general

[eternal niche]

who needs style control

[zealous panther]

… ok

[eternal niche]

reading for stupid

[zealous panther]

???

Alright

[eternal niche]

i have gemini 2.5 pro

who needs reading

[zealous panther]

I have both gemini and chatgpt lol

[eternal niche]

nice bro

[keen beacon]

He's just a troll

dont care bout him

[zealous panther]

Ilr

I know

Im not raging dw

[obtuse heart]

much better if you ignore him so he doesnt get entertainment

[rare python]

@patent aspen are you confident about gemini 3?

[zealous panther]

Hmmm it depends really nobody knows

[eternal niche]

even gemini 2.5 pro better lol

obviously gemini 3 would be better

[zealous panther]

I mean it can either be a plateau and so on

[quartz light]

the main problem for me is that long responses never fully work, it only worked once so far

[zealous panther]

If openai hit a wall then maybe all labs might hit their own walls

Or the same wall

[stray aspen]

but google labs have better technology

[zealous panther]

The talent is crazy

[quartz light]

when is gemini 3 releasing

[zealous panther]

I think the range is 2 weeks to 4 weeks

Ehhh actually…

[eternal niche]

yes and gemini 2.5 pro just better

[zealous panther]

Yeah

I really want gemini to be better though

[obtuse heart]

if gemini 2.5 rn is comparable to gpt-5 then its gonna absolutely be good

[zealous panther]

Its context is just so stupid

The inchat memory istg

[eternal niche]

not comparable

its better

[zealous panther]

It doesnt remember istg

[obtuse heart]

gemini?

[rare python]

gemini app is cooked. Too much spaghetti code

[zealous panther]

YEAH

[eternal niche]

lol

[rare python]

but Gemini on API is good

[zealous panther]

And the god damn tool calling bro…

[obtuse heart]

for me gemini has had the greatest memory

[solid brook]

I think bro's dad works at google

[zealous panther]

Used both

[eternal niche]

my brother

[pure falcon]

Gemini vs Zenith - who wins?

[zealous panther]

Idk

[eternal niche]

zenith is gemini 3

[zealous panther]

Ive used boh

Gemini tool calling bro istg… it makes my blood boil, it gets so much info wrong without searching once even if i say “search”

Like…

[eternal niche]

subscribe

[zealous panther]

I have pro

[rare python]

this is a known issue

That's why Gemini CLI is bad

[eternal niche]

my brother

[rare python]

bad tool calling

[zealous panther]

Yeah like i want it to be better than gpt but its these little things

[maiden fulcrum]

Do you guys think Gemini 2.5 Pro is better than ChatGPT-5?

[solar nebula]

yeah tool calling is pretty much the only issue with gemini

[zealous panther]

Its these little things that makes me default to gpt

[rare python]

I expect them to perfect tool calling since they released Gemini CLI

[pure falcon]

No, zenith is a stronger GPT-5. OpenAI went with Summit, I imagine because of cost

[zealous panther]

Istg yeah

[maiden fulcrum]

@zealous panther

[zealous panther]

Im pissed thouh pike

[zealous panther]

No

Overall its a no from me

Im still using both

Gpt 5 vibes feel better

[maiden fulcrum]

I really need to learn how to use OpenAI's API

[zealous panther]

And im a google shill lol

[zealous panther]

I mean you really need that extra processing ?

[maiden fulcrum]

I mostly use AI Studio

[rare python]

gpt 5 has one flaw I don't like is the concise, jargon dense writing style

[maiden fulcrum]

What do you mean?

[rare python]

It's so hard to understand GPT 5 unless you are also knowledgable about the topic

[patent aspen]

The worst possible scenario I could imagine for Gemini 3 would be if the entire AI industry is plateauing at the same time and GPT-5 was just the first sign

[obtuse heart]

isn't that why they added the new tone update no?

[eternal niche]

btw guys google will buy openai

[solid brook]

One other thing. I think they hardcoded a limit to gemini 2.5 pro. It cant generate more than 500 lines of code in one response

[zealous panther]

Preferences

[zealous panther]

I was talking about that too since i think thats very likely

Like openai isnt just gonna fall off

[patent aspen]

I mean they very well could

[zealous panther]

If its the small percentage that google deepmind cooks and releases good stuff

Im lovin it

[whole sundial]

https://fixupx.com/JustinLin610/status/1953836466997801401

Junyang Lin (@JustinLin610)

this is it!
︀︀
︀︀it means that u can use qwen code for free unless u need more than 2000 runs every day!
︀︀
︀︀i hope u can better enjoy qwen3-coder through qwen code!

Quoting Qwen (@Alibaba_Qwen)
︀
💡 You get 2,000 free Qwen Code runs every day!
︀︀
︀︀Run this one simple command:
︀︀npx @​qwen-code/qwen-code@latest
︀︀Hit Enter, and that’s it!
︀︀🚀 Now with Qwen OAuth support — super easy to use.
︀︀Try it now and supercharge your vibe code! 💻⚡
︀︀Github：github.com/QwenLM/qwen-code

**🔁 2 ❤️ 3 👁️ 26 **

FixupX · Aug 8, 2025, 3:11 PM

[zealous panther]

Yeah but the chances arent as much as its a general wall for all ai labs

[solid brook]

It will be bad for us. Competetion is good. If google beats everyone it would be pretty bad for us

[rare python]

@torn mantle

[zealous panther]

I sure as hell hope so too

[eternal niche]

google already did it

[zealous panther]

Openai has to be pushed again

We need another r1 and all

Nah though that webdev jump is just

Incredible

[clever estuary]

Bruh

[zealous panther]

Same as when gemini 2.5 pro was released

Even more impressive i think

[stray aspen]

whats gonna be chinas response to gpt-5

[eternal niche]

yes gemini 2.5 pro even more impressive

[clever estuary]

Seethe and cope

[zealous panther]

I mean they have like 100 ai labs ong

[leaden sun]

maybe it’s about time to switch to a complete new architecture?

[zealous panther]

Agreed

[eternal niche]

lets switch

[zealous panther]

If gemini 3.0 doesnt perform that is

We goota wait for that

[whole sundial]

z.ai is coming out with a new GLM 4.5 Vision model next week that I believe is capable of native image gen like 4o

[zealous panther]

I mean grok 4 failed expectations too

Soooo

[leaden sun]

I was expecting got-5 to be made from scratch with a new architecture, but what came out seems to be a strategic move I rather than showcasing scientific progress

[eternal niche]

loool openai RIP

[zealous panther]

Ffs yeah

[stray aspen]

z.ai is great

[rare python]

Read from this and below. GPT 5 has that same style as o3

#general message

[zealous panther]

Openai is in a really bad spot rn so if google does release they’ll have to release a really good model too and more

[solid brook]

I mean it attracted all the degenerates

[zealous panther]

Ani bro…

Like god…

[leaden sun]

Their glm 4.5 reminds me a lot of gpt 5 actually

[solid brook]

Yes truly "AI revolution"

[zealous panther]

Oh YEAH SOMETIMES O3 WRITES GOD DAMN LIKE IDK I CANT UNDERSTAND IT

[patent aspen]

I think it's more multi-faceted than that. If you do the same thing long enough, there will be diminishing marginal returns. I think we're starting to see diminishing marginal returns in certain categories while other categories have tons of room for growth. Ultimately I think that means progress going forward will be more jagged. We'll see more improvements from research breakthroughs, architectures, etc and less from post training hill climbing.

[zealous panther]

youre asking me to have a phd to understand it ??? Like what

[rare python]

yeah that's what I'm talking about

[zealous panther]

I havent encountered the same provlem for gpt 5 for me…

[zealous panther]

I think so too.

[whole sundial]

they never said this explicitly, but they responded to one person saying that they will release an improved vision model and they also said to me when i asked about image gen that there will be "an exciting launch in 3 days" (this was a few days ago before they pushed it back to likely Monday) and that I "won’t be disappointed"

[zealous panther]

There needs to be more researching done and such

I also think that gpt 5 is a step in like usability

Gemini 2.5 pro has a provlem with overthinking at times

[whole sundial]

so I strongly believe that they will be coming out with a large GLM 4.5 vision model that also has 4o-like image gen capabilities

[zealous panther]

Question though

Do you all think gpt 5 would have gotten a better score if it was thinking for everything ?

Gemini 2.5 pro did that

[whole sundial]

which is honestly needed because it should give OpenAI some tough competition in image gen

[solid brook]

Gpt 5 high activates just from saying think hard right?

[whole sundial]

people already really liked GLM 4.1V, and that was a 9B model

[zealous panther]

Not even sure

Ill check on that

[solid brook]

I do think it might be. On lmarena when i tell it to think hard it thinks way longer

[zealous panther]

Yep no odea

[zealous panther]

Same on the app but it just might be med or something

Theres thinking med too

So yeah

[leaden sun]

They’re supposed to publish a technical report on glm 4.5 this week… maybe they’ve postponed that?

[zealous panther]

The slander in x and this disc about openai bro…

[wicked root]

Anyone know when the next set of votes will be added to the leaderboard?

[zealous panther]

Nope depends on the team

[solid brook]

Guys why do i feel people get paid to sht on gpt 5?

[stray aspen]

craig how do you make gpt-5 use high reasoning effort

[solid brook]

.

[leaden sun]

so the rollout of new architecture is reminiscent of the process of moving from on-premise to big data cloud? I see the parallels, it’s a safe and careful approach

Not without embodiment am afraid

[stray aspen]

if you tell gpt-5 to think very hard it solves the siimple bench questions correctly

[pliant cliff]

wait gpt-5 is peak?

[terse shuttle]

the best

openai cooked

but gemin 2.5 deepthink is the same i think

[leaden sun]

AI cannot be commodity if water and electricity are scarce, and the current geopolitics point to even more scarcity….

[hollow imp]

Same as what

[terse shuttle]

as gpt-5

[hollow imp]

Btw guys why this yupp ai "o3 pro" feeling worse than o3 and 2.5 FLASH

@deep adder what do you do in your work hours?

[stray aspen]

chat on lmarena

[hollow imp]

I agree

[leaden sun]

Human body is not designed to sit in office 12h a day, we should play outside and enjoy the sun shine more, imaging those bankers and politicians having more sunshine in their life, world would be so much more peaceful right now? 😂

Oops sorry

They’ve tried to hire me, but I was too lazy to do their assessments 😅

[echo aurora]

We've been seeing really nice weather the last few days, I'm looking forward to getting out this weekend

[keen beacon]

Education system indoctrinates
Work system enslaves
Your labor is only for the benefit of oligarchs and by definition against that of society and yourself

[hollow imp]

And have you realised your TLP?

Your life purpose

@leaden sun where'd you go

See this

Nah it's not like something that will immediately click into your mind. Some people take days to figure out this. They start thinking about it as soon as they get the time to.

I also think it's really important

[leaden sun]

nice quote there ✨

[hollow imp]

It's not a quote

It's a personal experience

Of Robert greene

[leaden sun]

Take his advices with a grain of salt will ya?

[delicate rapids]

anyone tried gpt 5 vs claude 4 sonnet

for coding

many people r saying gpt 5 is better than claude 4 sonnet

did u compare with opus

can u elaborate more why

why it is better

[stray aspen]

because i tested both

and gpt-5 won

[delicate rapids]

for backend?

[patent bane]

is it available for Team users?

[delicate rapids]

idk,i didnt test this model much due to its bad initial results

[jagged abyss]

Hey Hi all , I am new to all this.

Help me out in exploring this AI world.

[patent bane]

you gotta figure out yourself

[echo aurora]

Hey sorry for taking awhile to get back to you on this! Got confirmation it's the reasoning effort high version.
@torn mantle @fading moth

[jagged abyss]

I am more interested in rendering videos from script through AI, any help will be appreciated.

[echo aurora]

welcome welcome! You should check out #1397655624103493813 for more details on how to use Video Arena

[prime mulch]

@echo aurora why isit violation

[stray aspen]

its a dude swaallowing a sword

[prime mulch]

No

This is the image

[stray aspen]

idk

[dusky aurora]

I think it sees the ear as a female breast

[prime mulch]

🤣🤣🤣🤣

I think it saw this image as something else

It works with other images

Thanks

[echo aurora]

I'm not sure tbh

We've seen feedback that the content filter can be a bit overzealous; however, that's mainly been for text based prompts. I haven't seen too many reports of false positives being flagged for images.

[prime mulch]

Yea it works with other images i think it thought this image as something else

[primal orbit]

content filter for text is ok now

[echo aurora]

Yeah there were some changes to reduce the false positives

[flint skiff]

tried both gpt 5 on highest reasoning and opus 4 and tbh opus 4 just feels better

I even prefer sonnet in cc over this

I tried cursor, roo and kilo code but all of them fall short

not sure about debugging yet though but for actually building UX and building apps from scratch its a pain compared to using cc

[delicate rapids]

Gpt 5 > claude opus 4+4.1

[flint skiff]

didnt you literally just say you dont like it?

[flint skiff]

this

[delicate rapids]

that was grok 4

[flint skiff]

oh lol

what do you like more about gpt 5 though

[ocean vortex]

"research grade" lmfao

It's like GPT5 only ran 10 times, like, give me a break...

[neon idol]

OMG

Thank you very much

[echo aurora]

No problem! Apologies I didn't have this information for you yesterday

[neon idol]

Dont worry

[obtuse heart]

mines the opposite, tried gpt 5 ( default ) and its by far a mile better than opus 4/4.1

[delicate rapids]

hi pineapple,why r u pineapple

[flint skiff]

oh wow interesting

[obtuse heart]

i think prompts are also a factor on the output

[echo aurora]

an old coworker made me the avatar one day with google slides and I loved it

[flint skiff]

whats ur prompt pipeline

u just write super detailed?

[obtuse heart]

i usually write a detailed prompt as a base, then get gemini 2.5 pro to improve it, and then polish and it works like magic

[flint skiff]

hmm alright ill try that

[obtuse heart]

does that bypass the ai detectors lol

thats hella insane then

jeeeezz

[ocean vortex]

opus4 is leagues behind. There can be some niche things it's still better at but at this point it's almost like comparing gpt4.5 to gpt5.... Not really productive or useful and those very limited narrow things do not outweigh the vast gpt5 advantages

[teal mantle]

I might want to try chatgpt team versus supergrok versus google ai pro
Which one then
But team takes two people

[obtuse heart]

its a long one too so they have a lot of info to work with, yet it still bypasses

gpt-5 is really great lol

[delicate rapids]

guys how to make whatsapp bot

[flint skiff]

how do u guys use gpt 5 tho?

like what pipeline

[delicate rapids]

is there dev portal for whatsapp

[hollow imp]

Microsoft copilot?

[obtuse heart]

lmarena

[flint skiff]

for coding

[obtuse heart]

i just use web based for now cus idk how to integrate it into cline for free

[flint skiff]

cant you just use ur api tokens? its cheap anyways

[obtuse heart]

yeah but im a cheapskate, free is free

didnt they announce that cursor will be offering free gpt-5 for a week?

[flint skiff]

its free right now

but medium reasoning

nerfed pretty hard I think

[ocean vortex]

the one where it's auto switching between thinking and not is only avail on chatgpt website I believe....

[obtuse heart]

unfortunate

[flint skiff]

u need to be a paid cursor user though

[ocean vortex]

I wonder why no API for it. No metrics even

it probably doesn't work nearly as well as thinking always lol

[torn mantle]

ah i see, thanks 😊

[teal mantle]

Should I try to get gpt team, supergrok or google ai pro
Or settle with plus?

[obtuse heart]

what about kilo code, arent they offering 20 dollar fee, for free?

[teal mantle]

Need something with abundant usage for non-coding

[flint skiff]

I got 5usd for free on kilo

[obtuse heart]

supergrok is pretty bad at that lol

hella slow

[delicate rapids]

gemini 2.5 pro = gpt 5

[ocean vortex]

Plus. Pro or Ultra is waste of money. SuperDork is not an option

[teal mantle]

how about google ai pro, do they give enough 2.5 pro?

[obtuse heart]

i think you should settle with 2.5 pro for non-coding usage

[delicate rapids]

gpt+ is best overall

[teal mantle]

I meant for usage limits since I would like to use something isn’t ai studio free (they still train on the inputs I think) or the dogshit free quota of 2.5 pro

[ocean vortex]

It sounds neat on paper and at a glance at it. But if it performed without compromises I'm sure they would have included metrics... Also would be no point to include thinking always version in model switcher

[obtuse heart]

ah then idk really, might be best to ask someone else whos more knowledgeable on that part

[ocean vortex]

I mean... it's probably gonna use search without thinking a whole lot. So that would be closer to gpt4.1 search than o3. Though I haven't tested this in detail yet

[hollow imp]

2.5 pro is dogshit?

[teal mantle]

The free rate limits is

[hollow imp]

The Gemini web one?

[teal mantle]

Also the app one
I just wonder what is the paid quota

[hollow imp]

App?

[hollow imp]

I was using Google ai pro free trial for 30 days and I didn't come across any stoppage or limit at all ( I didn't use it as much as these geeking ppl in the server )

[eternal niche]

because gemini 2.5 pro the best

[hollow imp]

https://tenor.com/view/monkey-music-monkey-monkey-listening-to-the-music-monkey-chill-vibin-gif-24320967

[hollow imp]

Vro I'm not going play with you PLEASE I'm busy

[minor adder]

??

[torn mantle]

on paper yea

[stray aspen]

stop ragebaiting guys gpt-5 is SoTA

[torn mantle]

but gpt5 doesnt have a 'deep thinking' feature like grok 4

they are using a more efficient method

like in gemini

grok 4 heavy spawns many instances called agents/like seperate models which takes a lot of time

meanwhile gemini deep think & gpt5 pro starts from a single node like ToT

oai is confusing all of us with these benchmarks

we have like 4 models

we really hit a plateau

this doesnt look good

i still have hope on gemini tho

[tawdry delta]

deep thinking on grok 4 is disgusting

[torn mantle]

its not really practical

you cant use it

[tawdry delta]

i still think gpt-5 medium being much much cheaper than grok 4 effectively

while being a tiny bit better for general use coding

makes it a good enough leap

[tawdry delta]

grok 4 non-heavy

[minor adder]

I'm currently using gemini for the most part.

[torn mantle]

same

i use it like 24/7

[delicate rapids]

Is it available on all gpt app now?

Ur one free?

I am feeling bad that i am relying on ai too much,like study,every day task,coding,images,logo

Ai is destroying creativity

[patent bane]

it is

it's a harsh truth and we have to face it

even though I use it a lot

I am a Team user and haven't seen gpt-5 pro yet

hmm

[wintry tinsel]

Same here I love spamming images and files for free

[maiden fulcrum]

Why ChatGPT-5 is solving 5.9 = x + 5.11 wrong?

[blazing bison]

Because routers like I said before the release, sucs

[neon idol]

What should be the amwer?

[blazing bison]

Where is the unified model @deep adder

[maiden fulcrum]

0.79 but it answered -0.21

[neon idol]

@echo aurora Sorry but on copilot the GPt 5 that they use is the gpt 5 most powerful model?

[blazing bison]

No, is the worst one

[neon idol]

Gimme prompt?

[stray aspen]

no

microsoft copilot has the worst gpt-5

[neon idol]

So the nano?

Or the mini?

[blazing bison]

No, is the gpt 5 chat

A router

That never routers

[neon idol]

So is like a fake gpt 5?

[blazing bison]

Test gpt 5 chat on api

Bro

You talk with 9999 different models

Its clearly

Even the speed of the answer change with the same prompt

[stray aspen]

is gpt--5 better than gpt 5 chat

[blazing bison]

On chatgpt it's a pure router

When you select thinking sometimes it's o3, sometimes o4 mini level of answers

You dont know which model they are routing you too

Gpt 5 is the worst release possible for consumers

[stray aspen]

so the API is better

[blazing bison]

Yes

[neon idol]

Bro I just want to know what is the gpt 5 that copilot use

[stray aspen]

the trashest gpt-5

[blazing bison]

Is the gpt 5 chat bro

I already answered it

[neon idol]

Ok but is the high medium or small?

[blazing bison]

Its impossible to know

Wait

Its the base model

Gpt 5 chat

There is no high medium or small

[neon idol]

So the medium

[stray aspen]

however if you tell it to think harder it gives better answers

[neon idol]

Ok

[blazing bison]

In the api the model gpt 5 chat is a separated model

[neon idol]

Thx but I think that I Will use gpt 5 on the orginal chatgpt app

[blazing bison]

It's broken

Its not routing correctly

Openai staff said they gonna fix it next week but anyway

Its the worst using a thinking model and dont know it you gonna pick the best one or the mini one

[maiden fulcrum]

Solve: 5.9 = x + 5.11

[wheat onyx]

yeah i think lots of people are using mini/nano when they shouldnt be

[blazing bison]

Exactly

[maiden fulcrum]

how?

that's not the app

[stray dock]

brief review of gpt 5 anyone (LM arena or native or both)

[stray aspen]

i am using the API

[maiden fulcrum]

it's free?

[stray aspen]

yeah on lmarena

or yupp.ai

its great

[maiden fulcrum]

i thought on their website

[stray aspen]

but tell it to think hard at the start of prompt

[maiden fulcrum]

how

[stray aspen]

you put "think hard"

[maiden fulcrum]

okay

[barren prairie]

The only good thing about gpt5 is that I can choose the covo color and turn it into pink

[eternal niche]

😭

[rich compass]

what the hell with lmarena

[tight tide]

Hello @tiny palm I am telling about your veo 3 video generation on YouTube and youtube removed my video is it illegal to tell this how to get veo 3 for free ?

[blazing bison]

Bro yupp is fake

[terse shuttle]

are u about this?

[blazing bison]

Its not free and its not unlimited

You are lying

[iron meadow]

Why did my chat history get cleared just now

[tight tide]

that was for clickbait nit telling to get unlimite d

[rich compass]

yeah

[terse shuttle]

@echo aurora what happened with website? why is connecting error here

[blazing bison]

Yeah but this can be the cause of yout video being removed

[echo aurora]

uh oh

[iron meadow]

@echo aurora Site out.

[tight tide]

so becaus eof title video can get remoevd

[craggy torrent]

is there an outage?

[echo aurora]

okay thanks

reporting

[blazing bison]

If people report too much yes

[tight tide]

ok so that means after review they can not het it back

[blazing bison]

Bcs people come here and see that it's arena and can't select the model they want

They reported you

[echo aurora]

Yeah looks like we're experiencing an outage. I'm really sorry everyone!

Team is looking into right now!

[mortal coyote]

@echo aurora

[echo aurora]

Hoping to get a fix out asap!

[craggy torrent]

alr

[terse shuttle]

already reported

[leaden cove]

pineapple

[echo aurora]

I'm not sure tbh, this is a question better suited for YT

[leaden cove]

i got my session terminated or what

my chats are gone

literally all

[echo aurora]

Yeah that's all going to be related to the outage we're having

[leaden cove]

omg is it gonna get fixed and we will get chats back?

[craggy torrent]

i can select a model now

the website works

[maiden fulcrum]

me too but i cannot see "think hard" button

for gpt-5

[craggy torrent]

i dont think theres a button for it

[celest folio]

i was having ai make me a config 😭

[maiden fulcrum]

some dude said there is

[maiden fulcrum]

.

[toxic egret]

Fixed :D

[echo aurora]

is it gonna get fixed
Yes for sure

will get chats back?
Unclear, but if I had to guess unlikely. Unfortunately, when chats in the past have disappeared they don't make a return.

Would want to note we fully understand how frusterating this is, and are focussed on deving features to prevent these kinds of problems.

[leaden cove]

omgg bro

[toxic egret]

U guys are fast lol

[leaden cove]

im gonna get mad if ill not get mine chats

[narrow dawn]

does GPT-5 Has a limit?

[prime mulch]

No you can't get your chats back

[dreamy sparrow]

nope

[narrow dawn]

kk

[dreamy sparrow]

i bought the chatgpt pro for the gpt-5 high

it's ass

i regret it

:)

[leaden cove]

blud i remember that happened to me an days ago like when that was an outage i just simply refreshed the page and got my chats back xd

[echo aurora]

I am sorry for this. We fully get why losing these chats is incredibly frustrating.

[dreamy sparrow]

not really

[craggy torrent]

you bought chatgpt plus?

[dreamy sparrow]

yeag

yeah

[wheat onyx]

5 mins
https://www.reddit.com/r/ChatGPT/comments/1mkae1l/gpt5_ama_with_openais_sam_altman_and_some_of_the/?sort=confidence

[craggy torrent]

why would you?

[dreamy sparrow]

i wanted to test the gpt-5 thinking:(

[leaden cove]

BROOO YESS I GOT IT its workinng

tysm

[craggy torrent]

Oh. K

[echo aurora]

Your chat history still there?

[dreamy sparrow]

and trust me when i say it can't solve a complex math question

[leaden cove]

yess

[dreamy sparrow]

Gemini 2.5 pro solved

[leaden cove]

could u like add an email registering ?

[craggy torrent]

what was the math question?

[dreamy sparrow]

lemme get it wait

[wheat onyx]

which GPT5

[dreamy sparrow]

there

[leaden cove]

to like login from other browser to ur acc with chats

[dreamy sparrow]

thinking mode

[echo aurora]

woot!

add an email registering
user login is something we're putting thought into, I'll be able to share more with the community when I can, but overall we understand why this is an important feature, esp when things like outages could lead to loss of chat history.

[neon idol]

Why chatgpt 5 is good only via API

[leaden cove]

www lmarena

[craggy torrent]

honestly i have been waiting for this

[prime mulch]

I think without email registering is good

Its help me lot without registration

[wintry tinsel]

How much memory do the chatbots have on lm arena?

[leaden cove]

nahh we need it

[echo aurora]

registeration could be optional, I'll be sure to pass this feedback onto the team

(optional for the user is what I'm getting at)

[prime mulch]

Optional registration is great option ❤️❤️❤️

[leaden cove]

i got an cloudflare error on chrome in the lmarena site so i couldnt do anything about it for a long i tryed switching to an another browser and it helped me but there was 0 chats so

[neon idol]

How can I do a registration?

[leaden cove]

we need that login as

[craggy torrent]

once it gets added

[leaden cove]

asf*

[echo aurora]

Unable to atm

[prime mulch]

He said they Will add in future

[lime coral]

The coping never end

[neon idol]

Oh sorry I didn't read correctly

[prime mulch]

Is it possible to add midjourney image model in lm arena website or in this discord channel

[echo aurora]

I can't make promises about this stuff, but I can promise it's something we're thinking about for sure

[echo aurora]

iirc they don't have an api

[leaden cove]

btw lmarena is the really best ai site, i remember i used poe thing site where was an claude 3.5 and etc it was so horrible cuz i got an limited time to chat w the model on free plan

[prime mulch]

Roger sir

[neon idol]

@echo aurora Do you have the idea to add the app of LMarena?

[wintry tinsel]

??

[leaden cove]

add the site to the main screen xd

[prime mulch]

I looking forward to see the opensource devolopers find a way to clone midjourney or anything better than that

[stray aspen]

Bro are you serious

[echo aurora]

sorry to say I'm not following

[stray aspen]

You have to put it in the prompt

[prime mulch]

Website is better we really don't need an app

[stray aspen]

Why would we need an app

The website is great

@echo aurora Was the guy from the interview the lmarena ceo

[echo aurora]

An app is a possibility, if it's something you'd like to see I'd encourage you to voice this in #1395093524516573399

[echo aurora]

from the AMA earlier?

[stray aspen]

Yes

[echo aurora]

Nope, that was Thijs, who's an intern for the summer here

Who I wish would stay here forever but is off to Stanford for the upcoming semester 😭

[leaden cove]

yo btw could u add an veo3 to video models?

[prime mulch]

Can u add image models in this channel with some low restriction

[prime mulch]

Its aldredy there

[stray aspen]

thats nice

[echo aurora]

Yeah it's there, but unable to select specific models to use (like direct and side-by-side mode)

[leaden cove]

its random?

[echo aurora]

I don't think so. We'd like to keep this channel free for general discussions.

[leaden cove]

gambling bruh

[echo aurora]

yeah

[prime mulch]

Its soo random

[stray aspen]

does this new run away model have audio

[prime mulch]

Yeah and this channel thought me a lot in few days and its first time for me to speak with someone that directly can solve issues in the website that i use

[dreamy sparrow]

oh

[wheat onyx]

rollout so bad that they're doubling limits

[stray aspen]

so thats why people were saying gemini is better

[velvet parrot]

Has anyone recovered history?

[prime mulch]

Not possible

[velvet parrot]

Really? Because I have my chats on my phone, but not on my computer.

[ocean vortex]

They are saved as your browser's temporary files I think. Not stored online (?)

[dreamy sparrow]

Gemini will forever be better

gemini 3 is 100% gonna be crazy

what about grok 4 heavy tho

lol

[ocean vortex]

heavy dork

[dreamy sparrow]

indeed

[wheat onyx]

autorouter broke yesterday
they will say what model youre using
doubling of limits
can still use 4o

[mortal coyote]

@echo aurora stuck at this

[surreal creek]

Reloading the page normally works 🙂

[echo aurora]

This is likely due to the outage

[keen beacon]

I am sure they will fix the situation

[blazing bison]

So apparently openai is lying on swe bench too

Wtf

[dreamy sparrow]

that is indeed correct..

[blazing bison]

https://x.com/deedydas/status/1953853659848016138?t=-7Ose0v15-AatPUdvtX17w&s=19

Deedy (@deedydas)

Ridiculous that OpenAI claimed 74.9% on SWE-Bench just to prove they were above Opus 4.1’s 74.5%…

By running it on 477 problems instead of the full 500.

Their system card only says 74% too.

X · Aug 8, 2025, 4:19 PM

[dreamy sparrow]

yeah im sticking to gemini 2.5 pro

🙏

[blazing bison]

Gpt 5 is a good model if you use the api

But api only

[dreamy sparrow]

ehhh

ehhhhhhhhh

Gemini is kind of smarter in ai studio...

with code execution

and search

so

yeah

Gemini still tops

[reef pawn]

I think Open AI is overvalued company

[stray aspen]

Yeah

It's great

[blazing bison]

Bro you need to stop believing in yup

[stray aspen]

Im using lmarena

[stray aspen]

Wdym

Its a great website

i dont know why people say its fake

(i dont work for yupp)

[dreamy sparrow]

it is

lol

[reef pawn]

@blazing bison your PFP is from Ranking of Kings, right?

[dreamy sparrow]

ur telling me they have claude 4.1 opus

and not grok 4 heavy

nuh uh buddy 💀

[reef pawn]

I want Deepseek and Gemini to win the AI race

[stray aspen]

grok 4 heavy doesnt even have an API

and its horribly expensive

[dreamy sparrow]

they already WON

[reef pawn]

Real

[dreamy sparrow]

gemini with code execution

can't fail

trust me

ai studio

[reef pawn]

As of now it's GPT-5

[dreamy sparrow]

watch what Gemini 3 will do

[reef pawn]

Waiting

[stray aspen]

it does fail

gpt-5 does better

[dreamy sparrow]

lmao no.

[stray aspen]

it first shots stuff

[dreamy sparrow]

Gemini aswell

[stray aspen]

on gemini i had to ask for polishing

[dreamy sparrow]

ai studio or app

[stray aspen]

ai studio

[dreamy sparrow]

not sure

i asked it complex questions

solved it

Easily

deepseek too tbh

idk why people don't talk about it

it's so good

it's the longest model that thinks im pretty sure..

took 1k seconds for a question

[stray aspen]

it reasons for too long

plus server is busy errors the whole time

plus no vision

[flint skiff]

@blazing bison gpt5 is super slow for me on the api

vscode, roo and kilo code

using high reasoning

same for u?

[blazing bison]

no

it's the fastest model of all reasonings

but i'm using reasoning medium, high is not necessary

[novel crater]

anyone gonna question the guy generating weird ass children videos in the video arena

[blazing bison]

which arena

[novel crater]

video arena

[stray aspen]

lmao

[novel crater]

#video-arena-1

[stray aspen]

that video is so bizarre

[warm fulcrum]

they have a cashout feature "u cant even use it". u can literally use any model u like for free and unlimited as long as u just give them feedback

so

its fake bro

[stray aspen]

its not fake lol

the models arent fake

[warm fulcrum]

sure man

seems like they can really afford

opus 4.1

[blazing bison]

bro they don't have money to offer that models

[stray aspen]

opus 4.1 is the more expensive models

which cost 300 points

[blazing bison]

they have scammer crypto shi*

[warm fulcrum]

is bro really saying

300 point is the same as real money

[stray aspen]

i dont know how they profit

but im sure the models are not fake

[blazing bison]

they don't bro

[warm fulcrum]

its probably just

[blazing bison]

they offer models from openai? Yes, but the gpt 5 reasoning for example is fake

you're using o4 mini

[warm fulcrum]

yea

i tried gpt-5-high against the actual gpt-5

and

it sucked ass

from their website

[stray aspen]

it worked fine for me

its great

[warm fulcrum]

cause u probably ask it basic ahh questions

[stray aspen]

no

[blazing bison]

the results you showed,o4 mini did an exactly like frontend

[stray aspen]

i was coding for roblocks

[warm fulcrum]

ok buddy

[stray aspen]

also i tried simple bench with that gpt 5 in yupp

he nailed it

[keen beacon]

I ask it basic ahh questions. Don't have any ideas to test the model

[blazing bison]

bro is lying or is marketing for them

[stray aspen]

i dont work for yupp.ai lmao

[blazing bison]

bcs a lot of ppl tryed the same on X

and the results are absolut trash

only you is saying this

[keen beacon]

Yes, I actually use them, but I like the idea of having all the AI in one, like LMArena

[blazing bison]

or you are lying or idk

[blazing bison]

the link lmao

[stray aspen]

what link

[blazing bison]

bro is marketing 100%

[warm fulcrum]

honestly

i tried yupp.ai for like

a week

[blazing bison]

why don't say just yupp, you had to put the entire link

[warm fulcrum]

but idk man it just sucked

[blazing bison]

it sucks

[stray aspen]

idk i just call it that

[warm fulcrum]

its like watered down versions of the actual model

[stray aspen]

nah

thats microsoft copilot

[warm fulcrum]

theyre naming gpt-5 nano as gpt-5-high

[stray aspen]

yupp.ai uses legit models

[blazing bison]

you can proof that?

[warm fulcrum]

confirmed by?

[stray aspen]

yes i compared

thats my proof

[warm fulcrum]

source: turst me bro

[blazing bison]

lmarena has partneship with google, openai, that's why they receive models before release for tests

[warm fulcrum]

@blazing bison does lmarena get their models from the actual company?

[blazing bison]

some of them yes

[stray aspen]

they are donated

[warm fulcrum]

i know that

[stray aspen]

they donate them the models

to lmarena

[warm fulcrum]

any model from openai is handed to them

but how are yupp ai getting their models 🤔

[stray aspen]

with the api

[blazing bison]

they don't bro

[warm fulcrum]

so they just give it to users for free

[blazing bison]

you don't imagine the cost of this

[stray aspen]

yeah but its limited

[warm fulcrum]

how?

u literally get more points for feedback

more than u had to spend

also this lol

[stray aspen]

you dont

[stray aspen]

i tried that

it doesnt work

[warm fulcrum]

yea

clearly

i said it didnt

[warm fulcrum]

yes u do bro

this guy is ragebaiting

[stray aspen]

you dont dude

sometimes you get nothing

sometimes you get 30

[warm fulcrum]

https://media.discordapp.net/attachments/947323370754691102/1024753179595903066/erm_mods.gif

[stray aspen]

and sometimes 200

[warm fulcrum]

dude they even made it easier to get points

u get like 1k every 5 prompts

[blazing bison]

bro they offer o3 pro

o3 pro

[stray aspen]

i dont

[blazing bison]

for credits

[warm fulcrum]

fakeee

[blazing bison]

do you know how much o3 pro costs?

lmao

[warm fulcrum]

i tried it

[blazing bison]

and their o3 pro answers in 30 seconds

the official one takes 10 minutes

[stray aspen]

no it takes forever

[warm fulcrum]

wait does

o3 pro

show its thinking

[blazing bison]

the summarized one

yes

[stray aspen]

no

they dont show

[blazing bison]

on api they show

[warm fulcrum]

im talking about thje actual one

not on yupp.ai

[blazing bison]

yes it show summarized ones

[warm fulcrum]

maybe every yupp.ai model is just an opensource model fine tuned

lol

[blazing bison]

prob is

[calm seal]

wth

[warm fulcrum]

what was this prompt exactly?

and also

[dreamy sparrow]

claude is kinda ass

[warm fulcrum]

100 credits for gpt 5

[blazing bison]

they put the gpt-5 name on system prompt

[warm fulcrum]

is a steal

[dreamy sparrow]

only good for coding

..

[stray aspen]

this

[warm fulcrum]

lol

ok

[blazing bison]

bro they can put any model name on system prompt

[stray aspen]

but this one is real tho

[dreamy sparrow]

and free on lmarena tho?

[blazing bison]

it don't

[warm fulcrum]

with limits

[dreamy sparrow]

wait what

[stray aspen]

actually 50

if you make your chat public

[warm fulcrum]

why would u do that

[dreamy sparrow]

wait really

[stray aspen]

to spend less points

it halves the cos

cost

[dreamy sparrow]

what if i ask how to get a gf

:)

[warm fulcrum]

?

idk bro

the model came out yesterday

i dont know everything about it

[dreamy sparrow]

gpt-5 is good for math

Api

only

bruv.

every i time i say anything

u say no lol

[stray aspen]

its great for math

[dreamy sparrow]

we breath oxygen

no lol

no lol

no lol

no lol

like

what the hell

[stray aspen]

are you trippin

[dreamy sparrow]

yes

[white hatch]

gpt is censored af

[dreamy sparrow]

what

they are stupid tho

guns.lol is stupid

[warm fulcrum]

bro

...

[dreamy sparrow]

u have no bio bruv

why are u talking

[warm fulcrum]

this is even worse

[dreamy sparrow]

what basically

name

[warm fulcrum]

its corny

[dreamy sparrow]

or

[dreamy sparrow]

the hitlocker?

[warm fulcrum]

everything in the image i sent

[dreamy sparrow]

ONE ONLY

[warm fulcrum]

its more corny than using guns.lol

[dreamy sparrow]

not really

[warm fulcrum]

brudda just keep the convo about ai

b4 pineapple come in here

[dreamy sparrow]

gpt-5 is ass

[stray aspen]

its actually a nice model

i dont know why people say its garbage

[dreamy sparrow]

yes

ofc it is

o3 pro or gpt-5 tho

[stray aspen]

gpt 5

[dreamy sparrow]

o3 pro thinks for like 20 min

i saw

[stray aspen]

way cheaper and comparable performance

and faster

[dreamy sparrow]

i don't really care about speed

i want them to take more time for more accuracy

[stray aspen]

thats not real

deepseek takes forever and does some stuff wrong

[dreamy sparrow]

what's not real

[dreamy sparrow]

coding or matg

math

[stray aspen]

everything

[keen beacon]

Qwen3?

[dreamy sparrow]

what about Gemini 2.5 pro

[stray aspen]

i dont know about qwen 3

tried it for coding and it sucks

and never use it again

[white hatch]

It's just nice

[dreamy sparrow]

it's crazy with code execution ngl

[keen beacon]

My fav open source model currently is Kimi K2

it has a pleasant personality

[dreamy sparrow]

Kimi as straight up ass for math

[keen beacon]

I dont use it for math?

[white hatch]

Haven't tried it yet btw

[stray aspen]

i think its meant for coding anyways

[dreamy sparrow]

gpt-5 has code execution?

[keen beacon]

Why do people assume I'mma do math with AI?

[dreamy sparrow]

didn't know that

[keen beacon]

I only do trivial stuff mostly

day to day stuff.

[dreamy sparrow]

wait does gpt-5 actually have

code execution

[keen beacon]

I think in canvas mode, yes

[white hatch]

Not really sure, but if you look in thinking progress you can actually see some python code

[dreamy sparrow]

oh

[white hatch]

Sometimes

[dreamy sparrow]

so it DOES have it

[white hatch]

Maybe it uses tools

[dreamy sparrow]

never used it

[keen beacon]

It was handy for pics

analyzing them

[dreamy sparrow]

o3 is better than o3 pro or what

u said o3 pro is ass

do u think gemini 3 would be better or gpt-5

if u say gpt-5 im gonna do bad things to u

Alright

HOW

[keen beacon]

why you so agitated?

[dreamy sparrow]

idk

in text and solving real life problem things

Yeah

[ocean vortex]

it isn't really gpt4o though

[dreamy sparrow]

i do:(

[ocean vortex]

OpenAI are cogsuckers for naming it like that

[keen beacon]

I have tried it out. It does work pretty well for video inputs

[stray aspen]

openAI is the least open AI company

[dreamy sparrow]

lmao

100% true

grok is more open 🙏

[stray aspen]

grok is dogwater

[ocean vortex]

Well to be fair... they did release open-source model finally

[dreamy sparrow]

dog sh

[ocean vortex]

So they are open in ways

[stray aspen]

just to justify their name lmao

20 cents

[dreamy sparrow]

i just bypass yupp

[ocean vortex]

8000 yap credits

[dreamy sparrow]

why do u pay

[stray aspen]

used to be 1 dollar

wait

nevermind

[keen beacon]

Well, the whole service seems a bit controversial atm because of its alignment problems

[stray aspen]

6000 points is more than 20 cents

[dreamy sparrow]

damn

[stray aspen]

1000 points is 20 cents

if im not wrong

but cash out is unavailable anyways

[ocean vortex]

wtf is yupp?

[dreamy sparrow]

i could buy a burger with that in Egypt

[stray aspen]

a website to use ai

[dreamy sparrow]

a website that uses 30% fake ai

[stray aspen]

it doesnt

it uses real models

[blazing bison]

it does

[ocean vortex]

ikr. But they do scam all tourists with like 20X prices lol

[keen beacon]

Why are we talking about Yupp anyway?

Is there any good thing about it?

[stray aspen]

i dont know

[stray aspen]

free o3 pro

[blazing bison]

it's not o3 pro

[dreamy sparrow]

ur telling me there's a gpt-5 high.

[blazing bison]

bro is lying

he is yupp mod or something

[dreamy sparrow]

that thinks same as gpt-5

[stray aspen]

that must be a regular gpt-5 with a system prompt

[ocean vortex]

There's actually gpt5-high-high

[stray aspen]

and they call it high

[dreamy sparrow]

i tested it with lmarena gpt-5

[ocean vortex]

high reasoning effort + high verbosity

[keen beacon]

Is it correspondent with the performance of the Pro model or is it a hoax?

[dreamy sparrow]

it's fake in yupp

thinks for 10 seconds

:)

[stray aspen]

i dont know ive never used the o3 pro from chatgpt

[keen beacon]

Welp, better be wary of its performance, lol

[stray aspen]

they also have free 4.1 opus

[keen beacon]

damn

[stray aspen]

they have a lot of models

and image models

[dreamy sparrow]

fake one

I've tested and saw on yt

lol

[stray aspen]

how do you know its fake

[ocean vortex]

it does it's just formed differently:

        "effort": "minimal"
    }``` 

this is still effort attribute of the reasoning object = reasoning effort lol

[dreamy sparrow]

saw vids

:)

[stray aspen]

what video

[dreamy sparrow]

thinks more in the vids

[stray aspen]

whats the reasoning effort of lmarena gpt-5

[dreamy sparrow]

less in yupp

hmm

[keen beacon]

If you did not see Sam Altman said they have problems with the dynamic thinking effort and whether the model should choose thinking or not.

on reddit

said they will fix it

that's why the model has seemed "dumb"

[ocean vortex]

this only applies to their chatgpt website. On API there's no model with auto thinking

[neon idol]

How is good claude opus 4.1 in Text?

[ocean vortex]

it's either thinking or not thinking

[dreamy sparrow]

bad

[neon idol]

Is good in coding or bad in everithing?

[obsidian shell]

for me its good

but good within a context

its good for coding micro tasks and sounding human

the rest...

no

[dreamy sparrow]

CRAZY IN CODING

[keen beacon]

Ah, I see

[ocean vortex]

no good

[stray aspen]

is this real or is it tripping

[ocean vortex]

set it to ultra high

[stray aspen]

im gonna set it to crushingly amazing

[obsidian shell]

bumpiness

[dreamy sparrow]

how do we set it

:)

[keen beacon]

What do you expect when 700 million people use it?

[dreamy sparrow]

where's the thing to set it

[small haven]

hows gpt5 so far

[keen beacon]

They have problems with inference

[dreamy sparrow]

in lmarena

[stray aspen]

yeah they are pretty much serving the whole world in their website

[solid brook]

Dude its high

[obsidian shell]

i like the word

[dreamy sparrow]

yo how do i set it

[solid brook]

I told it to think very hard on something. It thought for 3 or 4 minutes

[keen beacon]

There's no toggle for it on LMArena

[obsidian shell]

there is no ultra high model

[stray aspen]

so gpt-5 is hallucinating

[dreamy sparrow]

then what do i do

[stray aspen]

on lmarena

[dreamy sparrow]

just like grok 4

it's saying it's grok 1.5

🙏

[stray aspen]

the grok 4 on lmarena is stoned

[keen beacon]

You can only ask for it on #1372230675914031105 if you wish. For now, no chance.

[dreamy sparrow]

oh

ok

[solid brook]

@dreamy sparrow

[neon idol]

Sam Altaman: GPT 5 will be the most powerfull ai and will sostituate human. Goofy ahhh

[dreamy sparrow]

ooo

that works?

i didn't even know

lol

[solid brook]

Yep

[dreamy sparrow]

lemme actually try

[solid brook]

Just say"think very hard"

[keen beacon]

I see. Will have to try it out.

With my "boring" questions

[stray aspen]

elon musk said grok 5 will be crushingly good

[neon idol]

Ahahah