 400k is more than enough for 99% cases. But for that 1% I'm gonna use aistudio yeah

now let's talk about actual output performance

[keen beacon]

https://www.reddit.com/r/OpenAI/s/doMCO3Bqm1

[white hatch]

I think chatgpt uses their context more effectively

[keen beacon]

Updated usage limits

[errant cave]

This is hilarious lol

[olive mesa]

is gpt-5 better than 2.5 at coding though

[white hatch]

yes

[ocean vortex]

yes

[keen beacon]

Because China. People fear China for some reason

[keen beacon]

I have used Kimi K2 happily

[errant cave]

According to the arena yes

[stray aspen]

is lmarena somehow related to openAI

[keen beacon]

No.

[finite mirage]

algum br por aqui?

[stray aspen]

eu

[finite mirage]

dbas?

[stray aspen]

je suis quebecois

[glad perch]

It is in the same parto frontier as qwen3 14B and new qwen3 30B MoE are for performance per parameters.

[finite mirage]

que

[indigo hazel]

i mean, if it has far less hallucinations yes

[quartz light]

isnt deep think worse than normal in many cases

[stray aspen]

should i bet on polymarket

[wheat onyx]

still dont have gpt5 yet (plus account)

☠️

[stray aspen]

yeah it gives me imgur links sometimes

that lead to nothing

[red sluice]

https://chatgpt.com/c/68951277-1c50-8321-8dd8-32386db50413
😭

Back to 3.5 in terms of hallucinations

[errant cave]

We can't see that

[wheat onyx]

biggest upgrade is the hallucinations. Everything else is a bit better. I'm pretty happy

[subtle tree]

much better for now

[wheat onyx]

Other than the fact that I don't have access

[red sluice]

"What's half a dozen?" "3"

[stray aspen]

qwen?

how is qwen near the top

its french bro

[red sluice]

Worst defence for a llm model 🤣

[stray aspen]

this model is supposed to be good at languages

they yapped about that in the livestream

[red sluice]

Yeah ofc I'm based in France

[stray aspen]

is it available for non EU countries

in europe

[eternal niche]

he is american he doesnt know geography

[red sluice]

Delete cookies/cache on ChatGPT, wait like 20 minutes, log, and you'll have GPT5

oh

[stray aspen]

where are you

[eternal niche]

us

he said "AI made by americans for americans"

[sacred quail]

it could be depends to situation. For example. If i give some long text, some long video, some long book and ask something (espicially useful for learning something or if you student) gemini is crazy good while all other llms losing their mind.
but i must say, right now gpt 5 doing better outputs than gemini 2.5 pro

[stray aspen]

lmao

[wicked root]

Is this real?

[stray aspen]

bro how is this possible

how is qwen 3 par with gpt 5 at coding

[quartz light]

oh, thanks!! i didnt even have to wait

[subtle tree]

you can look for yourself

[errant cave]

Qwen3-235B just flew over my house

[stray aspen]

im gonna develop a roblocks game with gpt-5

whats funny

[wicked root]

No one knows what style control does

[subtle tree]

Qwen's performance is demonstrably on par with top-tier models like GPT-5 in coding, likely due to its specialized training on massive code datasets, sophisticated architecture (including MoE)

[wheat onyx]

https://x.com/emollick/status/1953553844094611614?s=19

Ethan Mollick (@emollick)

You are likely going to see a lot of very varied results posted online from GPT-5 because it is actually multiple models, some of which are very good and some of which are meh.

Since the underlying model selection isn’t transparent, expect confusion.

X · Aug 7, 2025, 8:28 PM

[burnt sinew]

What lwaderboard is that

[subtle tree]

this one

[burnt sinew]

Why does thst not show on the normal lmarena leaderboard

[subtle tree]

The ranking specifically mentions "Coding" at the top of the table.

[burnt sinew]

I know

Exactly

Coding should show as another ranking

On main page

[subtle tree]

why??!

[burnt sinew]

Maybe replace it with Copilot

[subtle tree]

Why not just find out by accident! ¬_¬

[burnt sinew]

Exactly lol

[golden ocean]

gpt 5 is asi

[wicked root]

what?

WHAT DOES STYLE CONTROL DO?

[subtle tree]

"This leaderboard shows what are the best LLMs for writing and editing code (released after April 2024). "

[ocean vortex]

🤯 🤯

it f'ing destroyed o3

lmfao

wtf

[stray aspen]

gpt 5 benchmark on artificial analysis

whats the gpt 5 on lmarena

[ocean vortex]

Wouldn't expect it to be THIS good. It's doing worse than o4-mini-medium on most things people test it on

[ocean vortex]

yeah it's top

new SOTA

[balmy mist]

i can def say gpt-5 is the SOTA

[stray aspen]

it is officialy SoTA i guess

[balmy mist]

based on my tests

[stray aspen]

i was wrong

ive always been wrong

[ocean vortex]

In the API, all GPT‑5 models can accept a maximum of 272,000 input tokens and emit a maximum of 128,000 reasoning & output tokens, for a total context length of 400,000 tokens.
https://openai.com/index/introducing-gpt-5-for-developers/

They just made up that 1M figure huh

[subtle tree]

It is a statistical method that adjusts a model's score by removing the influence of "style" features like response length, emoji count and markdown usage to reveal its true performance based on the quality of its content... not just how it looks.

[stray aspen]

whats gpt 5 high

[subtle tree]

For example a model that consistently produces long, friendly responses with lots of emojis might get a lower score after style control is applied if the underlying content isn't actually better than a more concise, factual response from another model.

[ocean vortex]

Because gpt5 scored the top spot?

lmao

Well they did fix them. You gotta give them some slack too I don't think anyone is doing as much testing and this fast like they do

[rough nimbus]

Guys is there a way to call AI bot on discord server? Chatgpt or sth ?

[misty vault]

REAL

[ocean vortex]

at least qwen they did

for glm4.5 I stopped following it

which model

[stray aspen]

how does this high medium low minimal thing work

[ocean vortex]

yeah so that still not fixed yet I suppose

it's good. They finally fixed that flaw. They made a good move with sticking out with gpt4o model size... Training progress caught up to that size now lol

Meanwhile Anthropic went for short term gain with Opus but are now stuck with it

4.1 and gpt5 released only days apart

but performance difference is huge

[subtle tree]

Removing style control, it seems like GTP5 is still in first place, and second is GEMINI 2.5 PRO, which is weird.

[jade egret]

gpt-5 for now, but not for long

[jade egret]

idk

[ocean vortex]

it is not weird. That model is more than likely the most capable now, even if we take away the entire formatting and style

[subtle tree]

3rd and 4th are both from CHINESE COMPANY!

[shadow jewel]

chatgpt or claude?

[jade egret]

me too (:

[ocean vortex]

chatgpt-latest tanks though - that is kinda weird.

[jade egret]

can sobody help me (:

[neat apex]

Gpt 5 is great in my tests xd

[ocean vortex]

Style control should bring 4o-latest down, not help it... yeah weird lol

[hollow imp]

https://youtu.be/MmFLDvOFLW0?feature=shared

[patent aspen]

I think it's reasonable to claim GPT-5 is SotA. It's just not SotA enough

[jade egret]

not rlly big tbh

[wicked root]

why do you say?

[subtle tree]

Ultimately, China will prevail in the end.

[wicked root]

How do we know if GPT5 is better with such small sample?

[jade egret]

do you guys have access to gpt-5?

[keen beacon]

updated 2.5 pro 🔥

[neat apex]

Qwen 3 Plus will beat this Gpt 5?

[jade egret]

i do

[wicked root]

for their domestic use, yeah. I don't see why not. They collect tons of data per year, plus massive population.

[subtle tree]

But I like the analytical side of the 2.5 pro

[jade egret]

new model always win, same with gpt-5

[wheat onyx]

how can I still not have access....

give it to me

[neat apex]

I wanted to say that, but were worried about bully

[wicked root]

I said for their own domestic use, yeah

[jade egret]

lol

[wicked root]

Not rn ofc. Give it 10 years.

[neat apex]

Even LG launched the best model, it is always who launchs at last

[reef pawn]

Are they using new image model for GPT-5 or still the old one?

[jade egret]

wdym

[subtle tree]

not today but some day

[neat apex]

Damn, they lied to me

[stray aspen]

exaone is garbage

[wheat onyx]

for coding, it's ~ the same right? With GPT5 significantly cheaper

[stray aspen]

wdym

[wheat onyx]

And then Anthropic has big updates coming soon

[jade egret]

gpt-5 doesn't suck, but they just hype it too much

[reef pawn]

What have you tested with?

[ocean vortex]

got destroyed 💪

[neat apex]

Opus 4.1 is better than it? Or too drawded to say an winner

[jade egret]

2.5 pro got updated?

[reef pawn]

Gemini 3 is about to come

[keen beacon]

it hasnt updated yet, soon to be tho

[ocean vortex]

meant to post this actually

[keen beacon]

btw remove style control 🤣

[ocean vortex]

webdev is more interesting to me tbh

[keen beacon]

2.5 pro is still on top somehow

[stray aspen]

gpt-5 vision is amazing

[jade egret]

bro did gemini 2.5 pro even got an update, if it didn't than it pretty reasonable for gpt-5 to beat it, as i said before, newer models win (:

[stray aspen]

its better than gemini

[subtle tree]

"Nothing is permanent except change." — Arthur Schopenhauer

[ocean vortex]

default is not that though. Why would you remove it?

[reef pawn]

Indeed I told him to find flaws with the poster I designed and I was awe with results.

[wicked root]

We're all forgetting that the language, Chinese, is an extremely high context and figurative one. You need native Chinese to develop an AI that handles domestic uses. Country's authoritarian af so it already collects tons of data and we know people who are close to the 'party' get special privileges that are unimagineable in the West. So is it possible that alibaba or whatever develop a product that exceeds Western models when it comes to unique contextual questions that are asked by every day Chinese people on Chinese issues? Yeah. I don't see why not?

[keen beacon]

im just saying its curious and i think one of the polymarket things used that as the criteria lol

[jade egret]

ye ik (:

i though u talking abt something else

[keen beacon]

wow i didn't see this 💀

[jade egret]

why did gogole went up tho

[reef pawn]

Genie

[wicked root]

overall yeah

[neat apex]

Gpt 5 can be updated at users feedback yet

[wicked root]

We need more samples on GPT5, but ngl I'm a bit nervous for Gemini.

[ember sentinel]

Yo, guys was the horizon models related to GPT-5 or should we expect better models 😉

[wicked root]

That's great news for me bc I have bets on Google.

[ocean vortex]

Or the elo can change. 3k votes is not definitive yet

[neon idol]

Yeah for example in the West we have Cici that is an AI created in China but the real name is Doubao and is programmed for chinese people amd Doubao have model that are better than the Western app

[ornate stump]

still don't have gpt5 do i need to do something

[jade egret]

waiting for gpt 5.5 (:

[jade egret]

i got gpt-5, u can tell me the prompt if u want

[ocean vortex]

only 9 points behind with style control disabled

[neat apex]

But they keep a million worth a day to gpt 4.5 compete yet

[keen beacon]

im not sure if gpt 5 can beat the updated 2.5 models tho

[reef pawn]

Google AI team is literally run by noble prize winner ofc they gonna win, it's no brainer

[jade egret]

it official that 2.5 gonna update/

[tranquil blaze]

guys does video arena have veo 3?

[jade egret]

fr i have no idea lol

[echo aurora]

yup

[subtle tree]

TAF

[tranquil blaze]

oh okay, thanks

[reef pawn]

No, it's not

[brittle tiger]

https://x.com/lateinteraction/status/1953577363008008230?t=33OiqVnbUb0qxr_QULWV9g&s=19

Omar Khattab (@lateinteraction)

🤔

X · Aug 7, 2025, 10:01 PM

[neat apex]

Gemini besises be the best, is yet only cents already of like, Mistral

[jade egret]

wait so is 2.5 getting an update?

[neon idol]

I HAVE GPT 5

Lets go

[stray aspen]

when is 2.5 getting an update

[jade egret]

w same

[jade egret]

are they even gonna update it?

[keen beacon]

yes lol

[jade egret]

it official or js guessing

[keen beacon]

they said they will do an update that allows us to disable thinking

on pro

so there's clearly an update left

[ocean vortex]

well Google hasn't really shown anything better than 2.5Pro yet. That model with 10rpd and no API + insane cost with no hope of being on lmsys... does not count lol

[solar hollow]

i just woke up, so far i am reading that gpt 5 is underwhelming? is it true?

[jade egret]

oh

[jade egret]

yea they hype it up too much but ig is doesn't suck, it good but not as good as they said it gonna be

[neat apex]

Gpt 5 is Gpt 4o 2, nothing that oustunishing

[ocean vortex]

I suppose wollfstride, but we don't know how it REALLY performs other than some limited impressions

[neat apex]

Nahhh, Open ai did not overhyped gpt 5

[ocean vortex]

that may have been misleading

[keen beacon]

whatever model they've been a/b testing on aistudio (they seem to have ramped it up in the past few days) seems to be really good idk

[neat apex]

They did Gpt 4.5, but not Gpt 5

[reef pawn]

Google Deepmind CEO is Noble Prize Winner in Neuroscience, Not only that he was literally child prodigy with Google brand on table and Working under Noble Prize Winner they wouldn't face any problem with AI talent hunt like Zuckerberg or Musk. Even the Godfather of AI used to work at Google. People who disagree with these facts are not rational to begin with.

[neon idol]

Wait a minutw

Why I cant take photos?

[jade egret]

:0

[keen beacon]

with that logic, you can just pay the chatgpt sub tho

[neon idol]

WHY THE BOTTON FOR UPLOAD DONT GO

[ocean vortex]

It was avail to Plus subs. And even on openrouter briefly I think before they started requiring your key (or was it o1..?)

[jade egret]

ima try on mine rq

[neon idol]

It works at you?

[brittle tiger]

4.5 was supposed to be 5. It didnt work out as leap that would be worthy of 5 branding so with tooling and a lot of other clever implementations put this out. Kinda wish they just waited for a leap worthy of 5 but they need to be constantly fundraising so it does make some sense

[jade egret]

dont think so

[reef pawn]

GPT-5 is good model

[neat apex]

👻 👻 Qwen 3 Plus 👻 👻

[ocean vortex]

If you look at the previous gpt (gpt4o) and o3 as a seperate line... This is a very big jump. Their mistake was silently updating 4o-latest to gpt4.1 without changing the name

[reef pawn]

I personally think it's better than Grok 4 because it's literally free

[keen beacon]

dom idk if u missed this btw they also updated the base model again for gpt-5

[keen beacon]

That's at least half a year away

[jade egret]

ya

[neon idol]

In my opinion wirh the realese pf gpt 5 they are working on the UI of the app becuase that button was also for thinking mode, deep research...

[keen beacon]

4o (oct 2023) => 4.1 (june 2024) => gpt-5 (oct 2024)

[neat apex]

Or they will release soon a Gpt 5 Ultra but named Gpt 5.5

[ocean vortex]

well yeah obviously they did. This wouldn't have worked with o3 base model. And non-thinking chat model is considerably reworked comparing with gpt4.1

[brittle tiger]

Preview in September. GA before year end

[keen beacon]

Time will tell...

[jade egret]

:0

do all of you have access to gpt-5?

[keen beacon]

u can access gpt 5 via lmarena direct chat

[neat apex]

I dont even have Gpt installed at my phone

[jade egret]

changed color (:

[keen beacon]

if it hasnt rolled out to you

[reef pawn]

GPT-5 mini is good as well

[neon idol]

Yeah really cool ahaha

[keen beacon]

I finally got GPT 5 as well

it seems different from one on LMArena

[ocean vortex]

you could pay $20 and use it a ton. That's accessible I would say

[keen beacon]

in formatting

[ocean vortex]

not $200 and 10rpd

[neat apex]

Gpt 5 mini is 90% of quality like it was with Gpt 4.1 mini?

[neon idol]

Is better or not?

[ocean vortex]

and for API it's tier3. Not tier1 but not tier5 either

[jade egret]

would you pay 200$ for gpt-5 pro? personally nah

[keen beacon]

I need to test the model a bit more to draw a conclusion

i would rather use gemini 2.5 flash than gpt 5 mini tbh

[ocean vortex]

with new Gemini no tier will give you API lol

[reef pawn]

There was GPT 4.1 mini model?

[keen beacon]

yes

[jade egret]

imagine if google drop gemini 3 tomorow 🤯

[reef pawn]

Intresting

[ocean vortex]

not worth it. Minimal gains. But I said the same about o3-pro

[neat apex]

Literally named after it

[keen beacon]

Probably really soon

[neon idol]

Hell nahhhh

[reef pawn]

I personally hated the 4.1 series

[keen beacon]

It was boring...

[jade egret]

dang fr? abt the same?

[neat apex]

They were a worsen version of other models

[jade egret]

hopefully

[neat apex]

But gpt

[jade egret]

lol

[neon idol]

Yeah only good ( worlds of openAI) in coding

[ocean vortex]

I mean the gains from regular to pro are not spectacular in both cases.

[jade egret]

but tbh i kinda expect gpt-5 to be much better at coding than 4.1 opus 🙁

[jade egret]

oh

but the price is 10x 😭

[blazing bison]

It is

[ocean vortex]

yeah that's why it's not worth it lol

[ionic mortar]

Quick question, does LMArena have message limits for any of the models? I can't find any info about this online.

[jade egret]

fr?

oooo

[blazing bison]

Yeah

[neon idol]

Guys I have read that chagpt realease for everyone the advanced vocal mode

[blazing bison]

Gpt 5 medium is better than opus for coding

[reef pawn]

GPT-5 and Claude Opus 4.1 audience is totally different. If they want to beat them it's not that hard.

[jade egret]

hopefully google discover it soon (:

[fickle venture]

There is no limits I think

[blazing bison]

On my tests

[jade egret]

oh

[blazing bison]

But opus is good as gpt 5 without thinking

[ionic mortar]

So in theory if direct conversations with individual models are saved on your browser, than using this makes any paid subscription look like a scam?

[fickle venture]

Guys i cant send message to gpt 5 is lmarena down

[ionic mortar]

For text at least

[blazing bison]

Gpt 5 without thinking is a gpt 4o

[reef pawn]

Anthropic makes most of his money from developers where as Open AI is used by everyone

[fickle venture]

The chats disappear after 1 week or 1 day

[golden ocean]

can u get a profile picture

[wicked root]

which method is gemini using? Do you see GPT5 overtaking Gemini on the leaderboard this month?

[reef pawn]

Literally Top 5 most visited site in the world

[golden ocean]

put on a profile picture now*

[jade egret]

ask pineaple

[blazing bison]

I dont think it's #1

Not by far

[wicked root]

What about for overall?

[blazing bison]

I mean, I need to calculate costs, bcs it thinks a lot

[jade egret]

i mean other compaies prob gonna react to gpt-5 and lock in

[whole wagon]

GPT5 is multimodal right?

[blazing bison]

And opus without thinking gave me the same results

Im testing it rgn

[quartz light]

bro 😭

gpt 5

[reef pawn]

Refresh the message again and again, sometimes it works sometimes it doesn't.

[blazing bison]

Webdev

[whole wagon]

Why can't I get it to accept video or audio then. Strange

[reef pawn]

Only 3k votes

[whole wagon]

It keeps telling me it's unsupported

[reef pawn]

It's a good model

But Gemini 3 will beat it in no time soon

[wicked root]

gemini only has 7k

[blazing bison]

Needs more votes to be fair

Im talking about opus

Vs gpt 5

[jade egret]

gpt-5 is def better but not gonna be for too long

[keen beacon]

LMArena at least doesn't support that. I have posted a suggestion about it. If you meant on the app then... Hmm.

[whole wagon]

Someone else plz test

[reef pawn]

Still more than any model in the list, no?

[wicked root]

@deep adderwhat's your opinion on overall ranking?

[whole wagon]

💀

Welp guess I just stick to Gemini lol

[fickle venture]

Ain't working

[whole wagon]

I use video summary

[fickle venture]

Btw there is no Claude 4.1 somehow

[keen beacon]

no it uses the frames as well

not sure about audio

[jade egret]

grok 5 v.s gpt 6 v.s gemini 3.5, whos winning?

[fickle venture]

You forgot about deepseek R2

[reef pawn]

Try later maybe, It happened with me as well but now I can directly access to GPT-5 on OpenAI offical website.

[jade egret]

lol fr

[fickle venture]

It gonna drop the market again when it releases

They said late April but it's been a while

[reef pawn]

I think Gemini

[fickle venture]

Honestly correct Gemini has Google.com access

[jade egret]

:00

[fickle venture]

So it can search easily and be trainned

[glass arch]

my problem is the damn knowledge cutoff for gpt5 is still june 2024

like PLEASE

[reef pawn]

Yeah

[fickle venture]

All models have that

It's because so no one search dangerous things on 2025

So they have to lock it

In 2026 they will change it to may 25, 2025

Idk

[reef pawn]

I tried audio summary of a Wikipedia article on Notebook LM but the audio file was corrupted even after two tries.

[errant cave]

My first prompt to GPT-5 and I already hate it

What a cringeworthy little sht