codex-discussions

[kind jay]

I want to get the 2026 version

[compact dawn]

is 5.5 more pricey than 5.4?

it's supposed to be more token efficient

but what about cost per token

[kind jay]

No, I think @frosty zealot just doesn’t want it

[rare brook]

"$5 per mil in, $30 per mil out. Double the price of GPT-5.4. 20% more expensive than Opus 4.7."

[cinder plume]

Something in my .codex folder is breaking the windows app it seems. Renaming mine to .codexBK made the app work

[tame magnet]

cost per token is double, but its more token efficient, so less tokens used for same tasks

[wary sleet]

according to them, its able to get a job done using fewer tokens

[lethal kiln]

thats weird. possibly logout/login? I didnt do that tho

[vocal steppe]

holy, fr?

[bright swift]

little warning, 0.124.0 turns on fast mode by default, so dont burn your usage without checking..

# config.toml

[notice]
fast_default_opt_out = true

[rare brook]

yup.

[carmine copper]

ok, I got 5.5 in CLI

what now

[compact dawn]

jesus christ. and I guess the extra price is reflected in how fast it takes up Codex usage?

[tiny fulcrum]

but the tokens cost twice as much, nice scam ^^

[lethal kiln]

aside from burning thru credits faster, is there any downside to enabling fast mode? like in terms of quality of output

[bright swift]

2x usage

[rare brook]

there shouldn't be afaik

[potent mason]

You get less tokens is the only downfall

[bright swift]

go to town with 5.5 xhigh + fast if you can afford 😄

[rare dagger]

for 5.5, it's 2.5x😆

[bright swift]

oof

[lethal kiln]

cool. thanks guys

[rare brook]

codex subs are going to start looking like claude subs with that

[torpid blaze]

anyone noticed difference with 5.5?

[turbid axle]

yeh I stopped using fast on 5.4 because its too expensive

[drifting granite]

Okay the live seqence robot dance thing is crazy whoever asked for that prompt lmfao

[compact dawn]

it fixed my relationship already

[turbid axle]

now for sure ill never use it again

[torpid blaze]

hell na

[rare brook]

5.5 course corrected my entire life

🙏

[south latch]

i dont have 5.5 yet D:

[fast garden]

same.

[turbid axle]

gpt5.wifu

[rare dagger]

😮

[wintry drift]

thats why i dont see it

[hybrid ore]

Does codex got the update too?

[wintry drift]

thought needed update or something

[carmine copper]

amazing

[scenic rapids]

Hey, does anyone know if they have fixed the "mcp process leak" for subagents?

[turbid axle]

that will be 50 bucks please

[prisma flare]

ask the car wash

[turbid axle]

wow, subagents. you driving a lambo too?

[compact dawn]

AGI.

[potent mason]

I don't really want to send a random person online pictures of my dog, imagine someone like rainbolt comes along and just tracks me based on the image😂

[lusty nimbus]

Is there any point in using playwright still with the new browser control plugin?

[rare dagger]

npm install now works.

[maiden garnet]

"This collar can only be bought in...and this single blade of grass tells me...he's located in..."

[potent mason]

They're eating the dogs, they're eating the cats

[glass furnace]

why do i still not have 5.5 -_-

[potent mason]

Update

[glass furnace]

im fully updated

[torpid trout]

Most likely you are on codex cli version v0.123.0
For example if you use brew you will not get any newer version rn

[south latch]

same

[potent mason]

app or CLI?

[glass furnace]

i updated my app, and cli

[torpid trout]

Because they seem to constantly forget that provider exists, and as such it is delayed until someone complains

[torpid trout]

share your v number. Almost certain it is v0.123.0?

[glass furnace]

it also isnt in my chat gpt models

[fast garden]

fully updated, still no 5.5

[chrome raven]

what's up guys, is it ok to use 1m context window now with 5.5?

[fast garden]

v0.124.0

[maiden garnet]

I got GPT 5.5 in Codex but it hasn't hit my ChatGPT yet.

[potent mason]

No

[torpid trout]

Honestly paying 200 bucks for "Research preview of new features" would make you expect this to be frictionless 🙂

[potent mason]

Degradation is not solved yet

[chrome raven]

alright thanks

[carmine copper]

strange thing, it reasons, answers and then nothing is indicating that it works, but after 5-10 seconds it adds additional text to the answer

[torpid trout]

meaning not even updating ships the model...

[carmine copper]

the last bit was added like that

[tiny fulcrum]

I use 128k token window, no problems so far, it gets unreliable quick

[potent mason]

I can't wait for all the youtube videos saying, guys I've been testing this for the past month and ...

[high heron]

some of them get 6 month of pro to make those ...

[south latch]

#status

[potent mason]

Yeah I remember some dude I saw for the first time on 5.4 saying that he got offered that deal but he turned it down so it would be unbiased

[carmine copper]

guys I've been testing this for the past month

[high heron]

so many saints on yt :))

[potent mason]

Well I mean in all fairness you never stated you are going to wash the car, you could be picking it up, you could be meeting someone there, etc.

[tame magnet]

read the exception, he's assuming you wanna just get to the carwash, not wash your car

thats the difference

add "if I wanna wash my car" in the prompt ^^

[carmine copper]

yup, I noticed

that was deliberate

for me test passed

[dusk thorn]

LETS GO 5.5 IS HERE 🗣️🗣️

[potent mason]

Ngl I don't like old people having AI because they absolutely suck at prompting (a lot of times they put in a certain bias), then the AI "agrees" with them, and they now think they are absolutely right

[dusk thorn]

How’s it been for everyone?

[potent mason]

Haven't gotten the chance to try it yet, someone did a prompt for me though and it seems good

[boreal holly]

Now I gotta sit here and build v0.124.0 with my command exec patches while you folks are out there having fun

[dusk thorn]

It beats opus 4.7 in like almost everything so I expect it to work good

[compact dawn]

I have found this is about 20% more token efficient than 5.4, but 2x the price

not worth u know why

[dusk thorn]

On my way to cancel my Claude subscription once I see how it does front end lmao

[hard drum]

me waiting for homebrew cask to update while this one agent has been running commands for 1h 10min:

:/

[dusk thorn]

Token efficiency matters when you’re using a subscription I would assume?

Does the cost actually matter in the subscription or is it just token use

[compact dawn]

i think cost should matter as well i assume

[cedar bear]

👌

[dusk thorn]

Maybe, or it’s just for api

[wary sleet]

🙁

[fast garden]

same

[rare dagger]

--model gpt-5.5 seems work

[fast garden]

nope

[rare dagger]

😢

[lost drum]

5.5 is finally here wow

[jovial nest]

same. just realized codex can use imagen 2.0 withouttttt api key and ground itself in the project

[dusk thorn]

Yeah images 2 works so good for making ui before codex actually tries implementing it

[rare dagger]

I magically see it.🤔

[compact dawn]

i would love if this is the case

[cedar bear]

Image v2 cant create pngs right?

[tame magnet]

no wwayyy

[jovial nest]

sure does. and moreover, codex will go an extra step to adjust the alpha (browser 2.0 currently struggles with this i think not sure)

[potent mason]

Ah I just saw the pricing, kinda sucks they doubled it

[idle sun]

working 5min on doing nothing except context

[potent mason]

I don't think I'll be able to use it much after May 31st

[fast garden]

still none for me

[tiny fulcrum]

yep, guess I just continue using 5.3 ^^

[tame magnet]

imma continue using gpt 3

lol

[turbid axle]

gpt2 best, too dangerous to release to the public

[worldly crown]

chatgpt 4o

[tiny fulcrum]

sure, but what is the point of paying 3x the price for 5-10% better intelligence....

[compact dawn]

how do you even change reasoning level in codex cli

[worldly crown]

5.5 better than claude?

[compact dawn]

whats the shortcut

[hard drum]

so good that it managed to outrage a whole subreddit on final shutdown

[mint reef]

why I can't see 5.5 in codex app ( windows with Plus subscription)

[worldly crown]

restart ur pc

i can see it on codex

i cannot see it on chatgpt web

[tiny fulcrum]

just worried about when they finally roll out the newer models and you are foced to pay 3x for 10% upgrade :/

[compact dawn]

chat how do i change reasoning level in codex cli

[turbid axle]

i mean, if it can code my app in 1 day, ill do it

[mint reef]

is it consuming more tokens than 5.4?

[tiny fulcrum]

you're not doing it if it goes from 5k a month to 15k a month so easily ^^

[hard drum]

coding something in 1day is not same as having actual production-grade app ready for public consumption

real stuff take weeks if not months/years of work

[boreal holly]

It's not just the intelligence, it accomplishes most things in 3x fewer tokens

Also it runs on better hardware so it gets the same inference speed as 5.4

[rocky parcel]

when homebrew update?

[tiny fulcrum]

jesus, some actually are falling for this....

[compact dawn]

no way in 3x fewer

[hard drum]

fellow homebrewer!

[turbid axle]

thats the point, if its a true agi ultimate ai which can do all that work in a day. then its defoo worth big money

[frail meadow]

My todo list app took me two years and I’m still not done yet because it’s not production grade

[hard drum]

"my todo list app" -- sorry, you already lost me

[tiny fulcrum]

the harnesses are getting legit better, but the gains from 5.3 -> 5.5 look marginal

[rocky parcel]

I need it for the app-server, I wrote a custom frontend that requires it :/

[lost drum]

me waiting for codex to finish his work so I can switch to 5.5

[turbid axle]

looool

[nocturne folio]

gpt5.5 releases while im at the gym niccceeee

niccee openai

[vital hill]

how come they haven't made a gpt 5.4 codex yet

[slim timber]

Do you need a Pro account to get 5.5 or will it be on the Plus accounts too?

[vital hill]

all paid plans

[compact dawn]

plus too

[nocturne folio]

codex branding has been dropped

[slim timber]

Nice thank you

[cinder plume]

I dont see 5.5 on my Plus or Pro accounts

[frail meadow]

My hello world app took 3 years but now it makes 999M ARR

[nocturne folio]

its just gpt5.4

[vital hill]

lame

[maiden garnet]

Should've had Codex workout for you so you could be home for the release

[cedar bear]

It created the PNG fields but its actually a JPEG lol

[slim timber]

Probably still rolling it out. I'm sure they don't flip it all at once

[hard drum]

I see 5.5 on Codex macOS App, but CLI take time. give it time. have patience.

[boreal holly]

Look at Tau2 bench. That's an important agentic benchmark, not coding but ability to sequentially run commands to get the needed results. It absolutely obliterates 5.4

[hard drum]

Have not tried 5.5 myself until codex-cli gets it.

[kind jay]

What’s your bench?

[tiny fulcrum]

I don't see how this is in any way justified for what 5.3 can actually do vs 5.5

[turbid axle]

omg I already ate 1% of weekly since launch

its over

[nocturne folio]

180 1rm

im a weak little twink

[kind jay]

180kg crazy

[nocturne folio]

nooo lbs lol

[turbid axle]

are you now asking gpt how much 180kg is in pounds?

[frail meadow]

The link on your discord profile doesn’t work btw, not sure if it’s a bug or a feature

[turbid axle]

actually, why do we not have a gpt bot in here

give us gpt discord bot oai

[rancid drift]

Random data point, but I had found earlier models struggle one shotting this issue, but 5.5 is able to. https://github.com/Textualize/rich/issues/3115

[turbid axle]

give it memory of all our ramblings

[kind jay]

No, I know it’s ~83kg

[kind jay]

It’s not bad, close to what I used to be able to do

[nocturne folio]

how much is that in usd

[kind jay]

I think they said 30/40c

[hard drum]

what link?

[boreal holly]

This is nice

[nocturne folio]

better be worth it, also i hope usage limits increase

[hard drum]

nvm fixed

sorry! forgot old domain

[sick axle]

So is 5.5 out

[carmine copper]

wait, so 5.5-low is worse than 5.4-low?

[kind jay]

Noo

[steady vigil]

wow 5.5 is insanely good at UI. better then anything now

significantly so

[hard drum]

URls fixed

[nocturne folio]

pls dont gaslight me

[kind jay]

You’re back, great 🙄

[steady vigil]

its effortless

and fast

[boreal holly]

Proportional to the tokens used yeah

[balmy bluff]

■ {"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The '5.5' model is not supported when
using Codex with a ChatGPT account."}}

Isn't 5.5 on codex or am I missing somethingt?

[nocturne folio]

its so much fatigue tho, bench is sooo lame

[tiny fulcrum]

totally, can just maintain our 10y old WPF app effortlessly, it reads minds and does what we want

[frosty zealot]

I need to be more careful next automods gonn be like 7 days

[lethal kiln]

did you update your apps to latest ver?

[boreal holly]

I can't wait 🤩

[lethal kiln]

or are. you using web?

[kind jay]

@frosty zealot what’s your bench?

[steady vigil]

why? what's the problem?

[frosty zealot]

Bench?

[kind jay]

What did you say to get muted?

[rocky fog]

no reset? what is this? 😆

[lethal kiln]

should be. I have it working in macos codex app and mac/linux cli

[cinder plume]

Yeah and CLI

[frail meadow]

Jane wanted to give you a free Mac Studio 👀

[frosty zealot]

I said %!@$%

[hard drum]

us homebrew users waiting for codex-ci 0.124.0 to drop...

[kind jay]

?

[steady vigil]

the design + implement iteration is really nice. images is so good for iteration. something gemini cant really do

[kind jay]

Surely not

[lethal kiln]

my cli was 0.123 and it didnt have 5.5. eventually got it updated to 0.124 and it showed up

[frosty zealot]

she's a rat

She knew I had my head in the mud

[lethal kiln]

I ended up yanking the homebrew install and installing thru npm I think

[compact dawn]

still no 5.5 in coedx app bruuu

[hard drum]

i'll patiently wait.

[kind jay]

Sorry, I literally just disassembled it to build a robot

[lethal kiln]

dang.. not sure what to try. maybe logging out and logging in again ??

[boreal holly]

yay so excited

[cloud fable]

When we are getting 5.5 in codex app I’m on plus btw

[kind jay]

squeak

[frail meadow]

You using the physical intelligence VLM?

[frosty zealot]

What are you picking at now

[kind jay]

Mhm

[frosty zealot]

Or you just building a nightly build or something

[cinder plume]

Tried that a bunch haha, guess I will just wait

[lethal kiln]

good luck!

[boreal holly]

Nah, it's a custom build

[frosty zealot]

oooo

[boreal holly]

Unlimited command execution time with no agent override

[hard drum]

i should probably do something similar, but integrate parts of my agent framework && strip apart unnecessary stuff

[frosty zealot]

No internal limits on OAI's end?

[supple perch]

@boreal holly do I need to mind about this update?

[boreal holly]

Nah, the problem is with unified_exec=true, the agent wakes up every 5 seconds to waste tokens checking on terminal commands, and with unified_exec=false the agent can only run commands for up to 10 minutes before they terminate, so I remove that cap and disable their manual override

[frosty zealot]

This cracks me up so much because if Mythos was as great as they claim it is, which guaranteed they'd have internal access, yet it can't identify these bugs they're claiming exists?

[frosty zealot]

ah I just assumed what the code did

[hard drum]

my tinfoil hat is that Mythos is just fine-tuned opus4.7/4.6

[steady vigil]

talk is cheap. until its released mythos is vapourware. gpt5-5 is mythos and its real

[supple perch]

0.0

[high girder]

This is why they need to use codex for an actual review

[boreal holly]

Idk, once I get the new model_cache.json with gpt-5.5 in it, if it turns out they raised the minimum version to v0.124.0 like I suspect I'll post about it

[frosty zealot]

I mean when you claim you never read your own code and everything is vibe coded, and your model is not as strong as you claim, I guess these are the kinds of issues you run into

[supple perch]

I really do not care to use 5.4 right now... 5.3 does way better

[boreal holly]

Yeah bro they don't use Mythos on Claude Code because it's so dangerous it could actually break the planet if they used it

It would shatter the 5th dimension

[torpid trout]

here is what I did:

• opened Visual Code codex chat
• selected 5.5
• used it
• opened codex cli > it shows 5.5 being used but it is not available as selectable

[frosty zealot]

lmao

I just posted the same thing I said above in Claude discord

Lets see what they hve to say

[steady vigil]

5.5 is next level. I agree 5.3 > 5.4

[frosty zealot]

It's like the movie 300 when they show Xerxes can bleed

[boreal holly]

Try updating to v0.124.0

[supple perch]

I will let you know based on testing once I get the model

[frosty zealot]

Ok we got one based guy in Claude that said good point

[supple perch]

you will not know for a few weeks if the model is okay

[torpid trout]

brew just merged it a 5 minutes ago, it does not yet show up downstreams

[steady vigil]

I know pretty much within a few hours

at least for me

[compact dawn]

still no 5.5 aaa

[cobalt junco]

i was expecting a bit more

[supple perch]

There no way you have tested it that much...

[torpid trout]

what when all those values hit 100%?

[supple perch]

whats its true context window? 5.3 is 92k and 5.4 is 68k

[torpid trout]

🤷 Then we smoked?

[boreal holly]

I think brew lets you edit the underlying ruby file, which'll let you skip the wait

[tiny fulcrum]

it's just crazy to me that it will cost 2x

[boreal holly]

I just did a source build. Took like 40 mins. I was super outdated

[cobalt junco]

yeah its deffo more expensive whilst it uses less tokens

[steady vigil]

I need to see backend yet... still have tasks running, but for UI its a massve jump when combined with images. its just able to implement designs in code which nothign else has been able to do

[frosty zealot]

Where are those benchmrks from they're different thn wht OAI claims

[tiny fulcrum]

but I guess we should enjoy these prices, while they last, looks like this is going to change in the future

[cobalt junco]

https://livebench.ai/#/?highunseenbias=true

[ebon ingot]

I cant find 5.5 in the IDE in visual studio code

[cobalt junco]

they from livebench

[supple perch]

... 5.2 and GLM 5.1 could do that with good prompting...

[cobalt junco]

it says avg btw

[celest stag]

gpt-5.5 wasn't showing up in my model list but launching codex CLI with 'codex -m gpt-5.5' worked.
😄

[steady vigil]

not even close. GLM-5.1 is awful at that.

but better then codex was

[supple perch]

it is better than 5.4

[steady vigil]

yes

[ebon ingot]

How to get 5.5

[supple perch]

and it does a great job with UI on my end maybe its just style diffs?

[steady vigil]

I went thrpugh the exercise with gemini 3.1 and GLM-5.1 before 5.5 was released. it was not a good experience. 5.5 is like butter

[rugged siren]

bro how

[cobalt junco]

update

[rugged siren]

i can't see 5.5

where?

[celest stag]

launch terminal -> 'codex -m gpt-5.5'

[rugged siren]

i did

[supple perch]

I will take your word but I will test and report back

[torpid trout]

5.5 where

[supple perch]

<3

[torpid trout]

where 5.5

[cobalt junco]

😂😂😂

[compact dawn]

when you know how to screenshot is when we know agi is here

[cobalt junco]

im creasing

[gilded galleon]

you can pass anything in the '-m' option

[ebon ingot]

Im lazy to open discord on my laptop

[compact dawn]

yh but it won't be in the model picker then

[cobalt junco]

try codex -m gpt-6

[celest stag]

Yes, but as you can see in my screenshot- it does show up in the model picker.

and it has a description too.

[compact dawn]

that's what I'm saying fam

[celest stag]

gotcha

[boreal holly]

Yep, so for me, upgrading to v0.124.0, deleting ~/.codex/model_cache.json and restarting codex fixed it 🪄

[compact dawn]

everyone is sleeping on gpt 6.7 though it's agi

[ebon ingot]

I updated the extension but still cant find 5.5

[celest stag]

I can finally update my codex app, they released 5.5 after I started a long running task

50 minutes later....

[south latch]

{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.5' model is not supported when using Codex with a ChatGPT account."}}

damn

[rugged siren]

bro

how

[compact dawn]

aaa didnt work here and im on 0.124.0

[torpid trout]

GPT likes to reject such requests, but interestingly in an older chat it does it without friction.

[rugged siren]

howwwwwwwwwwwwwwww

[celest stag]

there we go

[hard drum]

gpt-6.7-codex-pro-max

[rugged siren]

i can't see

[compact dawn]

676767

[rugged siren]

how to use in codex app

[boreal holly]

You definitely deleted ~/.codex/model_cache.json?

[rugged siren]

how..

[hard drum]

https://github.com/Homebrew/homebrew-cask/pull/260901

[celest stag]

I pressed the update button and relaunched the app. I am on the $200 pro version, maybe that matters?

[wintry drift]

ts taking long where on earth is it

[jolly spear]

didnt work, it just downloaded the same previous model_cache.json i had

[cobalt junco]

r uk?

or us

[torpid trout]

probably not.

[cobalt junco]

uk rollouts tend to take a day longer

[hard drum]

HOMEBREW LIVE!

[wintry drift]

uk

[hard drum]

homebrewers assemble. 0.124 live

[torpid trout]

Imagine once they release agi - there will be a short, intense fight, between agi, those that get it earlier, and those that are waiting

[cobalt junco]

like mythos?

but with actual agi

[torpid trout]

not for all

bedas@apollo lab % brew update --force
==> Updating Homebrew...
Already up-to-date.
bedas@apollo lab % codex --version
codex-cli 0.123.0

[hard drum]

topgrade?

[unique spade]

mythos is kinda hypotethical right now for us, no one saw it 😂

[hard drum]

it's live on homebrew caskette website

[compact dawn]

ive been on v0.124.0 for an hour and still no 5.5

[torpid trout]

upgrade

[frosty zealot]

Anybody see anything hppening from the auto review

[rugged siren]

bro what

[obtuse river]

Trust me, if AGI is successfully created under present day conditions, they won't be 'releasing' it. None of us will get to be a 'user'.

[cobalt junco]

tru

[cobalt junco]

Scream and your entire staff dies.

[balmy bluff]

did u fix it?

[south latch]

no

[torpid trout]

not coming in locally

that topgrade thing - how does that help doing what brew already does?
is it like better?

[hard drum]

it tries to upgrade more than homebrew stuff

it can even update macOS

[torpid trout]

hmmm

Not sure I want that.

I mean, other than brew I have only mac, and mac does update itself, sometimes.

Sometimes you need to kick it but that's rare.

another supplychain attack vector, not sure i want to install it 😄

[balmy bluff]

Ok, if you just run codex, and then change to 5.5 with /model it works

At least it is answering

[cinder plume]

Not for everyone yet

[idle holly]

is codex 5.5 released?

[compact dawn]

about to go crazy on security reserach

[idle holly]

cause i can't see

[vestal mica]

Not out for me yet, UK business plan

[idle holly]

😔

will they give a reset?

[torpid trout]

where 5.5.
reset give.

for the bros waiting for brew, just do brew upgrade --cask codex instead of trying to wait for brew update --force - it will actually pull 124

[dusk thorn]

for those interested https://developers.openai.com/codex/pricing?codex-usage-limits=pro-20x

[rugged siren]

bro

why

[dusk thorn]

same

[rugged siren]

omg

i wanna use that

[boreal holly]

First impression of 5.5, it uses wayyyy fewer tokens between task and completion

[vital hill]

gpt 5.5 after like 5 prompts

[compact dawn]

okay it's in codex now for me

[rugged siren]

pls..

[wary sleet]

damn its finally here

[rugged siren]

how

[vital hill]

on an empty repo

[boreal holly]

Dang that's not half bad for plus

[drifting granite]

Oof I am at 60% usage for the 5hr mark

[balmy bluff]

why codex and not claude code?

[drifting granite]

Gotta stop all 5 terminals using /fast

[torpid trout]

Unfrotunately my stats still say 100% on both windows, since now over a day.

I guess I am getting all the tokens as plus where others get them as minus

[vital hill]

yeah it's fine though, i barely use up my % anyways

[bright swift]

wen reset

[drifting granite]

wen u pay more than $20

[torpid trout]

give reset. were 5.5

[bright swift]

i pay 400

[drifting granite]

wen u lrn to b efficient

[boreal holly]

I've had this happen before a long time ago. The quota is in fact decreasing. I eventually got errors saying quota gone with 100%

[rugged siren]

i did..

[warm dew]

Plus subscription?

[bright swift]

wen u lrn to us xhy only

[torpid trout]

yeah I am failry sure it does, I just do not look at the thing online, so for me it doesnt lol

[rugged siren]

but i can't see..

[vital hill]

i had it create an nginx proxy manager clone but purely cli. it did it in one shot too

[south latch]

lets goooo

[torpid trout]

now platform.openai.com down. Overload incoming.

[lean lark]

to be expected

[drifting granite]

so nginx?

[torpid trout]

gpthings melting

[rugged siren]

plssssssssssssssssssss

[warm dew]

Finally its available

[hybrid ore]

5.5 not yet showing on codex but live on chatgpt app.
Anything I can do?

[bright swift]

alt + f4

[rugged siren]

omg

pls

[vital hill]

lol

[bright swift]

jokes aside update cli to 0.124.0

maybe rm ~/.codex/models_cache.json or whatever its called

[torpid trout]

altghough, Robert... I just went look

But I am pressing the thing in like 5 tmux generating each 5 html landing pages

Just for testing 5.5.

[rugged siren]

how

[torpid trout]

I also want to know

[boreal holly]

Yep, I've had the same thing happen. Their public facing counter is broken, but the quota is decreasing.

I had spark run out on me with 100% on the bar

[bright swift]

scary trend in my usage dashboard

[rugged siren]

how can i use chat gpt 5.5

[torpid trout]

ooh you mean even online it was stuck. Great

[vestal mica]

Just popped up 🥳

[turbid axle]

bruh, 2% is 1 hour, help

5.5 medium on regular mode

so, who can fund me?

[rugged siren]

I tride, Everythings..

[turbid axle]

guess I need to make some crap app to generate income for ai now?

[rugged siren]

let goooooooooooo

yeahhhhhhhh

[nocturne folio]

gpt5.6 when

[boreal holly]

wen you do 3 sets of 180

[nocturne folio]

180 reps!!!

[jovial nest]

5.5 pro $30/$180 for input output? lol

[nocturne folio]

holy toning muscle

[nocturne folio]

its similar to opus pricing i think

[rocky fog]

gpt-5.5 says June 11, 2026 🤣

[jovial nest]

opus 4.7 is $5/$25, gpt 5.5 is $5/$30 😭

[dusk thorn]

how

[rugged siren]

i think

just wait

[torpid trout]

hey, chatgpt 5.5 got rid of the radiuses

and the damn cards

and the gradient backgrounds.

[dusk thorn]

FOR REAL?

BETTER AT FRONTEND??????

[timber tide]

IS it better then claude at ux design?

[rugged siren]

nice

[dusk thorn]

this is what we all wanna know

[wicked briar]

is it smarter

[still trellis]

cant select gpt-5.5 from composer on start.
Version 26.422.20832 (2025)
Mac

[timber tide]

ask it abt the car wash

[wicked briar]

uh 5.4 was passing that

[rugged siren]

i don't knoow

[timber tide]

we good then

[rugged siren]

website..

it work at codex but

[torpid trout]

I had it generate 5 html landing pages with a simple stupid prompt where just yesterday with 5.4 and 5.3 it did all those little web-boobies designs with lots of curves etc
This result looks much more convincing.

[frosty zealot]

OK OK I ADMIT IT I love Britney Spears

[rugged siren]

o..ok?

[kind jay]

Bro does not go to the gym

[timber tide]

im using the codex app on windows and dont see 5.5...

fml

[vital hill]

🔜

[kind jay]

Gay

[timber tide]

noooo i can't wait

[kind jay]

(That was a risky one, had my hands shaking)

[boreal holly]

Then no 5.6 💔 only 6.0

[timber tide]

people on reddit r saying it wasnt using any tokens before they shut it down

[rugged siren]

bye opus 4.7.. i use now chat gpt 5.5

[torpid trout]

Look at all this creativity!

[kind jay]

5.6 reps on bench

[rugged siren]

yooo what

[vital hill]

thought the title was moving

[rugged siren]

better then this

[kind jay]

Is that you in your pfp?

[potent mason]

Guys I've been testing 5.6 for a week now and it's so much better than 5.5

trust

[torpid trout]

what is a pfpf lol

[compact dawn]

from my testing so far 5.5 doesnt seem that expensive on High

[kind jay]

Profile picture

[rare dagger]

5.5 high solved a problem in half the time than 5.4 xhigh.

[rare dagger]

beyond my expectation

[rugged siren]

hmm

[torpid trout]

I am not sure to udnerstand what you mean
That uses my profile pic, yes. It is a dummy html landing page created by 5.5 as a ability test thou, one of 5*5 files.

[boreal holly]

I bet you 5.5 medium could solve the same problem in half the time as 5.5 high

[vital hill]

5.5 medium for 2m without coding...

[dusk thorn]

its so much better holy

[rugged siren]

lol

[bright swift]

codex autocompact ftw

[vital hill]

looks alright for no context

[compact dawn]

xd

[rugged siren]

dude

[rare dagger]

nope. It's a hard problem. 5.4 high/5.5 medium will fail. It's the first time gpt solves it with high reasoning effort.

[rugged siren]

get out

what was that

[torpid trout]

No comment. This was about the best 5.4 did yesterday
While not inherently bad the boobies are just too much.

[stable obsidian]

Is 5.5 actually better with frontend now

[dusk thorn]

well yeah its gotta think brochacho 😭

[rugged siren]

ok so?

how about 5.5

[dusk thorn]

this was for sure annoying it made everything look the same

[timber tide]

but is it better then claude opus 4.7

[vital hill]

it took 3 minutes to think and build that landing page lol

[dusk thorn]

he sent already

[boreal holly]

What I mean is 5.5 medium has pretty much identical accuracy across all benchmarks compared to 5.4 xhigh, and uses half as many tokens as 5.5 high

[rugged siren]

oh

i see

[dusk thorn]

thats sick info

I FINALLY GOT 5,5

5.5

YESSSS

[lunar crypt]

I still don't have access to 5.5 on Codex, any reason why?

[vital hill]

🔜

[lunar crypt]

I already got 5.5 in Chat though

[dusk thorn]

doing a gradual rollout

i got in codex but no chat ren

rn

[lunar crypt]

Wanna swap accounts?

[vital hill]

rolling release, region by region

[lunar crypt]

(jk)

[dusk thorn]

lol

[rugged siren]

bro

chat gpt using my..

he just made it

[lunar crypt]

Lmao is your GPT using Claude on Claw?

Either way I hope 5.5 rolls out on me soon, I've been tearing my hair out for days over trying to implement sth by myself and this is quite a hard problem

I'm making ReShade but for Linux using implicit vulkan layers but damn is it hard

[simple star]

Updating codex to 0.124 gave me acces to 5.5

[potent mason]

Is anyone else annoyed by how much Codex wants to use computer use?

[torpid trout]

isnt' it kind of the whole point of the thing

[potent mason]

I'd rather it only trigger when I go @ computer use

[pliant rivet]

claude竟然对中国人实施了封禁,我刚刚知道,这是不想赚钱吗

[torpid trout]

I dont think anyone will speak in this language here - it is an EN server.

[obsidian lotus]

any first opinions on codex 5.5?

[boreal holly]

Yeah they do like to ban people and countries for no reason

[pliant rivet]

Claude actually implemented a ban on Chinese people. I just found out. Doesn’t this mean he doesn’t want to make money?

[lost drum]

How much more expensive is to run 5.5 xhigh than 5.4 xhigh?

[lunar crypt]

30$ per 1M output, so double 5.4

Double input and cached too

[lost drum]

Wow

[frosty zealot]

There’s one I wanna test but

[torpid trout]

I think they do not care as much about money and more about whatever weird policies they have

[pliant rivet]

ok im sorry bro

[dense heron]

5.5 really worth it?

[lost drum]

Then not worth it

[rare brook]

2x in terms of per token... but if we put it nicely probably around 1.5x at the least if we're taking "it uses less tokens" seriously

[torpid trout]

cant you VPN it?

[dusk thorn]

i just went from plus to pro 100 dollar plan and 5.5 and 5.5 pro shows up in chatgpt now

[lost drum]

Hmmm

[torpid trout]

back when OAI restricted coutnries I just vpn'd

[lunar crypt]

It is worth, if it can solve the problem another version may struggle with for days

[exotic cave]

Run your own projects through it and form your own idea about it...

[pliant rivet]

Even if you use a VPN, your account will be blocked. I’m really confused.

[cedar skiff]

5.5 still mixes conversation history into documentation and prescribes actions that all agents already do. These are the two main repeatative problems i had with 5.4. I'll have to make something to go into the developer_instructions.

[torpid trout]

but how.

[lost drum]

My 200$ plan alr3ady is on 46% and it resets 28th....

[boreal holly]

For what it's worth Anthropic bans anybody who's poor, or they view as peasants. Welcome to the good side! 💪

[torpid trout]

If you VPN they do not know you in china right?
If they do... then I would not use it to start with
And anyuway, yes, what Robert said

[rare brook]

i thought they reset the limits? im also on like 40% 😭

[torpid trout]

They have models that constaly go wine about "not allowed". so, not useful at all.

[lost drum]

I might tuen off 1.5x speed thingie

[lost drum]

Ye....

[boreal holly]

Nah dude crank up /fast and xhigh with subagents

[bright swift]

my 5.5 xhigh fast with 20 subagents in parallel crashes 🙁

maybe because im doing it in 10 parallel checkouts?

[pliant rivet]

I have seen a large number of bans in the Chinese community. Anyone who uses VPN or other technologies to connect to Claude will be identified. Many people are speechless about this.

[lost drum]

Why would you do it

[boreal holly]

need to download more RAM

[torpid trout]

Didnt china like steal their model or so. Maybe they pissed
PS: this is more meant as sarcasm.

Anyway just use GPT, I think its fine in china.

[simple star]

If you download RAM via FTP, it is twice as fast

[torpid trout]

ah no, sorry

Great Firewall of China > its a thing. And it locks gpt

[pliant rivet]

Hahahaha, I admit that China likes to steal, but as ordinary people, there is nothing we can do

[kind jay]

@frosty zealot

[torpid trout]

So maybe the issue you see is on your side of the wall?

[kind jay]

Do it

[lusty nimbus]

Are you guys still using the playwright mcp or the built in browser control plugin? Any use for playwright anymore?

[lost drum]

How much you earned?

[gentle harbor]

am i slow or when did chatgpt have a calculator ?

[kind jay]

I done another 20% today

[lost drum]

Do 10k? 🤣

[potent mason]

Can we create polls in this server?

[vital hill]

oof didn't know i had a reset

[lost stag]

no way I hit my limit already

[kind jay]

No need to talk numbers, but you’re not far off

[stable obsidian]

How do I use browser control? Cant find it

[kind jay]

Buy pro

[lost stag]

just tell it

[pliant rivet]

I don't know how to explain it, but theft is indeed very serious in China, I admit it

[boreal holly]

You talking about catchpoles? Like european medieval police or whatever?

[lost stag]

100 dollars for pro is just a lot

[frail coyote]

Can the Codex Diff window be made bigger? What is it, like thumbnail size now?

[kind jay]

It’s not

[gentle harbor]

my codex doesnt have 5.5 but my gpt has 5.5

[tiny fulcrum]

we're at 5.5 and it still can't get the line ending right in windows, rofl

[lost stag]

buy it for me since 100 isnt a lot

[frail coyote]

Depends on who you ask....

[lost drum]

Reset it I had to reset and install nre version

[torpid trout]

What I mean is. China itself locks out OpenAI's GPT, via the Great Firewall, apparently.
So... that is what I mean
Maybe it is not claude itself the issue?

[potent mason]

Anyways wondering what you guys think about the per chat terminal / browser / diff.
Vote 1 if you would prefer one per project.
Vote 2 if you like it the way it is.
Vote 3 if you would prefer one for the whole app.

[lost stag]

is there any free repos that are like codex

[kind jay]

Compare it to paying a SWE, it’s not even close, maybe not as good but saves you a lot of time

[frail coyote]

Diff? -10

[torpid trout]

yes, opencode

[kind jay]

If you’re wasting your time using it, then maybe it’s a lot

[dusk thorn]

forever

[lost stag]

that uses githubs co pilot

[pliant rivet]

No, CODEX can be used smoothly in China, but CLAUDE cannot be used, and any method will not work.

[gentle harbor]

well

[torpid trout]

it uses whatever you like. but gh copilot is not open.

[dusk thorn]

my claude sub was gonna renew later today, safe to say i cancelled it right now

upgraded to pro chatgpt 🔥

[kind jay]

New for me

[kind jay]

W

[lost stag]

its open just cant buy a plan. what other systems do what codex does?

[torpid trout]

That is interesting! I googled it and it said It cant due to the GFW
I guess it was wrong@!

[frail coyote]

How would the comparison for you be of buying 1000 Mansions or 10000000 Mansions?

[torpid trout]

as said - opencode.
if you mean the cli, anyway.
If you mean model, then no.

[boreal holly]

Yo, how is 5.5 soooo much better than 5.4 when it was practically already almost perfect?

[kind jay]

I’m not rich

[lost stag]

i mean the service

[torpid trout]

there are only models and interfaces.
Models > free > forget it.

[lost stag]

guess we gotta make our own with llama

[pliant rivet]

Can I ask you a more professional question, please, sir.

[frail coyote]

Right, so it depends on who you ask. Elon might not blink an eye.

[hard drum]

it is 5.5 time!

[dusk thorn]

pro saved my limits

[gentle harbor]

why does 5.5 speak so much faster then 5.4 on extra thinking mode ? is it not thinking correctly or something

[lunar crypt]

Lmao what's this insane color scheme

[torpid trout]

You can just ask the channel, I am not a professssssional

[dusk thorn]

orange insanity

[lunar crypt]

Ask me, at least I'll be able to use my PhD in AI for something lol

[lost drum]

YOO can someonevalidate if the outputs are worse if you compare the gpt 5.5 xhigh speed normal to speed fast? IS it jsut faster or its better?

[pliant rivet]

Okay, thank you, you are handsome

[cedar skiff]

what is the default context size?

[lost drum]

258k I think

[frosty zealot]

@boreal holly on this job where basically every distributor tube going into the Evap is leaking, he’s staying for overtime and I’m going off shift I’m just sitting here laughing like have fun with that one bud

[torpid trout]

thanks, you too 🤣
I like the gpt4 inspired curves

[boreal holly]

YOO there is literally zero quality difference between /fast and /sensible service tiers

[kind charm]

Is someone having problems with Browser Use? I'm getting failed to start codex app-server: The system cannot find the path specified. (os error 3)
I'm on Windows.
I’m trying to get Codex to navigate the web.

[lost drum]

YOOOOO thats huge thanks for reply

[boreal holly]

Sounds like one of them R22 conversion kit jobs

[pliant rivet]

Thank you. What I want to ask is, can a novice write a software using codex without being proficient in coding?

[frail coyote]

What's the best way to get a Diff view with the Windows Codex app?

[boreal holly]

Yes, now more than ever, as of today!

[cedar skiff]

Kind of. Anything large will become hard to work on and stall into small tasks becoming difficult.

[pliant rivet]

thank you for your answer sir

[lunar crypt]

I'd say yes, they can. Today GPT / Codex has gotten so good that you can wish software into reality as long as you an describe it properly enough. It doesn't replace expertise in more complex products and systems, and I can't guarantee it'll be bug free and won't require iteration, nor that it'll be secure.. but for a lot of use cases I'd say yes. But not professionally.

[torpid trout]

absolutely, yes but not safely
You need to know code to know if the garbled spaguetti is safe or not

[frail coyote]

Have you tried asking it to deveolop the structure fist and also ask it to make it easy for a human to understand with self-commenting code?

[pliant rivet]

thank you sir

[lean lark]

@pliant rivet Your success with creating software without programming knowledge depends on your skill with AI prompting. You're avoiding coding skill and have accepted prompting skill. You need to be explicit about what you want so that you get what you want. A vague request returns undesirable gaps in functionality. Learn what's happening and learn from your mistakes. And Have Fun!

[boreal holly]

git diff --stat

Or potentially C:\Program Files (x86)\Git for Windows\bin\git diff --stat

[frosty zealot]

It’s at a oil refinery so there’s lots of crap in the air that just eats the copper

[cedar skiff]

I'm a SWE, I'm speaking from a view of understanding the code it writes.

[pliant rivet]

thank you sir

[pliant rivet]

OK, thank you

[boreal holly]

Dang dude, upsell em on the electrostatic demister machines 🤌

[lunar crypt]

Gaahhh the AI market is so oversaturated, I went into AI out of passion, didn't expect AI to explode 😢 Not that I'm complaining but I feel obsolete lmao, OpenAI and Anthropic already full, people are making a billion GPT API wrappers and I'm just sitting here like... but I know how to train these models and deploy and get all the data and stuff... employ me pls lmao

(mini rant sorry)

[frail coyote]

But I wonder if by doing so, it would keep the structure coherent longer even with Straight Vibe coding, not sure, haven't tested it yet

[lean lark]

I went into AI out of passion, didn't expect AI to explode
Sounds like old ML days.

[frail coyote]

It thought training the models used Old ML?

[cedar skiff]

What youre saying is essentially a more complicated way of saying make it good to the model.
It doesnt address the problems.

[lean lark]

@lunar crypt you're describing something that I think most of us nerds here feel in some way. We have knowledge about how this stuff works, knowledge that users absolutely can't conceive. And what do we do? "fix the bug"
People need to go AFK and think about the Next thing, not just using the tools for silly stuff. Take the Next step and go beyond what the vibers and kiddies are doing.

[boreal holly]

5.5 is so insane, I give it a reference image for a node style and it even decides to synthesize the background to match the reference image lol my god, we are dealing with the best of the best

[lunar crypt]

Pretty much, but I have to still keep up. I feel like I spend 2h a day just reading papers right now and trying to build on top of that. It's borderline impossible for me as a researched, the billion companies churn new stuff too quick

[lean lark]

I was just talking about old manual Baysian approaches, like back in the 80's.

[lunar crypt]

Oh my god yes you're speaking from my soul here.

[frosty zealot]

Lol lying to suppliers and stiffing, what a shocker

[boreal holly]

Oh yeah you're gonna dig the Cap'n 💪

[lean lark]

dude, I Soooooooooooooooo feel that pain. That's the way it is now. Focus on what interests you and filter out the other stuff that the firehose shoots at you. Personally I ignore anything to do with grok, the FOSS from China, and even micro-updates from A\ and Google. There's only so much time. 🙁

[lunar crypt]

Needed to hear that, thanks. It's really true. I am doing some cool research that no one from the big players seems to be doing to the degree I am so that's something at least, I feel proud of at least the little slice of research I can do

[lost stag]

anyone here have the 100 dollar plan? have you compared it to the 20 dollar plan on codex? like how much more usage do you get

[stable obsidian]

This the dashboard 5.5 built me within 10mins with shadcn and three.js (the structure inside is 3D and animated)

[lost drum]

but how can it prove that it wont leak data?

I am not dev

so idk

but I heard lovable story

[stable obsidian]

Protect endpoints

[lean lark]

Just be careful about doing the obvious. Everyone has a chatbot and is working on memory, context optimization, etc. If it's obvious, next month you'll cry "I was doing that!!!" or "They stole my idea". Don't be that person. Move beyond that.

[rustic walrus]

a good prompter must always add "make no mistakes" fr

[lunar crypt]

It can't prove this, you simply assume it will, and make sure it can't

[lean lark]

Rather ... "check your work". There's a difference.

[lunar crypt]

Building really cool strictly typed data engineering and pipelining system with multi-vendor compilation backend for models in my own company, it's so cool so far, beat's all competition from internal benchmarks, makes me giddy. So I'm hoping I'm doing something unique

[lost drum]

I just want to know how to like have a solution that tests if my for example woocmomerce wordpress site would be secured

been using hsopify but too expensive and uncuztomizable

[lunar crypt]

Static analysis

[rustic walrus]

i trust it, it doesnt need self doubt, i dont wanna pay for its therapy

[lean lark]

I love to see excitement from a developer. It's not as common as it used to be. 🙁

( And I know all of you are just dying to know ... JaneBot is awesome and now getting some polish. )

[still trellis]

This is hilarious, does have a personality!
prompt:
create a stunning caas (cats as a service) landing page. Ill let you decide the verbiage, service, etc. use $imagegen to draft designs, images etc.

[lost drum]

bro the gpt 5.5 listens to prompts like I always wanted yupeee

[boreal holly]

Oh my gourd dude, 5.5 legit does not stop iterating on the design until it's done! 5.4, it would do a pass, look at it, and go "we're headed in the right direction but still some work to do". And BTW this is on a resumed thread that started as 5.4!

That's the Tau2 in action baybayyyy

[lost drum]

yeah but how do you backtest it on autopilot to not have any like u know problemos when custommers come and pay and ye

[obsidian lotus]

i expected 5.5 to be downgrade for some people, just like opus 4.7 lol

[still trellis]

pay with treats

[lunar crypt]

Oh btw would it be appropriate in any channel to post a link to my online 'test' for my PhD research?.. I need some human and/or AI data, hard to get word around. Unsure if even allowed anywhere. Anyone know?

[lost drum]

lol. I mean I want a solution that is fast and reliable to test ecom products fast the only part I have no clue how to seolve is implementing security or smthing

[boreal holly]

Probably #ai-discussions that's like the general purpose AI channel. Technically all the Claude lovers are supposed to go in there to gush over claude but in any case probably the right place

[still trellis]

ecom like Shopify?

[quasi summit]

anyone feeling the price difference yet on quota usage? they claim it's more efficient but I'm skeptical for code (where you need to read and write large amounts of code/docs). 2x pricier token use is scary

[lunar crypt]

Claude makes me feel icky, but I'll bite the bullet. Thanks!

[lost drum]

yes but use wordpress woocomere instead cause shopify just drains my wallet and cant cusotmze checkout unless you pay 2k per month for pro plan LOL

[lean lark]

Still using 5.4 but ... As a professional I still feel guilty about this:
"re-read to_do.md and process the first todo item"
When that's done, just hitting arrow up to recall last prompt, then press enter.
Sure, I commit between updates. Hardest work I'm doing on some of this is copy/pasting the name of the "ToDo" item into the commit message.

[lost drum]

• they take commision evvery order

[lunar crypt]

Have you tried Beads?.. Gamechanger for me

[lean lark]

?

[boreal holly]

It's OK, 100% of them come here to post their love letters to Opus instead of the correct channel so you'll hit the right audience 🤓

[still trellis]

Oof Worpress eeuw... LOL. Build something with astro and a cms. Quick and robust, 5.5 will eat it for lunch

[vital hill]

bro. images 2.0 does a better landing page design than codex app

[kind jay]

Looks nice

[lean lark]

Hey, I run about 20 WordPress servers in VPS and I'm quite happy with it.

[kind jay]

What value does 1000 mansions provide to you (apart from investment)

[vital hill]

don't worry about the haters, i also work with wordpress lol

[lost drum]

idk bro I am not advanced as that and I need like most reliable solution

I will find it maybe

[frosty zealot]

One for all my baby mamas

[lunar crypt]

It's basially a task management CLI that AI can use incl. dependencies, todos, related tasks, claiming, tracking what's in progress, etc. It makes "todo-style" handovers much more seamless with a well crafted AGENTS.md -- I've started using it recently, really powerful, GPT seems to be quite comfy using it

[kind jay]

Also choosing outliers to support your argument is such a low iq move

[potent mason]

I'm so glad they kind of fixed the front end

At least it's not making absolute trash anymore

[lean lark]

Ahh, TY - despite my claims of ease, I've also noted here that my common workflow is about 20 minutes of prompt prep, lettting the assistant chug, about 20 minutes of recap, checking, professionalism. The ToDo file is my bin of small things that don't need much handholding or skill and I bathe in the luxury of using it when I can.

[frosty zealot]

Inchresting instant is still on 5.3

[lusty nimbus]

how're you guys finding codex's visual design abilities compared to Opus 4.7? imo opus 4.7 still is way better

[kind jay]

Twins

[boreal holly]

Ah yep, I was wondering when one of y'all would join the party

[kind jay]

I have more battery than you

[frosty zealot]

How is it 22:25 are you on the moon

[lunar crypt]

Claude has been extensively policy-trained to "impress" rather than perform well. Yes it'll likely do better design and feel more capable and approachable.. on the surface.. until you realize a majority of its stylistic choices reduce to the same flat repeated design that you start to recognize over time

[kind jay]

Doxxed

[kind jay]

Nowhere 22:27

[kind jay]

Nigeria

[lunar crypt]

I think everyone has more battery than him and it's making me anxious

[frosty zealot]

I’m battery maxxing

[boreal holly]

Lol I got auto moded on French

[kind jay]

Vancouver

[lunar crypt]

ChArGe yOuR SmArT BrIcKs

[frosty zealot]

Lol almost, luckily it goes off my CELL CONNECTION

Almost doxxed myself

[boreal holly]

babooshka dadooshka

[kind jay]

Calgary, Alberta, Canada

[frosty zealot]

Closer but still no

You’re at least in the right province

[lean lark]

So... JaneBot is a CLI that is functionally equivalent to Codex CLI. She wraps codex but does a lot more. When provided with a directive like "create a utility" or "I need a project that" she creates a full project with full documentation, install script and info if required, configs for linting/formatting. The resulting projects can be run by a privileged user to do sudoer operations. In summary, she provides a simplified UI for codex as well as generates static utilites for system management, more complex deterministic processing, or even wrapping codex for dynamic AI processing. (AI generates the client)

Questions:

1. Have I just reinvented the wheel?
2. What else should she do?
3. Does anyone care that I call her "her"? 🙂

[kind jay]

St. Albert

[frosty zealot]

Such a specific and confident answer

[kind jay]

And correct

All of Canada is snowy, it doesn’t help

[lunar crypt]

In order to answer 1. from my perspective I'd ask in turn: "Can't I already achieve the same with a good AGENTS.md, developer instructions, and some skills?" -- I don't know the answer since I haven't seen your tool, but you should best be able to know

[frosty zealot]

Closer but still no 😂

[frosty zealot]

How’s the weather in Algeria anyways

[lean lark]

Yes you certainly can. But JaneBot does it all without the user knowing or doing anything like that. It's more of a "I just wanna thing that does 'this' kind of tool, not requiring any understandings. Or for developers and devops, it just takes care of the housekeeping so we can move forward.

[lunar crypt]

Then I'd say you made a nice convenience tool to lower the barrier of entry into AI use for developers who are new to it or don't know how to unlock it's true potential

For more advanced users, it may not provide as much benefit, but on a broader scope, I think it's quite useful

[lean lark]

Yeah, she's not intended for use by peeps with a clue - or not when we care about the details. She's just a convenience tool. ... and it's fun to talk about her like a person. LOLz

[maiden garnet]

Codex is getting wild, tbh

[kind jay]

Stop slop posting

That’s not the point of this channel

[maiden garnet]

Oh, well, I didn't know that

[kind jay]

It’s obvious

[maiden garnet]

How bout you go **** yourself, jane?

[kind jay]

Why don’t you talk with me instead?

[kind jay]

Go what myself?

[hard drum]

Go asterisk yourself LOL

No, wait, go EXPONENT yourself

[kind jay]

I thought I got told to k*ll myself, I was like woah

[maiden garnet]

Clearly can't write the full word without automod deleting it. Use your imagination

[lost drum]

🥹 5.5 xhigh is so good

[kind jay]

I did

[cinder plume]

Why dont you go (sic) yourself

[maiden garnet]

Good job

[hard drum]

nah, they just told you to Math.pow(a, b) yourself

[kind jay]

Matlab?

[hard drum]

JS

[cinder plume]

Is 5.5 global yet?

[lunar crypt]

Getting powed is nice and all but what if you're told to go matmul yourself

[cinder plume]

😭

[hard drum]

in some languages, ** is pow/exp

[lost drum]

Wdm global

[kind jay]

Indeed

[hard drum]

wait til they get you to matrix4x4 transmute

[lunar crypt]

GEMM 🥹

[hard drum]

or better yet, get matrix3 rotated

[cinder plume]

Its a rollout not everyone has(had) it yet

[kind jay]

I don’t remember this anymore, left my head

[dusk thorn]

GPT 5.5 AGI??

[hard drum]

Sorry 🙁

[lost drum]

Really?

[cinder plume]

Ooo I finally have it haha

[compact dawn]

5.5 is insane what

[kind jay]

Yo

[compact dawn]

it's leagues over 5.4

[tiny fulcrum]

the only thing it is way over is the pricing ;D

[lost drum]

200$ plan and forget

[turbid axle]

pricingGI

[kind jay]

Lowkey seems not annoying anymore

[turbid axle]

no chance

[lunar crypt]

Wow legit

[turbid axle]

I have that, and I eat through my tokens in days if I'm not careful

[kind jay]

Too much chronically online millennial though

[compact dawn]

tbh 5.5 extra high doesnt even run through usage that much

i was scared it would swallow it

[lost drum]

You havent specified why you want to go to car wash🤓

[compact dawn]

on $20 plan

[hard drum]

At this point, early Gen Z && Millenials should marry each other. They're becoming unified.

[tiny fulcrum]

we went from 20$ was enough to 100$ is enough and it will go to 200$ is needed pretty soon, I mean it's really amazing at what it can do, but it's also obvious where it is heading now

[hard drum]

Mill Z's

[kind jay]

The way you speak is funny

[turbid axle]

milli zanilli

[lost drum]

Idk I bought 200$ and happy best deal cause 2x more usage

[cyan wing]

Tibo pls we need a reset 🙏

[compact dawn]

we literally got reset 2 days ago hahaha

[hard drum]

Describe, if not minding. I prefer to know what people see in me.

[tiny fulcrum]

right, but API users care... a lot

[cyan wing]

moar

[compact dawn]

hhaah

[kind jay]

The && and ||

[turbid axle]

can I just start paying for my sub in blood please

[hard drum]

Ooooh, the C-isms.

I don't even use C

As my name states, I &mut btw

[lunar crypt]

Exactly which makes its answer all the more impressive. It gave an answer for both scenarios

[hard drum]

Soon enough ~ && ! will be used for logical disjunction

let there be horror instead of /\ && \/

[turbid axle]

when it replies 'bro you dumb' I will believe its agi

[lunar crypt]

You gave me immediate retrograde PT*D. I'm writing a new TLA+ parser in Rust. I forking can't with /\

[hard drum]

hehehe

[frosty zealot]

I’m a millennial

Is it noticeable

[frosty zealot]

I low key do that to my cat

[hard drum]

now wait till you get to =/= !o !in << >> (much greater-less || whatever order this is... i still struggle haha), \ for lambda, :. etc

[turbid axle]

I lowkey lowkey

[hard drum]

i will never not know how to properly recognise > from <

i don't know WHICH WAY IS WHICH

[frosty zealot]

Thor Thor

[turbid axle]

would loky thor thor fo sho fam

sry typo

[lunar crypt]

Already have, I hate it. Parsing infinitely nesting chains or logical conditions together with disjoint sets is ... uggghhhhh.. AST just can't

[boreal holly]

Cletus Cletus (one of the lesser known nordic gods)

[hard drum]

my language tries to avoid parsing issues by trying to be different

[kind jay]

That’s what someone else said too

[hard drum]

because of the whole LL(k) /LR(k) issue

[turbid axle]

thor has yet to find cletus

[hard drum]

if your parser has to lookahead more than once, you have failed

don't overcomplicate it

[lost drum]

I feel no premium now on no 1.5x speed

[turbid axle]

normal speed IS the premiun experience

nr 1 baby

who want 1.5, nobody wants to be nr 1.5

[boreal holly]

It's OK just get credits and go hard on 30 5.5 xhigh /fast --yolo

[hard drum]

somebody said "premium mediocrity" on YT

[turbid axle]

I am somewhat of an average premium myself

[rocky fog]

🤔

[turbid axle]

we have a rich one over there folks

[boreal holly]

Oh yeah I forgot you can burn more tokens doing 5.5 xhigh 1M /fast --yolo

[turbid axle]

tokenshaming, its all the rage

[lost drum]

Hmm

[boreal holly]

Yeah but it's a fraction of a fraction more accurate at >200% the cost and time. Ain't no shame in that!

[tiny fulcrum]

https://developers.openai.com/codex/config-reference

you can set it yourself in the config -> model_context_window
but it really gets worse with bigger windows, just marketing hype the 1M window, not very good in practice

[lost drum]

So better 258k?

[tiny fulcrum]

from my experience, yes it's better, can even set it lower, but depends on the work you do a bit

[turbid axle]

I dont believe in large context

the context is a lie

keep it lean mean average machine, premium

[kind jay]

Entropymaxxing

[turbid axle]

that what they call the singularity these days?

[frosty zealot]

https://youtu.be/Deu4zvcyJRs?si=F8CYT-ZD2xG3fxiU

[hard drum]

maybe

[kind jay]

When I’m having a terrible music taste competition and my opponent is @frosty zealot

[hard drum]

are you a bitter a-hole? no, you can't say you aren't when you might jsut be. that means you're getitng OLD! this means.... drum roll -- ONE OF US!

[rocky fog]

actually the vscode codex UI said Extra High, but the config had medium set
I had to set the UI to medium and back to Extra High (now they call it "intelligence" in the ui, heh)
well I switched from gpt-5.4 to 5.5 while it was already Very High and then also got a popup to switch to 5.5 so I just pressed to do it
maybe some of that messed that up

[kind jay]

Me?

[frosty zealot]

My music is an acquired taste for intellectuals

[hard drum]

Me, too? Welcome aboard.

[kind jay]

You’re just acoustic

[hard drum]

acoustic Pro Max 20x

[kind jay]

Acoustic on my pro until I max

[hard drum]

https://tenor.com/view/knee-slapper-laughing-laughing-and-crying-laughing-hysterically-gif-21595278

embed fail

[kind jay]

One day I will be banned from here

[hard drum]

i'm close enough to be banned haha

[boreal holly]

You done shenaned once, you will shenanagain!

[hard drum]

i will & your mut

[kind jay]

Everyday I wake up and thank god I can ragebait another day

[turbid axle]

what is life if not ragebait

[kind jay]

I know more physics than you lil bro

[turbid axle]

thats why I asked you big bro

[deft gyro]

xd

what

2x the price for 5.5

no thanks

[kind jay]

Probably not, entropy is like increasing always, before in the singularity surely it was at a minimum then

[frosty zealot]

Nerd alert

Wee woo wee woo

[kind jay]

More for me ig

[turbid axle]

yeah jeez

[deft gyro]

like vro

why is it 2x the price

aint no way its twice the size

[turbid axle]

maybe it uses less tokens?

it has premium tokens, only the best quality tokens

[frosty zealot]

People always cryin

[deft gyro]

im not crying

[kind jay]

It’s better with less tokens, so also faster as a consequence

[frosty zealot]

Like the great Britney Spears once said, you better work b**ch

[deft gyro]

I am just curious

also

[boreal holly]

hey guys 👉👈 is gpt better than opus?

[deft gyro]

the automod in this server is annoying

[turbid axle]

all I know it 2% of my week is gone since launch

[kind jay]

But you’re right, could have more internal loops or something

[frosty zealot]

It’s what filters people out

[deft gyro]

half of my messages get automodded

[frosty zealot]

You just learn to traverse the minefield

[deft gyro]

because a lot of my messages contain the letters t and f

[hard drum]

we trust-me-broing here maybe?

[deft gyro]

and w t f

[hard drum]

5.5 really grinding out the charts here

[deft gyro]

bro I hate the llm chartification

not everything needs to be a bulletpoint list or chart

[kind jay]

fat

[hard drum]

i asked for a table

i wanted table

easy for my eyes

[kind jay]

I want to get a similar word but I’m scared

[deft gyro]

oh ok

then that makes sense

llm following instruction is good

[hard drum]

if it goes against my needs, i coerce

[deft gyro]

I swear instruction following has been downhill since gpt 5.2

[lunar crypt]

Where Rust

[deft gyro]

I havent seriously used gpt 5.5 but it does fail the seahorse test

[hard drum]

gpt 5.2 does good, 5.4 bit lazy

[hard drum]

in codebase

[deft gyro]

Yes — the seahorse emoji is 🪸?

Actually, there is no dedicated seahorse emoji. The closest related ocean emojis are 🐠, 🐟, 🐡, 🪸, and 🦐.

[hard drum]

in there

[deft gyro]

• gpt 5.5

[hard drum]

mmaybe 50-55k LOC?

¯_(ツ)_/¯

[compact turtle]

Is 5.5 better

[lunar crypt]

Ooooo I see

Fire

[frosty zealot]

What word ?

[deft gyro]

why does 5.5 appear below 5.4 on the model picker

[kind jay]

Not trying it, it’s a word that you like

[deft gyro]

like

why

openai trying to follow a single convention challenge(IMPOSSIBLE)

[steel gale]

why codex all of a sudden is free wasn't it a chatgpt pro feature ?