general

[ocean vortex]

disagree

😇

[solid brook]

guys competetion is good.

[eternal niche]

US were created by Europe lol

[solid brook]

yeah "americans" are just bunch of europeans

only came to the land few hundred years ago

[stray aspen]

why is this thing not working

[ocean vortex]

What is currently unfolding in US, Europe (or EU) has already lived through and hopefully put behind...

[echo aurora]

lets not go down this road pls

[solid brook]

I get it here is not the place for political stuff

[stray aspen]

google needs to lock in

[devout vault]

google will prob win again against gpt-5

with their new gemini 3 model

[red tangle]

lol gemini 2.5 deepthink is probably better than gpt5

[stray aspen]

how do you know that

[red tangle]

i have

it's miles better than o3 pro

i have completely switched from chatgpt to gemini usage because o3 has been such a poor showing

[devout vault]

gemini is completely free with almost no usage at all

[red tangle]

o3 pro thinks for 20 minutes and is like at max 5% better than o1 pro

[devout vault]

gemini 3 is waiting for gpt-5 to release

so they can compare

their models

[solid brook]

the deepthink is very limited

[devout vault]

gemini 3 is already done training

a month ago

[ocean vortex]

they added nice feature to google gpt5 launch time conveniently. If you google for "time 10am pt" this will convert to your local time. How nice of them 😇

[stray aspen]

i love google

[obsidian shell]

gpt5 in 20 minutes?

[solid brook]

only 2 hours and 20 minutes left

[stray aspen]

no

[warm fulcrum]

guys why it feel like openai slowed time

[red tangle]

gpt5 isn't even out lol

[solid brook]

i did but they nerfed the hell out of their pro model

[stray aspen]

yes i noticed that

[devout vault]

gemini 2.5 pro was so powerful before when it first released which shows google true strength

[red tangle]

[devout vault]

gpt 5 - pricy
gemini 3 - free

[solid brook]

good competetion is always good for us

[stray aspen]

gpt 5 50 euro per token lol

[red tangle]

openai has already lost on API and enterprise

[white hatch]

lol wtf

[red tangle]

anthropic

[devout vault]

claude is like a human when it comes to coding

[red tangle]

why would you use a 10x more expensive model

[devout vault]

gpt-5 will be more smarter than claude opus 4 for sure

but gemini 3? idk

[red tangle]

um

im pretty sure they will drop something this month

they've been hyping it

[solid brook]

claude is so expensive. someone needs to beat them at coding to challenge them

[blazing bison]

I got access to gpt 5, the one that plus users will receive

Too much hype

[red tangle]

the markets are idiots

[devout vault]

how is it?

[stray aspen]

show us

[red tangle]

lmarena no style control google is dominating

it's not even close

lol no they arent

[solid brook]

Source: Trust me bro

[devout vault]

i think animated is lying

lol

[red tangle]

theres somehow a 75% chance openai has best model by end of august on polymarket 😂 😂 😂

based on lmarena no style control

i did

$1k

[blazing bison]

Well, no one need to believe me 🤷‍♂️

[red tangle]

it's my first time betting on polymarket

[devout vault]

thats what a liar would say bro

no proof at all

[devout vault]

LMFAO

[blazing bison]

What do you want, prints?

Share screen?

[devout vault]

you cant do noone of those

[blazing bison]

??

[red tangle]

wall street is better than polymarket traders

HF analysts actually know how to do math

a lot of smart money has googl rn

like a loooot

[white hatch]

When gpt 5 was announced?

[patent aspen]

It's a small enough market that I could move the market

[ocean vortex]

lol

[solid brook]

imagine if it does not

[ocean vortex]

impossible

[devout vault]

what is st

[stray aspen]

wall street

[solid brook]

yes impossible. that is the reason i used the term "imagine"

[patent aspen]

I wouldn't read that as "the market thinks it's a 94% chance"

[red tangle]

all of the good HFs are subscribed to semianalysis

[ocean vortex]

I think it's gonna be made avail immediately. Too big of a release not to. At the very least to their Pro subs same day. But likely more

[red tangle]

and have really smart analysts constantly looking at compute & energy that google vs openai are building

whereas polymarket is people who invest based on twitter vibes

[solid brook]

with this much hype they have to release a version to free users

[patent aspen]

Semi analysis is good if you want invest in the supply chain of the relevant companies

[warm fulcrum]

i believe u

[devout vault]

@blazing bison is saying a lot of baloney

[warm fulcrum]

hoW?

[blazing bison]

Proof?

[warm fulcrum]

he would never do such thing

[devout vault]

show a screenshot

rn

[blazing bison]

Of copilot?

🤓

[solid brook]

this guy is such a troll

[blazing bison]

Im not the only one with access btw

[stray aspen]

theres no gpt 5 on copilot yet

[devout vault]

ye

[warm fulcrum]

do u have the smart mode thing on copilot?

[stray aspen]

i dont

[blazing bison]

You need to change things in frontend to access

[blazing bison]

Yes

[warm fulcrum]

ye

i saw some other people get it

was it just random or did u have to do something?

[blazing bison]

Random

[eternal niche]

guys i have gpt6

[devout vault]

i have gpt-6 on copilot rn

[warm fulcrum]

how does it perform

does it perform like summit or zenith

[blazing bison]

Idk too much limited

[solid brook]

bro you're far behind I have gpt 7

[devout vault]

ye

do that

[blazing bison]

I can't send images

For example

[stray aspen]

guys ive just been granted access to gpt 8.5 pro high max reasoning 1 billion context

[devout vault]

and why not LOOL

[blazing bison]

Idk

[devout vault]

u can send photos here

[warm fulcrum]

wait wdym

too much limited?

rate limit?

[blazing bison]

They added a limit of output or something

Maybe it's not really gpt 5

[ocean vortex]

he sent too many naughty ones

[warm fulcrum]

it is

but finetuned

[devout vault]

ewwww

[wicked ingot]

finetuned?

[warm fulcrum]

ye

they did the same for gpt 4

[devout vault]

i like these ngl

[warm fulcrum]

price wont be likeable tho

[solid brook]

Guys what would you think will be the free tier model with no limits?

[torn bison]

a lot of insiders has bought openai

given that summit and zenith have already been tested in the arena, they have enough confidence to do so

[warm fulcrum]

lmarena so greedy

they didnt keep zenith

[eternal niche]

lol

[whole wagon]

why didnt lm arena give us the sota model for free smh

[solid brook]

2 hours left

[whole wagon]

[warm fulcrum]

exactly

so greedy

[red tangle]

no one is deciding to buy or sell openai secondaries based on summit and zenith

guarantee 99% of polymarket traders can't even buy secondaries

it's very few individuals

it's mostly funds that are buying up secondaries

[stray aspen]

i think horizon beta has given me the best plug and play main menu for roblocks out of all the LLMs ive tried

[rapid merlin]

they having a livestream in about 2 or so hours right

[stray aspen]

yes

[quartz light]

`how 2 get

[stray aspen]

your best bet is using horizon beta

[quartz light]

whats zenith

[warm fulcrum]

an ai model that was in battle mode (best gpt 5 version)

they said it was really good

but i never got the chance to try it

[hollow imp]

Actual chatgpt plus and pro sora?

[molten cipher]

Hey @echo aurora , I was wondering if I could talk with you through DMS or a ticket in this server?

[patent aspen]

I think they know but also don't have models that deal well with it. No AI company is profitable and wall street models generally are focused on 6-18 months in the future at best

[echo aurora]

Yeah, my DMs are open or you can DM @oak python

[quartz light]

bruh

[wintry tinsel]

China is in actual flames now, just because they open source good models doesn’t mean the country is holding itself together lol

[molten cipher]

ModMail isn't properly setup

[quartz light]

why'd they snatch it

[warm fulcrum]

idk

greedy mfs

[warm fulcrum]

why would pineapple take away zenith from us??!?

[echo aurora]

Hmm okay thanks I’ll look into, very odd. You can DM me

[molten cipher]

Sent a DM.

[raven helm]

GPT-5 benchmarks just leaked (It might be fake; take it with a grain of salt.)

[rapid merlin]

where did you grab this from?

[raven helm]

https://x.com/iruletheworldmo/status/1944375085478904118

🍓🍓🍓 (@iruletheworldmo)

excited for gpt5 july 31

i leaked grok’s bench scores so. this feels fair.

hot ai summer.

we’re so back.

lfg.

wild.

it’s a good model sir.

X · Jul 13, 2025, 12:35 PM

But might be fake

take it with a grain of salt

[rapid merlin]

most likely yeah

[zinc ore]

It's strawberry

He probably made those this morning

[rapid merlin]

excited for gpt5 july 31

strawberry?

[zinc ore]

Okay I mean on the 13th*

[rapid merlin]

wasnt strawberry o1

[zinc ore]

Look at the strawberries in his username

[raven helm]

He also posted this

[whole wagon]

bruh

[stray aspen]

fake

[raven helm]

But again; take it with a grain of salt

[whole wagon]

he could have at least attempted to make it look realistic

[rapid merlin]

yeah no

[raven helm]

What makes it look unrealistic? The insane scores?

[whole wagon]

you obviously do not get a base model performing like that on arc agi 2

[raven helm]

yea

From brockman

https://x.com/i/status/1953456219433455898

OpenAI (@OpenAI)

Dropping soon.

X · Aug 7, 2025, 2:00 PM

[stray aspen]

gpt 4.5

no way

[raven helm]

https://x.com/chetaslua/status/1953420691967016989

Chetaslua (@chetaslua)

🚨 GPT-5 ONE SHOTTED 🚨

SNAKE IN HEXAGON GAME , i gave it difficulty of hexagon and multiple snakes and it handle gracefully one shotted .

If anyone want to test GPT-5 send your prompts

X · Aug 7, 2025, 11:39 AM

[rapid merlin]

gpt 4-p

4 pixels

[astral jetty]

Very curious to see if the creative writing is less repetitive than even something like Gemini

[raven helm]

Yea, i think that will be getting an upgrade

(Hopefully there'll be less/no em-dashes)

[stray aspen]

horizon beta is so good at lua

[solid brook]

WTF

i feel fake

[raven helm]

Yea, no way a base model scores that high

or maybe it does

idk

[solid brook]

that is just way too big

way too big

[astral jetty]

I can kinda believe it, but I don’t think the gap will be that big between Gemini 2.5 and base gpt 5

[solid brook]

if this is true

gpt 5 is AGI?

[stray aspen]

no

[pure anvil]

agi is when spinny hexagon+snake

[solid brook]

i guess gpt 5.5 then?

I feel we are so close

[stray aspen]

we are close to AGI

[eternal niche]

agi is fake

[torn mantle]

not this guy again

stop reposting that idiot

[raven helm]

Ok, sure. Sorry.

[torn mantle]

is it you?

strawberry guy = you?

😡

[raven helm]

No No,

[torn mantle]

ok good

[raven helm]

I'm just saying i'll stop posting it here.

[torn mantle]

thanks

[raven helm]

No problem!

[eternal niche]

it is

[jade egret]

0_0

btw gpt-5 today right

YO

[stray aspen]

its out

47 minutes

for launch

[obsidian shell]

i dont think there will be a huge difference

probably 5-9%

they are just capitalizing on the hype they have been building

[storm needle]

https://github.com/openai/gpt-5-coding-examples/blob/main/apps/webcam-filter-playground/index.html

[keen beacon]

Ahhhhhhhhhh

[jade egret]

better be much better at coding than 4.1 opus (:

[quartz light]

https://websim.com/@rat/gpt-5-countdown-party-2

38 minutes

[coarse glade]

Guys quick question how do we do text to video on LMArena.ai

[echo aurora]

More info can be found in #1397655624103493813 , but the TLDR is use /video in #video-arena-1 #video-arena-2 #video-arena-3

[coarse glade]

So I can’t do it on the website

[echo aurora]

Video Arena is currently only available through our Discord

[torn mantle]

https://github.com/openai/gpt-5-coding-examples/tree/main

some cool demos

[torn mantle]

oops

[blazing bison]

Gpt 5 on copilot feels like 4o v2

[stray aspen]

30 minutes

[stray aspen]

show us

[blazing bison]

I hope it's not the same

[astral jetty]

Copilot is also kind of mid to begin with ngl

[stray aspen]

it is

[echo aurora]

Reminder we have our Staff AMA tomorrow with the dev behind our Video Arena bot, if you have any specific questions be sure to add them here

[quartz light]

please vote :)

[echo aurora]

https://discord.com/events/1340554757349179412/1400149736027328623

[quartz light]

👀

12 minutes!

vibe coded, lol
https://websim.com/@rat/gpt-5-countdown-party-2

[astral jetty]

256k sounds right, I don’t want to get my hopes too high

[blazing bison]

From my first impressions, it's not good, but I had good results with zenith

[spring rune]

I have this

[blazing bison]

https://x.com/OpenAI/status/1953498900230250850?t=JiAJn7rnprqOOM4ou91JqQ&s=19

OpenAI (@OpenAI)

wen GPT-5? In 10 minutes.

https://t.co/EOvjGJrHbj

X · Aug 7, 2025, 4:50 PM

Lesgoo

Hype or sota

Answer in 10 minutes

[wintry hamlet]

how do I use the bot in direct messages

[echo aurora]

You're unable to, it only works in #video-arena-1 #video-arena-2 #video-arena-3

[wintry hamlet]

oh ok. It would be cool if you could

[spring rune]

Hey pineapple when gpt-5 added in lmarena?

[echo aurora]

Be sure to share feedback about the bot in #bot-feedback

[proven raft]

Is gpt 5 released?

[wintry hamlet]

Ok

[astral jetty]

Is the demo in 5 mins or is the full release in 5 mins

[storm needle]

https://platform.openai.com/docs/models/gpt-5

[stray aspen]

gpt 5 is out

400k context

[proven raft]

Ooo

[stray aspen]

@echo auroraadd gpt 5 to the arena

[astral jetty]

What is that compared to Gemini

[astral jetty]

Give him time

[rapid merlin]

1 minute yall

[void elm]

the stream is in 10 minutes but when will it be released actually

[proven raft]

Yea it's cheaper than gpt 4.5

[clever estuary]

400k context

[void elm]

how much was o4

4o*

[proven raft]

200

[clever estuary]

4o
2.5 dollar input
10 output

[proven raft]

I think it was 200k for 4o

Most likely

o3 deep research is still the smartest one, tho with huge latency

[rapid merlin]

even a free tier? goddamn

[tame horizon]

What is the name of this site?

[whole wagon]

what the hell is this graph

makes no sense lmao

[brittle tiger]

still don't have audio or video input which is annoying

[raven oracle]

yes it does, bottom is no think top is with think

[whole wagon]

o3 is at 69.1

like the scale of the graph is impossible

lmao

[void elm]

5 increase is so disappointing

[keen beacon]

Very happy to hear about it

[tame horizon]

@clever estuary What is the name of this site?

[rapid merlin]

https://tenor.com/view/better-call-saul-faint-dead-nope-bye-gif-12483755

[clever estuary]

openAI api site

[rapid merlin]

graph made by chatgpt

[willow grail]

byebye anthropic

[tame horizon]

Thankyou

[willow grail]

byebye opus 4.x

BYEBYE

say hello to gpt5

[barren prairie]

Let s test and see if it is great or as always just a hype

[willow grail]

were.

case closed.

you are fanboying billy

[brittle tiger]

The disparity between GPT-5 Thinking score (incredible) and no-think (awful) is pretty crazy

[willow grail]

wait thats a crow?

[whole wagon]

how many thinking tokens is it gonna use kek

[willow grail]

omg we both likes crows

[rapid merlin]

yeah idk what to think about that

barely better than 4o

[willow grail]

is it not a crow?

[void elm]

not even a 1% increase over 4o is disgusting

[brittle tiger]

Thinking score is amazing and I don't know why you would use it without but definitely interesting

[void elm]

thats pathetic

[keen beacon]

Daaaaamn

Daamn

[void elm]

playground? what site

[stray aspen]

its out on lmarena

gpt 5

[hardy pecan]

HOLY

[stray aspen]

that was quick

[devout vault]

damn

[astral jetty]

I mean I’m pretty sure thinking is automatic now

[stray aspen]

is the gpt 5 in arena think

[blazing bison]

Lmao

[patent aspen]

GPT-5 seems to be weaker or comparable to Deep Think on all benchmarks without tools

[blazing bison]

But it answers faster

Its important

[primal orbit]

GPT 5 live direct arena!

[stray aspen]

gpt 5 crushed the webdev by far

[warm fulcrum]

yeehaw

[blazing bison]

And you can send more than 5 prompts per day

Also important

[obsidian cargo]

summit was GPT-5? Then what was zenith?

[warm fulcrum]

beast

[hoary plaza]

Ig it's time google will release something 😂

[spring rune]

FINALLY YAY!!!

[misty star]

@echo aurora I love you

[patent aspen]

Only a 21-point ELO lead over 2.5 Pro. We're good

[warm fulcrum]

its time for everyone to release something

[spring rune]

Love you guys

[stray aspen]

crazy

[hoary plaza]

Since they want #1

[exotic tartan]

WOW

[whole wagon]

LMAO all the bettors lose with openai

it didnt win without style control

[prime mulch]

Hell yeah

[primal orbit]

lmarena first with gpt5 on the web, gratz

[whole wagon]

[clever estuary]

they are deprecating everything???

[meager harbor]

what a joke, AI pioneer not even to be able to make a proper graph, embarassing

[prime mulch]

@echo aurora thanks a lot

[stray aspen]

cooked

[patent aspen]

I'm feeling pretty good ngl

[stray aspen]

the hype was a marketing stunt

[spring rune]

Hey brian! Looks like gpt 5 is indeed becoming underrated right?

[elder rapids]

dawg they're deprecating all the previous models

[blazing bison]

Much marketing little results

Hahahahahahha

[clever estuary]

crazy
that's actually crazy

[keen ferry]

gpt 5 worse than Gemini 2.5 pro?

[blazing bison]

Yes

[soft river]

I have a question, as soon as they started the live stream, they already published GPT 5 on the official website?

[blazing bison]

No

[ember sentinel]

guys, when? 🔥

[spring rune]

Well animated you know gpt 5 is way better than gemini 2.5 pro right?

[barren prairie]

Now Google can have a good sleep 😆

[hardy pecan]

no ones interested about non style control

[ornate agate]

I think this is the "google won its over" moment.

[blazing bison]

Im gonna be sure of it after testing

[primal orbit]

https://i.snipboard.io/mT7P1H.jpg

[sleek stump]

what video model this AI is using and how many videos i can generate in a day?

[indigo hazel]

i'd like to see the hallucination rate of gemini before saying this, i dont know

[meager harbor]

I predicted a 50 elo jump max above o3, I was right, AI is hitting a plateau

[fleet ocean]

8 videos

[sleek stump]

daily?

[astral jetty]

I’m gonna wait until Gemini 3 to fully say that

[stray aspen]

cant wait for gemini 3 to destroy OAI

[fleet ocean]

Yes

[sleek stump]

and model video model is it?

[primal orbit]

Let's wait for gpt5 thinking on arena

[fleet ocean]

You can't select a model. It will independently select.

[stray aspen]

horizon beta chat was disabled

[split kayak]

ok

[sleek stump]

ohh ,thanks for the info bro

[fleet ocean]

No problem

[astral jetty]

Isn’t thinking incorporated onto the main model itself

[split kayak]

ok

[barren prairie]

With a flash 😆

[spring rune]

Uh?

[civic flame]

@echo aurora can you shed some light on what zenith was? there wasn't jsut summit

[leaden meteor]

They can't. It's against their policy.

[storm needle]

gpt 5 pro

[stray aspen]

so summit was gpt 5

[cyan zodiac]

does anyone know what makes gpt 5 better than opus 4 at coding?

[keen talon]

what is the limit for video arena?

[primal orbit]

Dunno, there have to be different versions still.

[stray aspen]

eight

[indigo hazel]

i cant wait to see gpt 5 failing at your tests lmao, send images pls

[hoary plaza]

I remember seeing opus 4.1 in direct chat

Was it removed?

[primal orbit]

yes

[torn mantle]

As I've guessed summit = gpt5

It wasnt that obvious

[stray aspen]

what was zenith

[brittle tiger]

system card
https://cdn.openai.com/pdf/8124a3ce-ab78-4f06-96eb-49ea29ffb52f/gpt5-system-card-aug7.pdf

[stray aspen]

is the gpt 5 in lmarena reasoning?

[tidal schooner]

what do you think zenith is?

[stray aspen]

im guessing it is because it took forever to answer my prompt

[proven raft]

Gpt 5 has got 50 percent in hfe

[tidal schooner]

big question atm

[proven raft]

That's crazy

[fleet lintel]

Am I reading the evals incorrectly? GPT-5 looks underwhelming to me.. What am I missing?

[dreamy sparrow]

is every model in lmarena even real

[dreamy sparrow]

is it actually the real model

[stray aspen]

yes not sure for grok 4 tho

[meager harbor]

it's an improvement, just not that significant.

[rapid merlin]

it looks like it, it took like 2 minutes to answer an image examination question

[fleet lintel]

Things are slowing down but GPT-5 was supposed to be like multiple levels better ..

[proven raft]

Traffic

Many ppl are using it

[dreamy sparrow]

ye it's kinda real

most apps use it

[echo aurora]

yes

[dreamy sparrow]

ok

[stray aspen]

is it gpt 5 think or no think

[leaden meteor]

Why didn't lmarena add got 5 thinking on leaderboard yet? If it is zenith, it should have had enough votes like summit/gpt5main ?

[dreamy sparrow]

thino

think

[rapid merlin]

that's possible too, and i think the image you sent is fake

[keen beacon]

People are already complaining about gpt 5, lol

Just enjoy it

[eternal niche]

so gpt5 is crap

[dreamy sparrow]

gemini 2.5 thinks lesser than it

tho idk

maybe Gemini IS better

[stray aspen]

gemini 3 will destroy it

[red sluice]

Barely better than Gemini

[keen beacon]

It's too positive/sycophantic like 4o

Don't like it

[brittle tiger]

I don't think all of GPT-5 is thinking. It might do some to determine which to route to but they wouldn't differentiate between main and thinking if both were thinking.

[fleet lintel]

i think next version of gemini 2.5 is going to be better than gpt-5

[dusky aurora]

I can't wait

[fleet lintel]

I thought OAI was 3-4 months ahead of Google. I think it's behind now

[rapid merlin]

i wonder how strong the thinking one is

[primal orbit]

Gemini 2.5 was released in March. Google has to have much better version internally now.

[barren prairie]

But is was improved at coding over the time

[leaden meteor]

They have benchmarks for main and thinking. If they are same, why different benchmarks?

[fleet lintel]

polymarket sudden huge moved in favor of gemini models for Aug

[primal orbit]

it may have improved in coding, but it became sycophantic like 4o

[meager harbor]

the gemini gemini guy kilpatrick say this week was an exciting week so maybe gemini 3 experimental tomorrow ?

[patent bane]

oh no it's more censored

[fleet lintel]

gemini 3 might drop in Aug-end but most likely in Sept or Oct. I intially thought it would be Dec but they are accelerating the pace as per some sources

[barren prairie]

Google is now focusing at releasi ng some feautures on gemini app

[stray aspen]

LMAO

[dusky aurora]

Gemini is much less censred than the opposition

[leaden meteor]

Didn't we already get genie 3?

[meager harbor]

he said exciting week, not exciting day

[jade egret]

is gpt-5 better than gemini

[stray aspen]

slightly

[barren prairie]

But the story book too

[small haven]

Wait thats a steal

[fleet lintel]

I think wolfstride model is better than gpt-5 ... this is very dissappointing 🙁

[small haven]

Gemini 3 aint coming in aug tho

[meager harbor]

gemini 3 100%

[void elm]

what is this creature talking

[feral lichen]

Best ai for

Rbx studio

?

[void elm]

shut up

[stray aspen]

gpt 5

[void elm]

nah

best model is https://www.codecademy.com/learn/learn-lua

[keen beacon]

Yes. They are surely updating the gpt image model

[civic flame]

it literally wasn't though

[rapid merlin]

fr

[feral lichen]

I use them to help me, not just copy.

[civic flame]

it thought for less time than summit on 90% of tasks

lmao

it also had less juice

that was NOT pro

[eternal niche]

@deep adder hi

[civic flame]

they just picked the checkpoint that won in elo even though it was less performant

openai never help themselves do they

[whole wagon]

Oh dear....

[dreamy sparrow]

is the gpt in lmarena the pro thinking and not normal one

[jade egret]

is gpt-5 good? (better than claude 4.1 opus?)

[civic flame]

in logan we trust

[meager harbor]

so gpt 5 is 200 elo better than gpt 4 0314 (the og one)

[fleet lintel]

who is the only voter in favor of GPT-5 ? :). who is the crazy person? Reveal yourself

[red sluice]

Damn so GPT-5 was on lmarena since the 27th of July 🤯

[rapid merlin]

LMAOOOOOOOOOO

[eternal niche]

well...

[jade egret]

what does this mean, people dont like gpt-5?

[stray aspen]

craig eating his words again

[whole wagon]

underperformed expectations

[red sluice]

...

[stray aspen]

craig stop licking openai's boots

[jade egret]

lol

[native flame]

Hii, this gpt-5 has the thinking mode activated by default? Like the o3??

[dreamy sparrow]

Gemini 3 100%

lmao

[jade egret]

fr

[mental briar]

no

[dreamy sparrow]

it's google mate

[fleet lintel]

it's @deep adder 🙂 not surprised

[wheat onyx]

Absolutely huge

[dreamy sparrow]

it has thinking bruv

[mental briar]

but no by default

[dreamy sparrow]

well

yeah i guess?

[brittle tiger]

https://x.com/AiBattle_/status/1953508582927778188?t=xXydAMuMP2I9boJd_8QmqA&s=19

AiBattle (@AiBattle_)

GPT-5 (High) scores 9.9% on ARC-AGI-2, Grok 4 (Thinking) scored 16.0%

X · Aug 7, 2025, 5:28 PM

[keen beacon]

yes

[meager harbor]

2 years and half to get 200 elo boost, we'll see where we at in 2 years and a half but I predict max 100-120 elo if we keep the same pace but yeah sometimes people dont realise when a model hallucinates especially on hard task so an elo plateau is to be expected

[lavish orchid]

damn the context windows 😭 https://x.com/ahmetbuilds/status/1953511311175737370

Ahmet Dedeler (@ahmetbuilds)

GPT-5 context windows in ChatGPT:

8k for free users, 32k for Plus, 128k for Pro

same numbers as it was before with gpt-4o, bummer they didn't increase it since the model now has 400k context window

X · Aug 7, 2025, 5:39 PM

[keen beacon]

Can we people stop complaining already? Take a deep breath

[fleet lintel]

OMG!

[civic flame]

jesus

[steady vale]

grok 4 honestly kinda sucks

for any type of question that isn't just benchmark maxxing types

[fleet lintel]

If I remove style control... 2.5 > Gpt-5.. WTF

[civic flame]

1 point better

💀

[stray aspen]

8k are you serious

[verbal nimbus]

It's crazy and kinda scary how much Gemini 2.5 Pro hallucinates.

[rapid merlin]

32K? 🥀 🥀

yeah, go google go go go

[verbal nimbus]

Yeah, although if you look at the system prompt, I think the model doesn't enforce a limit most of the time on auto.

[void elm]

gpt 5 is a disappointment

gemini 3 last hope

[fleet lintel]

This release is breaking my heart :(. I had soo much hopes

[fleet lintel]

What if Google also fails 🙁

[astral jetty]

Yeah, I didn’t like grok 4 that much when I tried it

[stray aspen]

if its gonna have 8k windom for free users ill just use google ai studio

that sucks

[verbal nimbus]

For example, I told it to use Google Search. It told me it did. But you can see that there are no citations returned, which mean it didn't.

[void elm]

idk ask yourself why ai progress is slowing down all of a sudden

so much hype but so little progress

[rapid merlin]

I really hope it doesn't get dementia past the said limit, would be pretty damn hard to use for anything that isnt a simple task

[astral jetty]

We’ve probably hit a plateau

[whole wagon]

Google has incredible stuff upcoming there is no worries trust me. I am a bit shocked by this gpt5 release

i thought they would have cooked for sure

Tbh i tried gpt5 and the vibes are great

[stray aspen]

i hope google doesnt mess up

[verbal nimbus]

Then I told it to explicitly include the link from each website it found, and a quote from each. And it just started making non-existent links and quotes up. Like what the heck.

[empty stump]

Hi how good is gpt 5

[eternal niche]

craig still coping

[civic flame]

michelle please stop i can't see you through the tears in my eyes

[stray aspen]

its mid

[whole wagon]

like i tried actual help for real world coding and it was pretty good

maybe dont judge pure off benchmarks just yet

[verbal nimbus]

It's completely lying to the user about having used the search tool during thinking.

[empty stump]

4.5 type of thing?

[keen beacon]

A lot less hallucinations which is a big thing

[keen beacon]

Needs more votes

[native flame]

Well then how can I make the GPT-5 on the lmarena can think?? Like the reasoning models?

[feral lichen]

You know, the human brain can reach further than an ai

[barren prairie]

To be more down

[keen beacon]

?

[verbal nimbus]

Is it not thinking?

[keen beacon]

You guys are so negative

[indigo hazel]

guys i honestly think that the thing of hallucinations is still really good even if it's not the best model in every task

[void elm]

AI is NOT replacing jobs 💀

[void elm]

1 year wait for this trash btw

[iron meadow]

So

[blazing bison]

1 year?

2 years

[void elm]

more yea

[fleet lintel]

that;s one good thing about this rlease 😄 😄

[blazing bison]

2 years for a mid model

[iron meadow]

How does gpt 5 compare to opus 4

[rapid merlin]

saved to my gifs

[iron meadow]

Does it have more sophistication?

[whole wagon]

its not trash cmon lol. it is still SOTA its not just the jump expected

[verbal nimbus]

This is Gemini 2.5 Pro btw, not GPT-5.

[tall summit]

gpt 5 seems... better at translation than any other from openai

[rapid merlin]

censorship update

[blazing bison]

people will stick with claude, believe me

[void elm]

this jump is the same amount of improvement from claude 4.0 to 4.1

[leaden palm]

what is my timeline 😭

[zealous panther]

What was zenith

[blazing bison]

Dario did nothing and won

🤑

[native flame]

I was told it has not the thinking mode by default, Soo I don't know how to activate it :'v cuz there are no buttons

[stray aspen]

we dont nkow

[rapid merlin]

LMAOOOOOOO

[void elm]

HAHA

alr bro

[fleet lintel]

agreed.. still very disappointing .

[whole wagon]

not true

[empty stump]

What is airline

[zealous panther]

It said it was trained onopenai data so

[fleet lintel]

I fell for OAI hype again... this is like 5th time I fell for it

[barren prairie]

Maybe it was gpt 4o update but they wanted to call ot gpt5

[iron meadow]

lawsuit incoming?

[blazing bison]

gpt 5 is really a 4o v2

[zealous panther]

Literally any AI hype ever…except google thouh

[keen beacon]

Boys, "15.9% for Grok 4 vs 9.9% for GPT-5." it sucks. sad face emoji

arc agi

[verbal nimbus]

Is GPT-5 a router model?

[whole wagon]

sheesh

[rapid merlin]

gpt-4oo

[whole wagon]

that aint great ngl

[wintry tinsel]

Is there a gpt 5 heavy or a gpt 5 thinking yet?

[keen beacon]

This is the worst "upgrade" ever!

[fleet lintel]

50 ELO improvement over 2.5 gemini.. that was my mid-level expectation.. i was hoping for 65+

[jade egret]

so gpt-5 suck?

[wheat onyx]

Why are people thinking this is bad? This looks fantastic to me

[glad perch]

Maybe it's a hybrid model like clause 3.7 and qwen3

[red sluice]

It doesn't suck, it's just overly underwhelming and isn't the revolution everyone wished for, far from it

[whole wagon]

they hyped it as agi and beyond

[keen beacon]

Simple-bench needed right now! can't wait for those results

[jade egret]

oh

so they hype it up way too much lol

[whole wagon]

"a team of phd experts on demand" yeah no it is not that lol

[brittle tiger]

Solely expecations. o3 was bigger jump. people rightly or wrongly expected a big jump with 5 series

[wheat onyx]

looks like it kills on all tests, and 1/5 the hallucination

[keen beacon]

Thats what sam hype man does!

[jade egret]

guys please tell me is it even betetr than opus 4.1?

[weary flint]

Hello, how do I make videos in 9:16 size?

[fleet lintel]

it is not fantastic.. remove style control and 2.5 pro is better than gpt-5. that's how dissappointing the model is

[deft vigil]

Wow finally Sam hit the wall

[steady vale]

this long context performance is really good

actually game changing tbh

[whole wagon]

i really like gpt5 ngl. i actually used it

[zealous panther]

Jse it bruh

[meager harbor]

it's same improvement from o3 to gpt 5 than from gemini 2.5 experimental 03-25 to gemini experimental 05-06

[keen beacon]

Google DID "out accelerate" Sam

[weary flint]

Hello, how do I make videos in 9:16 size?

[keen beacon]

lol

[whole wagon]

the benchmarks arent capturing smth it feels

[deft vigil]

Any coding benchmark guys

[void elm]

gemini 3.0 deep thinking is gonna leave gpt 5 in the dust

[wheat onyx]

looks like GPT5 is much better at writing, coding, and overall knowledge, while having 1/5 the hallucinations. Pretty huge imo

[zealous panther]

Its topping webdev so idk why pissed

[fleet lintel]

Sama is saying the same on twitter...

[echo aurora]

You're unable to set the size of the output

[wintry tinsel]

GPT 5 is confirmation this crsp has hit a wall and we’ve been being grifted into thinking it hasn’t for over a year now

[zealous panther]

Disappointed again

[indigo hazel]

that's why i think is really good

[keen beacon]

Ran a writing task (albeit meme one) and it fails to Gemini 2.5

[stray aspen]

its good at lua

[feral lichen]

Hi

[verbal nimbus]

I don't think ARC-AGI is reliable for models released after the benchmark. Just look at how o3 (High) scores 68.8% on ARC-AGI-1 vs. Opus 4's 35.7%, but for ARC-AGI-2 they score about the same (Opus scores higher now). The current models will probably do badly on ARC-AGI-3.

[keen beacon]

Slop maxed model

No way

[wintry tinsel]

AI has hit a wall guys

[wintry tinsel]

They don’t know how to scale or improve it anymore it’s all investor hype

[keen beacon]

This is what Sam commented on the Chart crime "wow a mega chart screwup from us earlier--wen GPT-6?! correct on the blog though.
"

[fleet lintel]

@deep adder is agian the only voter in favor of gpt-5... are you OAI employee?? reveal to us

[feral lichen]

Ai has limits, the brain does not

[keen beacon]

"GPT-5 is here - and it’s #1 across the board.

🥇#1 in Text, WebDev, and Vision Arena
🥇#1 in Hard Prompts, Coding, Math, Creativity, Long Queries, and more

Tested under the codename “summit”, GPT-5 now holds the highest Arena score to date." THEN WHAT WAS ZENITH???

context: zenith was better

[stray aspen]

its gpt 5.5

[wheat onyx]

GPT5Pro?

[blazing bison]

reveal zenith

[feral lichen]

Does anyone know anything about monitors?

[whole wagon]

summit > zenith

[blazing bison]

no

[red sluice]

Yup if zenith is google damn son we're gonna have fun

[blazing bison]

no one agrees that summit >zenith bro

NO ONE

[keen beacon]

GUYS! GPT-6 COMFIRMED! (gone wrong!)

[whole wagon]

bruh

no way

[white hatch]

we'll see how gpt 5 will fix the bug in my project

[wheat onyx]

This crushes Opus on SWE Bench. 75% vs 67.6%

[fleet lintel]

it;s OAI model for sure

[red sluice]

oh

[primal orbit]

openai is not the same without Ilya.

[keen beacon]

We need ilya back tbh

[verbal nimbus]

Would be nice if there was a graph with Claude, and the price on the x-axis.

[blazing bison]

imagine if zenith is gpt 5 -mini

fun

[keen beacon]

His balding head made the company bold

[wheat onyx]

we'll see these from 3rd party

[void elm]

openai is over

not even gpt 6 will save them

actual bs

[stray aspen]

at this point deepseek r2 will cook them

[verbal nimbus]

Opus 4.1 scores 74.1% without thinking btw

[keen beacon]

Bro this is pure trash

coding specialized model

[fleet lintel]

nah.. OAI will remain top company for atleast couple of years... but direction is not great

[keen beacon]

Opus 4.2 will cook this

[narrow dawn]

lmarena doesn't works

[echo aurora]

how so?

[wheat onyx]

good to know, it's not on the leaderboard I saw

[narrow dawn]

i need to select the model and there is no model

._.

[deft vigil]

Lmarena is dxomark 2.0

[unborn lantern]

guys, how can i use gpt 4.5 in lmarena?

[whole wagon]

Admittedly I am extremely confused how they are expecting to reach AGI

[stray aspen]

is lmarena down

[fleet lintel]

lol https://www.reddit.com/r/singularity/comments/1mk70xx/summary_of_the_livestream_for_those_that_couldnt/

[whole wagon]

This does not seem the path to do so

[void elm]

google will be the first to reach agi

[verbal nimbus]

Opus 4.1 was just released yesterday I think. Only a small improvement on SWE-bench though.

[meager harbor]

SCAM HYPEMAN

[pulsar rain]

gpt-5 tend to go straight to the point compare to gemini 2.5 pro

[unborn lantern]

Yes. It's down

[void elm]

google has infinite money

everyone shitted on bard then they cooked

[keen beacon]

They think they can declare AGI by the end of the year (reports mention this in relation to the Microsoft deal) HOW???

[wheat onyx]

ah so it's basically on par for this test

[narrow dawn]

how long?

[echo aurora]

uh oh, I'm seeing the same, escalating now.

[opal juniper]

I have same problem

[unborn lantern]

Mybe 3 mints

[tired herald]

Yeah model selector be having some problems

[narrow dawn]

bruh :/

[opal juniper]

However the battle mode works

[wheat onyx]

Anthropic should have no problem for staying ahead in coding then, they said big improvements in coming weeks

[feral lichen]

Max tokens for gpt 5?

[narrow dawn]

i was just making my homework with it

[unborn lantern]

guys, how can i use gpt 4.5 in lmarena?

[keen beacon]

Guys! have we hit a wall??? first claude opus 4.1 scores 3 percent more and now this?

[stray aspen]

lmarena model selection is not working

[unborn lantern]

Lmao

[sleek crow]

finally a model that can generate a Minecraft clone

[verbal nimbus]

The subset of problems and the framework is a bit different, I think. And one has thinking whereas the other doesn't. SWE-bench bash-only mode would be more fair: https://www.swebench.com/. They'll probably show up soon.

[fleet lintel]

100%

[feral lichen]

Max tokens for gpt 5?

[white hatch]

looool

[stray aspen]

a chinese model also did it

[narrow dawn]

i missed one thing

[opal juniper]

What model is that

[fleet lintel]

i think OAI is definitely hitting the wall... let's see if google is in the same boat.. gemini 3 will tell us

[keen beacon]

"Grok 5 will be out before the end of this year and it will be crushingly good
" ELON ON TWITTER!!!

[sleek crow]

gpt-5

[unborn lantern]

Which model?

[verbal nimbus]

GPT-5?

[pulsar rain]

that's quiet impressive. try asking it all the problem with chunk generating, glass, water, vvv to see how it would solve

[narrow dawn]

something with gpt

[sleek crow]

is avaible on lmarena

[torn mantle]

is lmarena down?

cant select models

[keen beacon]

So the entire hypetrain depends on Gemini 3 since both Anthropic and OpenAI hit a wall

[feral lichen]

Max tokens for gpt 5?

[verbal nimbus]

Or try asking it to solve old-fashioned JS script-order loading issues. Claude had a seizure, lol.

[exotic tartan]

API said 128K

[keen beacon]

WE WAITED 2 YEARS FOR THIS???

[whole wagon]

Even the december odds are shifting...

[feral lichen]

How much is it? Idk

[echo aurora]

Yeah, we're looking into

[meager harbor]

yeah that's crap, gemini is 1 million

[stray aspen]

waited so long for this garbage

[torn mantle]

thanks

[keen beacon]

2 years!

[tired herald]

Model selector back on again

[exotic tartan]

why is it down?

[wheat onyx]

The AIDS chart

[unborn lantern]

guys, how can i use gpt 4.5 in lmarena?

[echo aurora]

I'm seeing the same

[barren prairie]

I hope thta deepSeek won t do the same thing , waiting ages for garbage

[pulsar rain]

1 million token point is barely enough if you want it to read all the text in a book 🤣

[thorn ore]

is chatgpt-5 a joke model

i think its fake

[steady vale]

JUST IN: GPT-4.5 got removed from chatgpt's website

[thorn ore]

Oh

NOOOOOO

[stray aspen]

lets what they give us. those people never say a word

[void elm]

bro is not a twitter bot

[fleet lintel]

anthropic is still great but only for coding.. i still have hopes on them in this area

[whole sundial]

gpt 4.5's dead, dead from api and dead from chatgpt web

[warm fulcrum]

great

[verbal nimbus]

Gemini hallucinates so terribly though. It sometimes doesn't even tell me I forgot to attach a document; it just makes one up.

[void elm]

its not even removed

[prime mulch]

Backfired? Lmaoooo

[void elm]

4.5 is still there

lies

[wicked root]

How much better is gpt5?

[pulsar rain]

it's just a cheap ass model for general uses anyway

[whole wagon]

bros had to make the most misleading graph ever

[keen beacon]

This is deeply saddening.

[warm fulcrum]

@echo aurora which gpt-5 version is it that's displayed on lmarena?

[whole wagon]

thats crazy work ngl

[unborn lantern]

Thanks for your info

[pulsar rain]

first impression: concise, straight to the point

[stray aspen]

its great for lua

[verbal nimbus]

Try pasting Claude's system prompt into other LLMs. Might help.

[wicked root]

Is gemini 2.5 screwed?

[warm fulcrum]

who uses lua noob

[tired herald]

Lmfao

[hoary elbow]

I overslept and I woke up realizing that GPT five is out

[stray aspen]

roblocks

[stray aspen]

no

[tired herald]

Theres two models called "gpt oss 120b" on lmarena rn

[keen beacon]

Why did they release this? If they did it for the normies, why did they hype it so much!?

[stray aspen]

yeah thats trash

[warm fulcrum]

whats wrong with it

[hoary elbow]

GPTOSS 120 B is a open source model made by GPT. ChatGPT says it’s just as powerful as 4o

[keen beacon]

Everything!

it does not match the hype!

[stray aspen]

no its bad

[pulsar rain]

go back to sleep again then 🤣

[stray aspen]

its the worst open source model ever

[unborn lantern]

Same or bugs?

[stray aspen]

plus it has north korean level censorship

like what were they thinking

[hoary elbow]

Why

[blazing rune]

They didn't release any benchmarks for the Mini and Nano versions of GPT-5.

[echo aurora]

hmm will take a look

[keen beacon]

GUYS! GPT-5 supposed to be the best model for cost to performance though.

[blazing rune]

probably means they are about at the level of 4.1 Mini and Nano

[hoary elbow]

Is it better than Grok though?

[pulsar rain]

it barely beat gemini 2.5 pro in bendmark.

[torn mantle]

its working now @echo aurora

[hoary elbow]

Ok

[stray aspen]

no

[keen beacon]

ITS NOT A REVOLUTIONARY MODEL! its an effcient model

[rapid merlin]

good old graphs made by chatgpt

[stray aspen]

it didnt beat grok in arc agi 2

[wicked root]

Polymarket’s saying gemini’s over

[torn mantle]

so far gpt5 is good

[verbal nimbus]

Google has the efficiency advantage because of their custom TPUs. It's kinda crazy that you get free unlimited use of Gemini 2.5 Pro on AIStudio with 1M context.

[rapid merlin]

are yall testing it from lmarena?

dont think they put it on their site yet

[echo aurora]

I'm still seeing a few things off, but yeah overall should be back

[wicked root]

Wait so is it over for gemini?

[keen beacon]

"GPT-5 results on ARC-AGI 1 & 2!

Top line:

65.7% on ARC-AGI-1
9.9% on ARC-AGI-2
" IT DOESN'T EVEN MATCH o3 FROM DECEMBER AT ARC AGI 1!!!

[pulsar rain]

Nothing will replace 1M context. just paste the whole book and it know everything

[hoary elbow]

No Gemini still better than GPT five

[echo aurora]

the standard version

[hoary elbow]

I saw the benchmarks

But at least GPT five is good

[warm fulcrum]

ok thx

[keen beacon]

LOGAN! say it! say the damn words! "Gemini Gemini Gemini"!!!

[astral jetty]

Does that automatically include thinking

[hoary elbow]

[unborn lantern]

They Didn't increase their knowledge cut off parameters

[verbal nimbus]

GPT-5 kinda falls short on the context window size and knowledge cutoff date.

[stray aspen]

this one?

[wintry tinsel]

Gemini may be the best now but google will neuter it once they have market monopoly we need heavy competition to keep Gemini good

[tired herald]

Damn

[tired herald]

Where is this picture from

[hoary elbow]

Is there another way to get GPT five to search

[fleet lintel]

why that old?

[hoary elbow]

Because Joe Biden is not the president

[torn mantle]

nah its actually so good

[hoary elbow]

Not anymore

[torn mantle]

i have many things to say

[whole wagon]

bruh

this is literally stale model wth

[stray aspen]

from openai models

[tired herald]

Dem

[devout vault]

Why is the cut off day 2024 oct

[verbal nimbus]

Bad data? Idk. I thought Sam Altman said it was a router model though. (few months ago)

[patent bane]

gpt-5 on chatgpt is dumber than the one in API????

[whole wagon]

thats how long it took them to get this release out 💀

crazy

[echo aurora]

Should be fixed now

[tired herald]

Openai logins are broken lol

[keen beacon]

"gpt-5 fast facts:

1. hits sota on pretty much every eval
2. way better than claude 4.1 opus at swe

3. 5× cheaper than opus

4. 40% cheaper than sonnet

5. best writing quality of any model
6. way less sycophantic" - OpenAI employee

Roon failed us.

Trash writing still

[verbal nimbus]

The mini version has even an older knowledge cutoff date

[void elm]

barely any better

[indigo flax]

“Black and white vector-style silhouette of a confident bearded man wearing sunglasses, modern hairstyle

[tired herald]

They ought to make a good 1m parameter model

[void elm]

openai has to retire and give its compute to google

[stray aspen]

google deepmind has to

[solid brook]

the nano

[keen beacon]

Secret Gemini models have better writing!!!

[tired herald]

Those are garbage

[whole wagon]

WTF

how the hell

[tired herald]

Google already have good ones, OpenAI dont at all

[whole wagon]

its a mini model bruh it should not take a year to train

[verbal nimbus]

The one in ChatGPT is a different model with a different cutoff date. https://platform.openai.com/docs/models/gpt-5-chat-latest vs. https://platform.openai.com/docs/models/gpt-5

[devout vault]

Gemini 3 and grok 5 will win 100%

[stray aspen]

the chatgpt 5 is dumber than the API

[tired herald]

Not even nano has 1m

[quiet moss]

is chatGPT 5 on the website yet?

[patent bane]

that"s what I said...

[rapid merlin]

so the one in chatgpt is lobotomized?

💀

[stray aspen]

didnt someone from google say this was gonna be an exciting week

[barren prairie]

I am not a google employe

[tired herald]

Of course they will, they have to release a "competitor" to this new trashai model

[sour spindle]

What model has had the biggest positive reception here?

[stray aspen]

gemini

[verbal nimbus]

Claude is the most self-aware and agentic still. Gemini doesn't even tell me I passed in the exact same attachment twice, whereas Claude is like "HoLdUp"

[hoary elbow]

Can I send a video here real quick?

[devout vault]

i was the first to say that LOL

[devout vault]

true

[tired herald]

Gemini 2.5 Pro already has many things better than gpt 5

Gemini 3 Pro will be groundbreaking

[stray aspen]

yeah its pretty full model

gpt 5 is garbage

[hoary elbow]

Gemini three pro might be better than Grok four I mean it has a chance to be better

Since 2.5 pro is better than GPT five

[stray aspen]

kimi k2 are you serious ?

[keen beacon]

HOW IS GPT 5 WINNING???

[hoary elbow]

I wonder how Gemini will be

[analog bone]

Where's GLM ? 🙁

[pulsar rain]

default gemini 2.5 pro praises all your question no mater how stupid it is. It get very annoyed

[hoary elbow]

Wait, it’s winning

I can’t believe it

[stray aspen]

glm aint sota

[void elm]

its such a minor upgrade

[keen beacon]

Stop with the A = A arguments. its a trash model sir.

[tired herald]

Now make a simple prompt against that behaviour and put it into system prompt

[verbal nimbus]

Try pasting the Claude system prompt into it

Gemini is definitely sycophantic, lol

[keen beacon]

What do you mean?

[tired herald]

Well, at least gpt 5 nano accepts pictures

[sour spindle]

Also does anyone have access to to 5 right now lol

[steady vale]

gemini is the most sycophantic model

[whole wagon]

this guy on the livestream is just bsing

[tired herald]

Lmarena

[whole wagon]

literally saying nothing

[stray aspen]

yes

[keen beacon]

It literally is not! arc agi is only one of the benchmarks where it falters!

[stray aspen]

on lmarena

[void elm]

filler words

[stray aspen]

now the livestream is just pure yapping

[whole wagon]

selling

[slow sail]

He tried his best

[whole wagon]

we will get back to selling kek

[verbal nimbus]

They're actually tied, as the confidence interval overlaps.

[pulsar rain]

is it this one? https://www.reddit.com/r/ClaudeAI/comments/1ixapi4/here_is_claude_sonnet_37_full_system_prompt/

[whole wagon]

thats not how the betting works

[keen beacon]

Will they do the pokemon Bench with GPT-5?

[whole wagon]

it is the actual value only

I find it funny they copied gemini 2.5 pro pricing exactly lol

like $1.25/$10

[hollow imp]

Wait is gpt5 a reasoning model?

[short adder]

Could we get gemini 3 this month？

[stray aspen]

yes

[keen beacon]

Roon did not fail us

[verbal nimbus]

The official one is here: https://docs.anthropic.com/en/release-notes/system-prompts. I removed unnecessary lines (Anthropic product info).

[void elm]

did you even watch the stream?

[balmy mist]

why some people hating on GPT-5?

[void elm]

because its like upgrading claude 4.0 to 4.1

negligible upgrades

[hollow imp]

Bro can I dm you
I want to talk a lot about gemini 2.5 and stuff. Not the news or about the model but about the uses of the model and my personal experience and cases

[hollow imp]

I didn't

[stray aspen]

they spit in our faces again

first with this garbage open source model

and now with gpt 5

[hollow imp]

What happened