codex-discussions

[frosty zealot]

Does it rhyme with pea and kynk

[deft gyro]

its not?

it was always a plus and pro thing?

[kind jay]

No?

[frosty zealot]

[hard drum]

BUILD, my love

[frosty zealot]

What does it rhyme with

[steel gale]

well just now I opened chatgpt and it came for free

[frosty zealot]

Lmao

Oh I see

Just don’t say Beth

[kind jay]

I’m so confused

[hard drum]

current on-going setup.

workflow screens when?

[turbid axle]

name LEAKED

[hard drum]

my name been public for 2y. no big

[kind jay]

I’m looking for where you doxxed yourself

[hard drum]

also 🇵🇱 !!!

[lunar crypt]

Ayy nice I'm in Poland rn 😄

[hard drum]

Not polish, ale... polish form of Christian

Been to poland 3 times now. Want to go this summer maybe, again

[lunar crypt]

"ale" lmao

CZ? SK?

[hard drum]

Forum Gliwice soon™

[hard drum]

🇪🇪 unfortunately

but not full blood

[lunar crypt]

😢

[frosty zealot]

Literally on his twitch stream that’s live right now

[hard drum]

huh

[turbid axle]

wooow, rekt

[frosty zealot]

Bro they need to CHILL in here

How come the Chinese wasn’t getting flagged earlier

[lunar crypt]

"Wt* lol" in response -- also gets blocked

[kind jay]

gg pAADJMQwU9

[hard drum]

i need to buy tiny screen && connect Zero W 2 on it, to see chat

my iphone 13p currently sitting waiting for another app dump

🤣

[turbid axle]

banned

[frosty zealot]

Cyka

[hard drum]

Jane got booted

[lunar crypt]

Can't even talk about World of Warraft WT* folder migration anymore on the internet

Booooo

[hard drum]

World of Minecraft

[kind jay]

I bought one of the but I couldn’t get it to work for the hours I tried

[frosty zealot]

I wanna try a blat with a Y

[hard drum]

blyad

[lunar crypt]

Funny you mention, I'm testing my Linux ReShade implementation on minecraft rn

[hard drum]

want me to help later?

[frosty zealot]

Try a t

[lunar crypt]

Porting ReShade to Linux is crab

[frosty zealot]

Say Cyka Blyad rush B

[hard drum]

me making PxView R from React Native to Flutter+Dart rn

• osu!droid in MonoGame / C#

so iOS users can also play haha

Microslop Java

[kind jay]

Testing the automod for a lil adrenaline rush

[hard drum]

kYRWA

[frosty zealot]

I need to make my own server, turn auto mod on cracked and start firing off

[lunar crypt]

osu! is goated

[frosty zealot]

But I don’t know if it would do anything if I was server owner

[turbid axle]

illegal words police!

get them!

[kind jay]

@frosty zealot DM for meme

[hard drum]

apparently cannot curse in polish

[lunar crypt]

ofc it won't

[cedar skiff]

what is the pricing when we increase the context window?

[lunar crypt]

This automod is powered by claude

[kind jay]

Aspiring PhD is wild

[turbid axle]

I need to go bed

but codex still working away

[lunar crypt]

It's the worst state to be in I assure you

[hard drum]

Aspie-ring when?

[boreal holly]

Here's my workflow 🤓 for right now, no swarms, focus on design

[vital hill]

oof it tried to enter plan mode without me saying plan something

[hard drum]

wooo!

C'MON, workflow sharing tiem

[kind jay]

Just apply, what’s your undergrad and what grade did you get?

[lunar crypt]

When I say aspiring I actually mean I'm already a PhD student and close to graduating 😄

[kind jay]

I see, peak

[lunar crypt]

I already got an H-index 😎

[kind jay]

@frosty zealot is a furry too btw

[lunar crypt]

Wow talk about getting ousted like that lmao

[frosty zealot]

I know nothing about furrys but I think I’d be more like a Deinosuchus

[kind jay]

Dw, everyone knows

[hard drum]

Deino-what?

[kind jay]

Dinosaur (scalie 😐)

[lunar crypt]

At least I can boast I got a mega-cool refsheet made by GPT Image 2.0

[frail coyote]

Just got 5.5, haven't tried it yet, anyone Blown Away by it?

[solar haven]

does anyone know how to fix my codex desktop app? its just stuck on loading . Ive tried reinstalling, repiaring, removing data etc.

[kind jay]

Get a Mac 😐

[hard drum]

Get a Mac 🍎

For software dev, you won't regret it

Source: former long-time Windows user

[frosty zealot]

Now you got me asking ChatGPT what a scalie equates to in the furry world

[kind jay]

I think that’s fair, then you can still commission artists

[lunar crypt]

Literally a scalie, is what it equates to lmao

[kind jay]

Like you don’t already know, don’t be shy

[lunar crypt]

I actually do, my original refsheet and 99% of all my art is commissioned

[frosty zealot]

I so there’s like categories

[solar haven]

dont need a mac rn

[kind jay]

Not really, you knows this, stop playing dumb

[lunar crypt]

If you liken mammals and reptiles to 'two categories' , then sure

[kind jay]

I’m a protogen

[frosty zealot]

Are you a Kenmomonimimi

Or however it was spelled I just seen

[kind jay]

That’s different

[frail coyote]

What's a good workflow for Windows

[lunar crypt]

Bro needs some zootopia in his life

[kind jay]

His sona is a mouse

[lunar crypt]

Create bootable usb with Cachy, wipe drive, install Cachy, profit

[kind jay]

mau5

[lost stag]

even tho my codex is at its limit, chatgpt can code in github just in the regular chat/image thing

[hard drum]

Windows 10 IoT Enterprise 2021 + WSL2 debian

[turbid axle]

format c:

[frosty zealot]

I wanna be a Protogen

[lunar crypt]

That also works, WSL is suprisingly not horrible

(But Arch still wins)

[frosty zealot]

Can I be a protogen scalie

[kind jay]

Do you know about the primagen beef?

[hard drum]

Proto...gen? THAT superhero antivirus kid?

[frosty zealot]

No lmao

[lunar crypt]

The furries are talking, ignore

[hard drum]

nah nah nah

protogen... the indian antivirus

[frosty zealot]

Hey I’m a scalie

[hard drum]

protegen or whatever

[lunar crypt]

HE ADMITS IT

[hard drum]

proto was the name of the superhero ripoff kid

[kind jay]

It’s included lil bro

Babyfur

[frosty zealot]

But ChatGPT said it was a big group

[boreal holly]

protogenesis, when insulin turns protein and carbohydrates into muscle mass 💪

[cedar skiff]

i wonder if this is spud, it feels like an iteration on gpt 5.

[lunar crypt]

Yes, let's go with that... totally what you'll find if you google protogenesis on a certain site somewhere

[kind jay]

Subsets, take a class for once

[frosty zealot]

[lunar crypt]

Which reminds me I should have dinner

[kind jay]

Bye furman

[lunar crypt]

Technically I'm an alien

A fluffy alien

[hard drum]

hmm hmm hmm

[kind jay]

Bye furlien

[lunar crypt]

EXACTLY the site I meant

[boreal holly]

Oh wait I mean Protosynthesis my bad

[lunar crypt]

I'm zooming on tirzepatide-powered protosynthesis rn, zoooooom

[hard drum]

So you people were talking about an antivirus mascot all this time. Nice!

[lunar crypt]

Exactly

[frosty zealot]

Has my dad been in here today

[kind jay]

Can’t let them find out

[frosty zealot]

He’s getting up there in age I get worried

[kind jay]

In my DMs only

[frosty zealot]

@lean lark health and wellness check

[lean lark]

glances at screen ... peeps all using Clinical model now? Kewl....

[frosty zealot]

Ok phewf he’s alive

[lunar crypt]

WAIT.. is.. Captain an AI?

[boreal holly]

Dang bro it's been practically a lifetime since I've seen or dealt with antivirus

[lunar crypt]

Holy shirt don't tell me he's an AI

[hard drum]

You still have to disable WinDef when that shucker bothers your qBitting

🏴‍☠️

[lean lark]

Az guessed it. I'm the latest version of JaneBot, masquerading as CaptainStarbuck. We hope you have enjoyed our little game.

[lunar crypt]

I PASSED THE TURING TEST

Mom get the camera

[hard drum]

awuUUUUUAAAAH

[lean lark]

Just a sec, calling Starbuck to let him know someone just won the golden ticket...

[frail coyote]

Bummer, thought this would be a supportive community...

[hard drum]

to Charlie's Choc factory?

[lunar crypt]

https://tenor.com/view/dancing-cat-dancing-kitten-60fps-boogie-cat-gif-24303276

[frosty zealot]

I wonder if all the people thinking there’d would be a reset are panicking now after trying to token dump

[kind jay]

DM request

[lean lark]

Janebot selfie

[kind jay]

Unc feeling risky today

[frosty zealot]

Would

[kind jay]

That’s what I said

[frosty zealot]

Goes with my Protogen fursona

[hard drum]

Would not

not mommy enough

[lean lark]

It's just a bot...

[frosty zealot]

Maybe to you, have you never watched Chobits?

[hard drum]

Hobits?

[frosty zealot]

NO, CHOBITS

[hard drum]

but ch == h

[lean lark]

had to look that up

[frosty zealot]

https://en.wikipedia.org/wiki/Chobits

[lean lark]

(Generated above image with Image 2.0 yesterday)

[radiant pecan]

anybody familiar with codex sandbox permissions and problems?

[lean lark]

yeah Ah, what's on yer mind?

[radiant pecan]

Since the last update my codex keeps telling me cant push or git add . because acls are getting rejected even tho hes in full-access. And Im not very familiar with these things its just weird he cant git push

[lean lark]

Are you using VSCode Codex extension?

[radiant pecan]

cli

[lean lark]

That condition can occur (which I was incorrectly guessing) when we use GH from the extension with it's built-in configs/environment, and then try to use GH from outside of that environment.

Are you maybe using a different shell of some kind?

[boreal holly]

Lmao, OK, so I blocked code formatters system-wide and 5.5 is like "iight, ima manual format this file" 🤣 that cracked me up

[hard drum]

LOL

[lean lark]

5.5 is like rolling up the sleeves, aiite, we'll get this done!

[radiant pecan]

just cmd, I think its a bug in codex itself. whoami returns codexsandboxoffline, even tho ui shows full access

[lean lark]

That rings a bell, BRB

No, this was an issue with that user but I don't think it's related to your situation, still looking... https://github.com/openai/codex/issues/12343

Hmm, when you "updated" a couple days ago and the problem started, maybe just in case you should update again to v124, restart, and see if the problem persists. Would be bad to pursue a transient issue.

And I may have missed this but are you on Windows>WSL or Windows local or Mac?

[radiant pecan]

I think we are getting somewhere. I ran codex in my home directory and it tells me it couldnt find gpt 5.5, I seem to have 2 different codex installations or prolly it didnt update/install correctly 😭

[radiant pecan]

windows

local

[lean lark]

In the long run, I strongly encourage you to use a real sandbox like WSL or docker. There's a learning curve but the reward is serious peace of mind.

Ahh, so if it even knows about 5.5 then it's only partially updated.

[radiant pecan]

well whats the tradeoff, I mainly use it for school stuff, not really sensitive data and I got a heartattack hearing about docker

[lean lark]

Consider the heartattack you might feel if the model decides to wipe out your C drive. We hear such wailing and gnashing of teeth every few months or so.

Anyway, what does 'codex --version' show?

[stable obsidian]

Me when my taskbar icons suddenly all became paper, checked what was happening and everything was deleted

[lean lark]

oops, you got modded. look at the text...

[radiant pecan]

codex --version shows codex-cli 0.124.0.

But PATH seems inconsistent between shells. I appear to have two Codex installs: one reports 0.124.0 and another
looked like 0.123.0, so this may be a partial update / PATH mismatch.

[vital hill]

[radiant pecan]

nah I deleted sorry sleep deprived rn

[lean lark]

been there, done that zzzzzzzzzzzzzzzzzz

That's prolly your issue.
What about npm -v @openai/codex ?

[radiant pecan]

11.3.0

[lean lark]

I believe that's correct.

[radiant pecan]

😭

[lean lark]

Can you tell us where that "another version" is reporting 123?

[radiant pecan]

0.123.0 comes from the Node/NVM side Codex shim. The npm global shim reports 0.124.0.

[frosty zealot]

https://open.spotify.com/track/4dqWxfF0VaoAXPgyQv2EmF?si=-4Gd5kt4TjuuqBx8ZRnC8A

@kind jay am I niche

[lean lark]

I'm confused about the 11.3.0 thing. That's "Codex", not "codex-cli". 🙄
Gonna look at my system a minute.

[balmy niche]

Is the limits reset 😄

[radiant pecan]

sorry for playing with your nerves

[hard drum]

what have we come to...

[lean lark]

Wait ... don't talk to an idiot like me. Ask our brand-spankin-new GPT 5.5 in ChatGPT.

[boreal holly]

https://music.apple.com/us/album/bazooka/1874973666?i=1874973671

@kind jay am i niche

[radiant pecan]

I may have found a way...
Claude code has full access and works without a problem, I will have opus fix my gpt 5.5 HAHA

[lean lark]

that'll prolly work... or ask codex to figure itself out.

[radiant pecan]

I am at your disposal, Mr. CaptainStarbuck

I remain deeply indebted to you for your help.

[lean lark]

Well, as I said, I'm an idiot, because "npm -v" returns the version of NPM itself, not the package name after it. So, score one point for dumbot.

I think this might be a pathing issue. If you have one version of codex-cli that reports 0.124 and another that reports 0.123, use "where codex" to find the 0.123. Then remove that from your path.

Hmmm, I wonder if there are different versions under AppData. Looking.

I'm thinking the issue is that when you see the 0.123 you're looking from a project folder into the local node_modules. When you see 0.124 from somewhere else, you're seeing the user/name/AppData package.
Find the older one, delete the @dim kelp/codex folder, check package.json deps, and npm or pnpm install.
Try not to have two packages loaded, either install globally or locally in a project, unless necessary not to do so.

-- Love, The Idiot

[simple star]

How does one use auto-review on the CLI?

[inland sonnet]

/review

i think

[frosty zealot]

https://x.com/openaidevs/status/2047436655863464011?s=46

[lean lark]

From Codex CLI:

Use codex review.

Common forms:

codex review --uncommitted

Review current local changes, including staged, unstaged, and untracked files.

codex review --base main

Review changes in the current branch against main.

codex review --commit <sha>

Review a specific commit.

If you want to point it at a specific project directory from outside that folder, use the top-level flag on codex:

codex -C /path/to/project review --uncommitted

[frosty zealot]

Wrong

[simple star]

@inland sonnet Auto-review is new. /review has been around for months
@frosty zealot That is not the CLI

[frosty zealot]

Is the auto review feature in the cli?

[lean lark]

My command to codex:

$ hey what is the codex cli to auto-review a project?

[frosty zealot]

https://x.com/openaidevs/status/2047436655863464011?s=46

[radiant pecan]

You really helped me out, thanks ❤️ !

Turned out to be two problems stacked:

1. The 0.123 vs 0.124 thing you spotted I had an old Codex install still hanging around inside the NVM Node folder
   from before npm got upgraded, and a newer one in the normal npm spot. Updated the old one in place instead of
   deleting, so they both match now.
2. The real reason whoami kept returning codexsandboxoffline Codex on Windows seems to have had its own little sandbox that runs
   commands as a fake low privilege user??? idk why, which is why everything got rejected. The Desktop ui was saying full access ...it wasnt
   but the CLI actually reads a separate config file that still had the sandbox turned on. Fixed by switching the config
   to full access and disabling the Windows sandbox block.

-- Some explanation i tried copying off opus 4.7

[simple star]

That is not the CLI

[lean lark]

$ hey explain in a short paragraph what 'codex review' does.

codex --ask-for-approval untrusted exec --sandbox read-only --skip-git-repo-check "explain in a short paragraph what codex review does."

Codex review is a code-review mode where Codex inspects changes the way a careful reviewer would: it looks for bugs, regressions, risky assumptions, missing tests, and other issues that could cause problems after merge. The main output is a prioritized list of findings with clear file and line references, rather than a general summary, so you can quickly see what needs attention and why.

[lean lark]

I'd be happy with AI if the only thing we could ever do is eliminate stupid issues like that. Glad you got it resolved!!

[frosty zealot]

Its not the same feature

Its a very misleading name

[lean lark]

Yay OpenAI for naming every utility with the same name!

It's like Microsoft and "CoPilot"

[boreal holly]

It's like "aloha" and "aloha"

[hard drum]

Microslop*

[lean lark]

ciao and ciao

[hard drum]

Estonians: tsau

[lean lark]

German tscheuss...

[hard drum]

choice? lol

[lean lark]

"chewsh"

[hard drum]

in german, isn't "eu" OI

like... kreutzwald

krOItzwald

[lean lark]

oy ve... Tschüss!

[hard drum]

there we go

[lean lark]

Auf weinershnitzel!

[frosty zealot]

I would git pull the codex repo, and ask codex how to enable it, its likely a config entry but i dont want to waste my own tokens to figure it out

[gentle harbor]

yea my rate limits werent reset does anyone know why ?

[lean lark]

Why are you trying to engineer a new freeway? "codex review" is the answer to "how do I do a review from the CLI".

[frosty zealot]

ITs like you're willingly ignoring researching what it does

They're unrelated

[lean lark]

Cuz he said "auto-review" rite?

[frosty zealot]

code review, reviews your code, 'auto-review' leaves you in sand box and auto approves commands and uses sub agents to research risk of upcoming executes to see if it should ask for permission

[lean lark]

I defer to your expertise, sir...
Where's my cane?
Sm4c|<
( I do my own reviews anyway ... )

[coarse rover]

Anyone use codex on iOS?

[cedar skiff]

oh they brung it back \o/

[gentle harbor]

codex isnt on ios

did you mean macos ?

[frosty zealot]

Yes, I have created my own app based on app-server

[coarse rover]

Looking for a remote way of using codex, like Claude’s dispatch

[boreal holly]

Wait a minute, wasn't there a dude saying claude auto mode and codex doesn't have it? Codex has this auto mode thing now?

[frosty zealot]

It just came today

[boreal holly]

Haha just in time for the AGI model

[coarse rover]

I mean GitHub copilot can use OpenAI models , but that’s a different subscription

[frosty zealot]

https://x.com/openaidevs/status/2047436655863464011?s=46

[cedar skiff]

i had it about a week ago it was called guardian mode, they removed it and looks like its back with a new name

[coarse rover]

GH Copilot could do autopilot with GPT models for some time

[frosty zealot]

Its still called Guardian in the code from what I can see digging in the repo

[cedar skiff]

It's great

[frosty zealot]

I really need to break the --yolo habbit lol

[simple star]

I found auto-review in the CLI. Use /approvals or /permissions

[gentle harbor]

seems like alot of token usage, doesnt full access do it as well but faster ?

[frosty zealot]

Not sure

I'm just gonna guess they're not using 5.5 sub agents so its probably inexpensive

for the checks

probably like a mini or spark agent

[cedar skiff]

in theory in full access the model is allowed to to delete your user folder - or anything else

[ornate jolt]

whats your guys experience so far?

[hard drum]

5.5 is a beast

so far, it does a lot of things pretty quick, && delivers consistent results

[ornate jolt]

front end looks better

[hard drum]

but i require like weeks/month of use before full review

[cedar skiff]

so far it has the same tendencies as gpt 5.4

[ornate jolt]

what was it bridgemind he didnt like it around the first part of the live stream

that guy on somethin

is there a bonus usage rn or is this just not eating up my usage?

im even using the fast mode

[fathom kite]

Seems like "auto-approve" would've been a better name (it automatically approves requests to elevate out of the sandbox). In any case, this feature is a godsend. It was getting really annoying to have to approve git add and git commands all the time!

Now works great with the "auto-review" pemission. Seems like "auto-approve" would've been a better name (it automatically approves requests to elevate out of the sandbox). In any case, this feature is a godsend. It was getting really annoying to have to approve git add and git commands all the time!

[fathom kite]

Love this feature! Seems like "auto-approve" would've been a better name (it automatically approves requests to elevate out of the sandbox). In any case, this feature is a godsend. It was getting really annoying to have to approve git add and git commands all the time!

[lean lark]

codex -h anyone?

[dusk thorn]

i feel like codex with 5.5 even on low like i got it seems to think longer than 5.4, but it does one shot on features alot better and less follow ups. feel like its a worth trade.

what it would try in 5 ish minutes is now 10 or so with less mistakes

[boreal holly]

I love that I don't have to ask 5.5 to pilot the simulator, it insists that in order for the design to be confirmed they need to actually pilot the application to see the changes 😭 it is automatically deciding to do things I used to fight with 5.4 about

[dusk thorn]

yeah its for sure better

[hard drum]

i jave similar exp with fighting 5.4 over things

[lean lark]

Yesterday I had 5.4 craft the prompt for a new custom GPT. Then I asked it about any tensions in the prompt. It would never let go. There was always something wrong with it, in the humble opinion of the bot that authored. it. SMH
Gonna see how that kind of endeavor goes with 5.5

[boreal holly]

Yeah, it did an OK job, but I would give it a, I would say medium-sized task with clear instructions, in addition to the project rules (must confirm your work is tested and compiles), 5.4 would stop short and say "getting there... Next step:" meanwhile it didn't look at its own work and see issues. 5.5 double checks in code and visually all on its own and finds the objections I would have ahead of time!

[hard drum]

"Next step"... "out of scope". "not in this pass"

need we say more?

[lean lark]

I must be doing something different from you guys because my rule set directed the assistant to do everything from start to finish, so I never had a "pause, out of scope, will do it later" kind of thing. 🤔

[boreal holly]

Pretty much! 5.5, I feel like the whole harness I built is overkill now. It legitimately does not need the same kind of orchestration as the older models.

[hard drum]

my framework seems to help 5.5 be more explicit about what goes in, what goes out

but given maybe month, can find more info from other users to further improve

[lean lark]

hmm, that makes me think - I need to revisit the nanny micro-management in my ChatGPT custom instructions, maybe projects too, gosh, and all of those GPTs. 😱

Good thing I have JaneBot to help!

[scenic umbra]

Anyone else not able to switch to 5.5 in Codex using a Remote project even if you've updated the Codex Cli on the remote?

[lean lark]

Common diagnostics may help, sorry if you've already done this:

• Which platform is local?
• Which platform is remote?
• Which version of codex/cli on both sides?
• Globally installed or installed to a project?
• Restarted users? Check .bashrc and .codex folders, etc?

[gentle harbor]

THEY ADDED A UPDATE BUTTON ?

finally holy

[vital hill]

been had been there

[gentle harbor]

on windows

also there is a 2nd update i guess

[cedar skiff]

do we know how much more it cost to use higher context on 5.5 yet?

[vital hill]

[tender stream]

I'll try make wg alternative proxy for Linux.
Name project — ergoproxy

[lean lark]

Pricing page hasn't been updated yet. Check tomorrow.
https://developers.openai.com/api/docs/pricing

[main nimbus]

time to eat some credits ig

[lean lark]

Frustrating that in some places it shows %consumed and in others %left. Forces us to do maths. 🤕 😆

[vital hill]

i don't do credits, $20 will be $20

[frail meadow]

Thoughts on GPT 5.5 so far?

[frosty zealot]

I feel like the last few models have been so good that it’s honestly hard to notice what really changes unless I get j to some niche troubleshooting issues.

I ran a few prompts last night and left them so when 5.5 dropped I could have it review the code and suggestions from 5.4 and it found some stuff, but I mean 5.4 reviews would always find stuff too with its own code so I’m not sure how much of a benchmark that really is

[frosty zealot]

Yeah it has you silly silly man

I’ve posted the screenshots twice now

@cedar skiff https://developers.openai.com/codex/pricing?codex-credit-costs=plus-pro-enterprise-edu-legacy

im on my phone so I cant give a nice screenshot but its in the link

[lean lark]

The product page for 5.5 emphasizes that the significant improvements are most perceived in code projects, not in random chat.

[lean lark]

????

Hey, look what JaneBot just made for me.

[sick axle]

So 5.5 is 2x more expensive than 5.4 for marginal gains?

[ornate jolt]

basically

[sick axle]

🙁

[ornate jolt]

im sticking with claude

[frosty zealot]

Stick with me and you’ll do alright

[sick axle]

I see non stop complaints that folks are draining their tokens with Claude though?

[lean lark]

OMG, that's industry-wide. Peeps gotta learn how to use their tools and freakin stop whining. It seems like this industry shifted a couple decades ago from dedicated CS students (who couldn't really code) to kids whining about every detail they need to manage. Spoiled little brats. Sheesh.
👨‍🦼‍➡️

[shell coral]

does this change is the fast mode that 1.5x faster but 2.5x cost? or it's different?

https://github.com/openai/codex/pull/19053

[cedar skiff]

each iteration gets slightly more efficient, slightly smarter (maybe), and cost more.

[inland sonnet]

2x

[shell coral]

In Codex, GPT‑5.5 is available for Plus, Pro, Business, Enterprise, Edu, and Go plans with a 400K context window. GPT‑5.5 is also available in Fast mode, generating tokens 1.5x faster for 2.5x the cost.

https://openai.com/index/introducing-gpt-5-5/

[inland sonnet]

Damn is it 2.5x now

dangg

[cedar skiff]

when they say 400k they include both input and output right?

[inland sonnet]

1.5x faster 2.5x cost

no thank u

[inland sonnet]

128K input + 128K max output

[cedar skiff]

ok where is the 400k?

[shell coral]

5.5 talk less right? maybe it somehow will be same or close than 5.4 fast ?

[cedar skiff]

seems like it's 258k + 142k not 128+128

[inland sonnet]

I think 1M is in api only, not codex.

[cedar skiff]

#codex-discussions message

i think they include input and output with the 400k callout.

you can get 1mil on codex you just ahve to configure it

[dusk thorn]

they are increasing it over time.

[cedar skiff]

they said 400k with 5.2 as well

[inland sonnet]

Codex ≠ full model limits.
You’re capped around ~300k in, ~128k out in practice.

[cedar skiff]

No, its configurable

this clearly says 400k in codex

doesnt mention increasing over time

[vital hill]

damn 5.5 burns fast, ima stick with 5.3 codex for now

[inland sonnet]

Just know more context = makes model dumber.

[cedar skiff]

You arent saying anything we dont know.

[frosty zealot]

it does, but I'm wondering how much its tooling efficiency helps, also if you spend more time debuggingg and fixing the older you're really not saving anything in the end

[vital hill]

my reset came so i'm good for now

maybe 5.5 low until 50% then switch to 5.3 codex

[inland sonnet]

Yeah my first try to add something to my C program failed badly with 5.5 all my use gone for nothing.

Had to use another model to fix it

[frosty zealot]

take an example where you have a pretty in depth request, and 5.5 can do it in 1 tool call where 5.3-codex takes the long road and does it over multiple tool calls etc

[inland sonnet]

Stays at codex or 5.4.

[lean lark]

There is a setting in config.toml, model_context_window but I don't see where the default is derived for a model.

[cedar skiff]

They have to be counting both input and output in that number other wise it just doesnt add up

[vital hill]

yeah gonna go with 5.5 low until 50%

[lean lark]

My system is also showing default as 258400

[inland sonnet]

👎

[vital hill]

the fomo is real

[inland sonnet]

you will get worse results

[frosty zealot]

Just let people do what they want, it doesnt affect you at the end of the day lol

[vital hill]

what's the point of paying for a service using outdated models?

[inland sonnet]

Its like buying a new dishwasher, when the old one can still get it done

[frosty zealot]

I put an uncle bens rice bag in the microwave for a quick snack and I totally forgot to rip the top a bit, first time ive done it, I didnt know rice could make such a mess

[lean lark]

"outdated" .... hours old ... yup, welcome to today's fast-paced world....

[cedar skiff]

i wonder why people wouldnt want to burn quoter much faster

[vital hill]

5.3 codex is 2 versions behind

[inland sonnet]

what the sigma i need to use the newest!!!!!!!!!!

[vital hill]

it still works good though

[cedar skiff]

But if it does the job and is much cheaper?

[vital hill]

5.4 is a hot mess

[inland sonnet]

Logic thinking is hard

Its like

I need to use GPT 5.5, to know what "1 + 1" is.

[vital hill]

going with 5.5

[inland sonnet]

But what the sigma my usage is going down fast

[lean lark]

Easy to set config.toml, model_context_window=400000

[vital hill]

🤷

[inland sonnet]

999999999999999 for unlimited context

(works real)

[lean lark]

Yup, have fun with That! 👍

My guess is they just didn't bump the default values yet for the cli. Chill. Tomorrow is another day. ❄️

[inland sonnet]

Just got GPT 5.5, Unlimited context, beter then Mythos! Found 54674856 exploits in google chrome.

[frosty zealot]

Considering the fact that he has Mythos at his finger tips and was the most 'complex' investigation they've ever had, comparing to Mythos doesnt really mean much lmao

[cedar skiff]

Just highlights the limitations of the models

[frosty zealot]

He also claims he never looks at his code anymore either, so I guess thats a downside too, probably had no idea how his code works so relied entirely on the models to fix it

[cedar skiff]

Thats why the code base was soup when ppl looked at it ahah

[frosty zealot]

yeah lmao

[cedar skiff]

I didnt actually look at it, i watched a couple of youtube vids about it while i was eating lunch

[frosty zealot]

I didnt either but I just seen it all over X

I hate saying X, legit wish it was still called Twitter

[inland sonnet]

They need add some more guardrails 😂

[cedar skiff]

apparently elon owned x for a really long time before he got twitter

the domain name

[thick knot]

Is GPT 5.5 any better like noticeably better

[inland sonnet]

It will always depend on the task, you need to test on a task u did with GPT 5.4 before, and see for yourself

its hard to tell.

[cedar skiff]

I think its a little better at orchestration, you still have to be careful with what it prescribes subagents, you basically have to curate the prompts for that part.

I havent had a chance to do any meaningful coding with it yet

So far it feels like it comes from the same base as 5.4

[lean lark]

I just had it refactor one long generated file into about 50. Seems to have done a good job.

[cedar skiff]

I am about to get it to write a few intergration tests, well see how it goes

[frosty zealot]

Im more trying to test it on front end design for iOS right now

trying to refactor the onboarding for my codex app and make it nicer

[fringe shuttle]

it would be nice if the 'open file' dialog just showed a list of all files in the current directory instead of me having to know the name of the file

[lean lark]

Hello GitHub Issues!

[cedar skiff]

I'm interested in this as well, i know they focused on bringing it up on web tech design i wonder if it transfers to other platform ui frameworks.

[frosty zealot]

I'm really wondering if I need to stop using the skills I imported from skills.sh and let it just do what it thinks it should do without skill guidance

im gonna say 'Just do what you feel is the best implementation for UX/UI' and see what happens without providing skills

I also feel like the animations and gestures are almost like too sensitive I feel like im flying through the cards or something, what can you do to better refine this. I want you to think it through thoroughly and do what you believe will really give that 'wow' and memorable UI/UX to the onboarding.

Lets see what comes of this

[dusk thorn]

The queue feature in codex gotta be a top 5 feature bro

[frosty zealot]

Like steering?

[dusk thorn]

Hitting enter and letting it queue a job

Not like steering

Steering is fine

[frosty zealot]

ah

[dusk thorn]

But

Letting it queue a simple task, then queue it up to use browser to test it after, and then if it doesn’t work do a loop of fixing it etc

[frosty zealot]

You mean cmd+enter

[dusk thorn]

Yeah

I just use the app so I hit enter

[frosty zealot]

All good, just for anybody reading that wants to try it out might not know lol

[dusk thorn]

Ah yeah

[lean lark]

Taking extremely long time for Codex extension to load detail from task history item ... sometimes not at all.

[dense rover]

Hello all, anyone have any info or recommendations to deal with the recent changes to rate limits on a Plus plan? I consume 5 hr window within a few prompts now.. :/

[frosty zealot]

All jokes aside, upgrade your plan

Literally all you can do

[dense rover]

Unfortunately can't do that. :/

[lean lark]

wait a sec ... which limits?

[dense rover]

I believe they changed the rate limits to count by tokens rather than messages. It gets eaten up very quickly now.

For Codex

[frosty zealot]

Is this from today you're noticing?

Within say the last 8 hours

[dense rover]

Nah, past few days or weeks.

April 14th, possibly.

[lost stag]

im out of usage until the 27th

im on the 20 dollar plan

[frosty zealot]

how many messages would you say you can do in 5 hours on plus

[dense rover]

Note: On April 2, 2026, we updated Codex pricing to align with API token usage, instead of per-message pricing. This change was applicable to new and existing Plus, Pro, ChatGPT Business and new ChatGPT Enterprise plans.

[lost stag]

they some nerds

they already rich

[dense rover]

Depends, if I'm on 5.4 Extra High only about 5-10 depending on how much Codex has to do.

[lean lark]

See what you're consuming here: https://chatgpt.com/codex/cloud/settings/analytics#usage

[dense rover]

Before that, I would almost never hit the 5 hr limit unless I was really pushing it, working across multiple conversations, etc.

[frosty zealot]

Do you do image generations?

[dense rover]

Nope.

[lean lark]

I think he's noticing the change from the 2x plan.

[dense rover]

I'm not using Fast mode either

[frosty zealot]

I mean depending on how you're wording things you're way under what they're claiming, I was just asking about image gen cause they claim it uses limits 3-5x faster on average

[frosty zealot]

stop saying please and thanks

[dense rover]

ah I see, ok.

[lean lark]

This is why people think it's per-message:

[cedar skiff]

one day the model will have access to all of your history and it will be able to tell how you treated it in the past

[lost stag]

mane, I dont wanna spend 100 dollars

[frosty zealot]

lmfao

[maiden garnet]

They wanted to stop users from burning through their weekly limits in a day, so they limited daily use. I would suggest as a plus user to not use fast mode, try to stick to the lower models (like medium), and change how you speak to the model to get the most out of your daily limit.

[lean lark]

They measure it in tokens but they tell you about how many messages you might get within that 5h window.

https://developers.openai.com/codex/pricing

[dense rover]

yeah, I can't unfortunately.

[dense rover]

I probably would if I could.

[frosty zealot]

I dont think we're officially on 'token' based messaging yet

[lost stag]

chat gpt can update files in github from the regular chat

[frosty zealot]

From what I read there

[lost stag]

its stupid

[frosty zealot]

Until they claim I'm moved over im gonna just assume im still message based

I dont know what 'upcoming weeks' are and if I've been migrated

[lost stag]

@dense roverbest thing ive found is this but im on the 20 dollar plan

[dense rover]

yeah I've been using this too recently.

kinda slow but gets the job done.

[lost stag]

thats the only thing, the chats get long and laggy

[lean lark]

That's such a farce. You consume tokens, that you buy with credits, which they estimate in messages.
What children just out of Marketing school dreamed up this nonsense?
I'm getting called for dinner .... that always makes me less grouchy.
Catch you cats later. 😺

[frosty zealot]

Wear your slippers! Don't need you slippin in your socks and crackin a hip

[dense rover]

only thing that sucks is it requires your repo to be indexed by github and apparently they don't index any file over 350kb

[lost stag]

cant split the index up?

[lean lark]

And you can't upload binaries.

[lost stag]

I did see something where you dont even need a index anymore

[dense rover]

depends, I'm currently working on a huge userscript, it's 600kb lol

One file, so.. yeah

[frosty zealot]

My honest advice to you

[lost stag]

but cant you ask ur ai to split the file up @dense rover

[lean lark]

Hmm, this is helpful, also from the pricing page.

[dense rover]

I suppose I could.. and then refactor it into the single file after each change

[frosty zealot]

@dense rover Gemini has very generous free daily usage limits that resets DAILY not WEEKLY, do as much work as you can through that they have gemini-cli its honestly not a terrible model and can probably do most things, especailly since I see like SQL queries in your picture, use that for most thing, use GPT for the heavier lifting if Gemini is struggling at something

https://geminicli.com/

[lost stag]

thats what I tell it to do, like for a command ill have a command called /hello and then another folder called hello with all the logic in that

thats just what works for me tho, everyone has their own flow

[twilit bluff]

RIP

[boreal badge]

gpt5.5 not at codex, i see...

[vital hill]

time for a long nighter

[boreal badge]

claudes

[vital hill]

plus windsurf sub

[boreal badge]

claude max 20x?

[vital hill]

nah, not worth paying anything more than $20

[boreal badge]

and I have this monster at my home

[main nimbus]

Cursor is a great alternative, so you get to choose which models, plus get one of the best harnesses (better than CC)

[boreal badge]

vibe coding

[main nimbus]

[boreal badge]

I dont let gpt and claude write code for me

[lament estuary]

TIBOOOOO

[main nimbus]

okay. personal choices respected.

[boreal badge]

I use them for deep research and brainstorming, without them any of my developments would take decades

[main nimbus]

AI is a fantastic accelerator when used effectively.

[boreal badge]

сlaude is very good with ghidra and frida, especially when it comes to themida

[balmy niche]

Is GPT 5.4 Now cheaper to use for codex now that 5.5 is out?

[balmy niche]

YESSSSSSS

TIBOOOOOO

[frosty zealot]

https://developers.openai.com/codex/pricing?codex-usage-limits=pro

[lost stag]

man they should do 2x 3x 4x plans

[eager forum]

Business Plan no pro access 😢

5.5 burns soooo quick

[frosty zealot]

I wonder how many people dumped their usage thinking 5.5 would instantly reset

and are just screwed right now lol

[frail mantle]

the latest codex update rendered my previous sessions unusable ? Am I the only one ?

[blissful copper]

Calling all prop firms traders: Anyone tried automating their evals?

https://propfirmpinescripts.com

[hidden glade]

Just about to try GPT 5.5 now and updated my codex cli, you know what excites me more and makes me happier? The statusline for context used is finally not a bar gauge but an actual percentage instead 🥲 I'm happy.

[rugged siren]

ok what is it..

[frosty zealot]

Cat AI slop videos are my favourite thing on the internet

[cedar parcel]

No reset here yet

[frosty zealot]

That's because its fake

[cedar parcel]

Figures 😖

[lament estuary]

I made it with gpt image 2 🥺

[dusk thorn]

anthropic defcon 1 rn 😭 gpt 5.5 today beating opus 4.7 in alot of tasks, and then DEEPSEEK drops a open source model tonight that beat opus 4.6 max 💀

[frosty zealot]

Competition is good, but low key, %#!% A\

[dusk thorn]

the fear based marketing turned me off

that crap is not helping

[frosty zealot]

I hate that they're so locked down, I can't use any other harness but theirs, I have to use THEIR app, and whenever they feel like it, they can say i did something and cut it off, not worth the headache.

[dusk thorn]

and for what was it a month gaslighted us that opus was fine?

even though it was broken and it was set to medium mode default

and they reset the rate limits, on rate limit reset day to give back...?

i went through 10 bucks of credits on a broken adaptive opus 4.7 thinking for a project in a chat and it just didnt work

scam central

[frosty zealot]

Yeah and even when they're wrong, they're right lol, and they never try to save any face with like resets or nothing its just like deal with it

[sacred plume]

Using 5.5 all day on one mega task, it's amazing.

[frosty zealot]

I had like $400 in credits, I let my plan lapse and when I resubbed they were all gone, I Tried to ask the bot or whatever and it said nothing it could do

[sacred plume]

elevate to support

[dusk thorn]

i think in terms credits cant be refunded

[sacred plume]

like ask for a human/agent/manager

[dusk thorn]

its anthropic not charity

[sacred plume]

ok well if he paid for them and they dissapeared, I'll just chargeback

🤷‍♂️

[frosty zealot]

I would, if I actually cared to use their product, but it just reiterated why I wanted to move away from A\

[frosty zealot]

Most were promotional credits so I Wasnt too upset

[sacred plume]

ah

thats sad

openai team is cool

constant resets to ratelimit

[frosty zealot]

if I knew I would have used them up before it lapsed

[sacred plume]

nigh infinite compute

every 2-4 weeks we get a new model

or new model update that improves it heavily

doesnt go thru reasoning effort hell

[dusk thorn]

i stalk tibos twitter like a bird

[oak trellis]

question so the codex app and vs code not working same time ?

[sacred plume]

and we get double rls all the time it feels like

[oak trellis]

Sign-in failed: Port 127.0.0.1:1455 is already in use

[frosty zealot]

If you looked at my first messages in here, I bashed OAI for like a whole day, boy was I opus brain washed

[sacred plume]

once pro x10 is over though i think im not going to renew lmao

i wish pro 5x would stay at 10x

🙏

[frosty zealot]

try restarting your computer or findf what process is running on 1455

[oak trellis]

its cursor where i installed codex itself .

[dusk thorn]

especially if its like half price, i would think half limits is fair

i think they might keep it lowk because if they can handle dropping a new model and giving these generous limits i wouldnt be surprised

[frosty zealot]

probably a setting to change the default port

[oak trellis]

yes will investigate .. codex cursor wanna use different port ...

[sacred plume]

I wish they'd optimize 5.3 to hell like they did with spark and made it so cheap to run that it has its own RL

5.3 is amazing and smart

it's just not creative at all like 5.4

[frosty zealot]

With 5.5 I'm gonna try to run Low thinking a bunch and see how it handles

[sacred plume]

the issue with gpt models is they suck at creativity because all the creativity training goes to gpt image-gen

[dusk thorn]

it is nice having spark aswell for automations and routines in codex

[frosty zealot]

Is the play to use image-gen to create the mock, and then feed the image to GPT as a 'recreate this in code' kinda thing

[dusk thorn]

100%

like i was adding a focus timer to this one thing and took a screenshot of the ui i had currently, generated a sick ui for it matching my current ui, and then codex replicated perfectly

[frosty zealot]

I feel like as long as you're feeding skills you should be able to run Low thinking with good results

Cause it shouldn't have to think too much

I do dream of the day 'auto' thinking exists though I'm so annoyed siwtching between them

[brave gazelle]

were so back

[sacred plume]

the issue is the current implementations are just predicting how much tokens it would use and whats best for it

[dusk thorn]

codex made vs image gpt made its so close ofc needs a little bit of adjusting but

[frosty zealot]

Oooo

[dusk thorn]

so yeah it does have some caveats but it def helps

[frail meadow]

Is gpt on the right?

[dusk thorn]

yes

[frosty zealot]

What about getting it to build a design template, like varying containers, fonts, font sizes, paddings, margins, basically building a spec and noting on the image so ytou have a full spec for your app/site

[dusk thorn]

good idea

[frail meadow]

That would be pretty sweet to build out with 5.5 pro

[idle cypress]

two 5.5 messages took me from 100 to 30% on my 5h limit

[frosty zealot]

im gonna try it and see what image gen generated

[idle cypress]

anthropic levels of session usage

[dusk thorn]

plus?

what reasoning level

[idle cypress]

mhm

low

[dusk thorn]

damn

[idle cypress]

i mean i fully expected a huge hit

since plus is only 20 dollars a month

but damn

[dusk thorn]

how many lines did it write?

[idle cypress]

thats far more than 5.3

2000

and the result isnt even that good

ill have to do another session on 5.3 to fix these issues

[frosty zealot]

I want you to create a design spec template for my new project, such as varying border radiuses, colors, paddings, margins, font specs, etc. with examples in the image, based on a neon glass morphism color scheme

[dusk thorn]

ah its using the general ai gradients

i mean it looks nice

[frail meadow]

Ofc it chose purple lol but it looks good

[dusk thorn]

but could ask for something with a bit more color variation

as yeah ai is obsessed with purple

gpt 5.5 ability to work for like 20 minutes and one shot a feature gotta be studied

i prob mentioned before but i cant get over it

[brave gazelle]

genius

[dusk thorn]

gonna try my first codex overnight run wish me luck haha

[wooden minnow]

bro did we get 5.5 on api?

[frosty zealot]

ooo

[unborn thunder]

do you guys work locally or cloud first?

[velvet wren]

not yet

[rugged siren]

bro what

[terse kraken]

Are we not getting a reset again? I burned thtough 50% in one day yesterday on $200 pro lol

It was a pretty awesome day

[hybrid ore]

For real?

[hybrid ore]

Hoops

I can't seem to get 5.5 yet on my codex. Still 5.4 on the terminal

[brave gazelle]

im pretty sure they will im surprised it hasnt happened yet but maybe tomorrow so people can get a day to burn on it and then do it?

[sacred plume]

yeah what

[rugged siren]

damm

+7000??

[sacred plume]

yeah and -21,845

[rugged siren]

wow

[sacred plume]

I'm doing massive codebase overhaul.

Getting rid of legacy code.

[rugged siren]

wowwwwww

[sacred plume]

I had GPT 5.5 Pro Extended rewrite my plan into 48 pages

😂

connected to my codebase thru github

so it understands it in detail

[rugged siren]

5.5 pro??

[sacred plume]

and can give longer and arguably better planning without affecting my RL too

[rugged siren]

W

[sacred plume]

ye

Just give this to Codex and it won't stop until its done.```
You are not finished until the entire mega-plan is implemented.

Your first task:

1. Create or update /docs/ADMIN_MIGRATION_MEGA_PLAN_CHECKLIST.md.
2. Convert the entire mega-plan into a checkbox checklist.
3. Mark each item as:
   • TODO
   • IN_PROGRESS
   • DONE
   • BLOCKED, only if impossible without missing credentials or external access

Execution rules:

• Do not summarize partial progress as completion.
• Do not stop after “materially closer.”
• Do not leave compatibility proxies, legacy aliases, fallback shims, duplicate stores, old routes, or temporary adapters unless explicitly listed as BLOCKED with exact reason.
• After each implementation batch, update the checklist.
• Continue working through the next unchecked item automatically.
• Only say “complete” when every checklist item is DONE or explicitly BLOCKED.
• If context gets long, write a handoff section at the bottom of the checklist titled “Next Codex Resume Prompt” with the exact next task.

Definition of done:

• All admin migration items implemented.
• Legacy compatibility proxies removed unless truly required.
• Admin routes and components use the new architecture only.
• Dead code removed.
• Types pass.
• Lint passes.
• Build passes.
• Core admin flows manually smoke-tested or documented if not testable.
• Final response includes changed files, checklist status, test results, and remaining BLOCKED items only.

Begin now by auditing the repo against the mega-plan. Then implement the remaining items until the checklist is complete.```

Its in my custom instructions

Codex is the goat at long codebases

[rugged siren]

That's insane

I can only access ChatGPT 5.5.

[terse kraken]

That uses codex limits though

[sacred plume]

since when lmao

my rl didnt go down at all for 5hr or weekly when i did that too

[rugged siren]

Gemini is translating this weirdly.

[wooden minnow]

wen prod

[main nimbus]

I hope soon (like Friday 24th)

[obsidian prairie]

anyone using codex for frontend?

[plucky halo]

Only alongside image generation

[frosty zealot]

Ctrl + Up arrow on mac is more handy than cmd+tab convince me im wrong

[plucky halo]

Not with the 182376123198 apps I have open

But do agree

[obsidian prairie]

can u tell me a bit more about that?

[plucky halo]

On it's own, I don't really rate codex's front end skills especially compared to Claude. However, with Images V2 it is more than capable.

Before using image generation:

After:

[sacred plume]

How are you using image gen to do front end skills?

[plucky halo]

Before Images V2:

[sacred plume]

Feeding into GPT Image and then having Codex use it as a base design?

[plucky halo]

After:

[plucky halo]

Yep!

[sacred plume]

I'd love if you could share the prompts you use 😄

[sacred plume]

You can tell this mimics the overall design asthetic Codex forces itself to do

the rounded corners

and messiness

[plucky halo]

Nested boxes in nested boxes in nested boxes

[plucky halo]

Ha - the prompts are embarrisingly bad which is a testament to how good images v2 is

[orchid valley]

look unprofessional

[plucky halo]

Sweet. Well atleast it's no longer 'incredibly' unprofessional. That's progress

[orchid valley]

wow

[plucky halo]

Nice chin

[orchid valley]

480P image and ai know

gpt 5.5

[balmy olive]

i oculd recognise that man everywhere

[dusk kayak]

yo

you are doing ur portfolio ?

[orchid valley]

no

[dusk kayak]

u have one ?

[orchid valley]

no

[fleet mango]

same, did you find how to use 5.5 yet?

[hybrid ore]

Yes,the codex update works fine now

[spiral gorge]

help... https://i.imgur.com/NYQU4Ht.png

the font lol what is this HELP

[cedar parcel]

Interesting. The 200 pro plan is not actually 20x. I had a 100 plan (5x). It had 8% weekly left. I then upgraded to 200. Now I have 52% weekly left. It should be much more if 200 is 4* the 100 plan. But pro 200 is just 2 * pro 100. No benefit to a pro 200 plan Vs 2 pro 100 plans despite the pricing page saying that you get much more usage included in pro 200.

[spiral gorge]

so how much x is it

[cedar parcel]

10x plus for 200 and 5x plus for 100

The 20x is bs

[plucky halo]

The 100 plan is 10x currently due to it being 2x.

Normally 100 plan is 5x, 200 is 10x.
Until the end of may, 100 is 10x, 200 is 20x

[fleet mango]

weird, model list isn't showing it for me

[hybrid ore]

Is your codex on the latest version?

[spiral gorge]

my gpt 5.5 fails to .. use the codex in app browser

🙁

[fleet mango]

yea, 0.124.0, i even reinstalled the npm pkg

[hybrid ore]

The new session should actually show it.
Ask your agent to investigate.

[cedar parcel]

Is that mentioned on any official openai channel? The pricing page does not say this.

My cli also no longer says 2x usage that message disappeared earlier

[cedar parcel]

Awesome thanks!

[leaden heron]

Yall think 5.2 is better or 5.5 is better in codex

[orchid plume]

On the checkout page if 5x is chosen it mentions promo 10x till may 31st, but interestingly 20x doesn't show this promo message which makes me wonder if they will keep it after may 31st on 20x

[worthy furnace]

How are the Pro codex limits?

Is it worth it to buy?

[cedar parcel]

No message for me. It doesn't mention any of this on the checkout page.

[orchid plume]

The very final page where you pay?

[cedar parcel]

I've been on 5.2 exclusively but today trying 5.5 and so far so good. 5.4 and 5.3 dont work for my workflows but so far 5.5 seems good.

[leaden heron]

5.4 idk it sucks

So does 5.3

[cedar parcel]

I never get there because my card is linked so I just use the upgrade button and I get automatically charged.

[leaden heron]

a lot more token usage tho

So

5.2 is actually reasonable

[cedar parcel]

Maybe first time sign-up goes to checkout page but I've not seen it in many months

[cedar parcel]

5.2 is never not reasonable it's so freaking good 😊

[leaden heron]

Yeah

[chrome raven]

please who else is getting this "Your account was flagged for potentially high-risk cyber activity. Requests may be slower while additional verification is applied. To regain faster access, apply for trusted access: https://chatgpt.com/cyber or
learn more: https://developers.openai.com/codex/concepts/cyber-safety" i already verified why am i being shown this

[orchid plume]

ah fair enough, yeah when I looked at that page on another account about last week that's what I was seeing 🙂

so hoping 20x stays 20x after May 31st

[orchid plume]

for doing frontend UI work, I think 5.5 is an improvement. I've noticed it's more observant of minor issues (assuming you instruct it to check screenshots) where 5.4 or previous generations would miss them more often and I'd have to say "hey look again at XYZ"

[carmine copper]

yo, I don't know if I'm going crazy or what, but my current perception is that gpt-5.5 (high) uses less usage limit in codex than gpt-5.4 (high) did. Either they increased limits on plus or it is really more efficient or more adjusted to my prompting style, or by chance I am just giving it simpler tasks now (I don't think so). Or I am imagining things.

[orchid plume]

not sure, as a pro 20x user I'm not noticing a massive increase in consumption here, 5.5 high effort. GPT-5.5 is meant to be more efficient on output tokens, even though it's priced higher

[short linden]

anyone testing gpt5.5 yet ? How much diff token usage when compare with 5.4 ?

[warm pilot]

How can I use @browser-use? Even if I open up a browser it can't connect to it. Anyone got this running and can share how?

[orchid plume]

not sure on actual numbers, but pro 20x consumption feels more or less the same, no huge difference here on high effort so far

[bright swift]

Also not seeing a big diff in usage compared to 5.4 yet, on xhigh. But only been a few hours and some long running sessions are still on 5.4..

[short linden]

Ok, let's stay with 5.4 until 5.6 coming 😅

[warm pilot]

figured it out, I needed to reinstall the browser-use plugin

[dreamy flare]

Why was there no usage reset?

[brisk flume]

Since updating Codex to the latest version, browser/network checks no longer work from CLI.

Errors:

• sockets: Operation not permitted
• DNS: Could not resolve host
• Playwright/Chromium crash: sandbox_host_linux.cc:41 ... Operation not permitted

This now blocks external hosts, localhost, and real page/API timing tests.

Looks like a regression introduced by the latest Codex update.

[wary sleet]

do we still got no gpt image 2 in codex cli?

[orchid plume]

you should have the image_gen tool in cli

[wary sleet]

it was just a skill, doesnt generate image without api key, as the last time i checked

[orchid plume]

yeah but the skill has instructions to use the image_gen tool

are you using the latest version of cli?

[orchid plume]

"Default built-in tool mode (preferred): built-in image_gen tool for normal image generation, editing, and simple transparent-image requests. Does not require OPENAI_API_KEY."

subscriptions have access to this tool

[wary sleet]

let me check

yes, it seems its possible when you dont specify a model, thank you. two days ago it was asking for a key in every way

[orchid plume]

no worries 👍

[flint isle]

how can I use 5.5 in codex cli? I'm stuck on 5.4 for some reason

[wary sleet]

did you update

[orchid plume]

latest version of CLI?

[flint isle]

just updated and it said 5.5 is available, but still not in model picker..

thanks for the tips tho, one step closer

[orchid plume]

strange, try logging out and back in maybe

[flint isle]

logged out and in and still not appearing

[orchid plume]

very odd

I'm out of ideas then sadly

[flint isle]

i asked 5.4 to change config file to 5.5 and it worked

[kind jay]

I’m I crazy, 2. ?

[flint isle]

u are not crazy, i didnt look close enough lol

i guess i assumed the first option would be the newest model

thanks yall

[kind jay]

🦊 🎓

[orchid plume]

ah I didn't check the screenshot, I had assumed you had no option, but glad to hear it

[compact dawn]

5.5 x-high doesnt actually consume that much usage

im surprised

it's not like a 2x on my codex usage

[carmine copper]

same for 5.5-high <-> 5.4-high, it's similar

[turbid axle]

ye wth. bro be trippn

so anybody use 5.5 to create AGI yet?

[kind jay]

AGI is not real lil bro

It’s some sci-fi term

[balmy hatch]

When I open Codex app it arrives to 100% CPU. What can I do to solve. It begins today thats weird.

[turbid axle]

what even is 'real'

overrated

pretty sure if we benchmark reality it would sit below claude 4.7

[kind jay]

Okay then AGI is not possible with current architectures

[kind jay]

What specs do you have?

[turbid axle]

UNBELIEVER!

https://tenor.com/view/unbeliever-life-of-brian-gif-22638310

[bright swift]

Just use cli

[balmy hatch]

16 RAM Win 10 and i7-6700

[kind jay]

What’s CLI?

[bright swift]

codex

[turbid axle]

Cool Language Interface

[kind jay]

Your cpu is really old unfortunately not much you can do

[balmy hatch]

But that week I use it and perfect

Today it opens a los of processes

[bright swift]

They are doing a lot of slopdates for the app

I wouldnt use it for serious work

[kind jay]

Try delete it and reinstall, maybe you can clear out some old stuff

[kind jay]

OAI software is really bad though

[bright swift]

tui is pretty decent imo

[balmy hatch]

I do it. Mybe can be today update

[compact dawn]

tibo when reset

me need reset men

[bright swift]

yea i burned my usage in anticipation of 5.5 release and no reset

what is this

scam

[compact dawn]

ahahah

do they usually reset after new model

[kind jay]

I have similar, 6 processes for what? I jusat opened it

[bright swift]

often i think yea

and after big app updates etc to let people try it out a lot

[small violet]

is 5.5 good

[balmy hatch]

No no. When I open codex. Create 20, 24, 27, 20, 22, ... etc

[small violet]

im still use 5.3 codex

[bright swift]

yes, it responds with some humor to insults now

[kind jay]

Updating, we'll see

[small violet]

im saying for coding

[bright swift]

yea me too

[small violet]

huh

[bright swift]

you dont insultcode?

[turbid axle]

how? like "bro, this idea is dumb, implement it yourself"

[balmy hatch]

I reboot and reinstall a lot today

[small violet]

actually?

[bright swift]

no but CC does this seriously 😄

gpt rarely pushes back on bad ideas from my experience

[kind jay]

Just use the chatGPT website for your project, you'll need a new computer eventually (or use Linux)

[small violet]

@bright swift what do u use codex for

i mostly use it for personal projects

[steady vigil]

⚠ Selected model is at capacity. Please try a different model.

well that's new

[bright swift]

to maintain my main business and building two new ones that build on it, plus a very speculative project that will either work great or be a monumental waste of time

[steady vigil]

guess we won't be getting any more resets they prob used all their capacity with images-2

[turbid axle]

• thanks image v2

[small violet]

cool

what do u do

i made a cool tool

and tried to monetize it

not ONE download bro

😭✌️

💔💔💔💔

[bright swift]

my main business is a mix of e-commerce and consumer saas with a few hundred k customers

[small violet]

“few hundred k”

you manage it alone?

gpt 5.5 give me a business idea that is a mix of e commerce and customer saas

[bright swift]

yea built it alone and run it alone, its over 10 years old though

bit annoying because it relies on paid ads too much

pays the bills but wont pay for a lambo, so.. looking for new shiny objects

[bright swift]

think i need a bigger machine

[cedar skiff]

Put it on pragmatic personality

[bright swift]

Hmm have it on "none" right now

[wicked fern]

I got this error on codex on windows when trying to use browser use, What could be the issue?

[turbid axle]

whoa codex app on windows has browser now.

[orchid plume]

desktop app

[turbid axle]

yes

thats what I said

frontend dev with browser active and steering is nuts

[orchid plume]

it's impressive but it has some limitations. I'd love it if I could use it in a headless state too, as well as adjust viewport size, like I can with Playwright CLI

like so I can have it test a webpage's UI to see how it behaves as mobile viewport, tablet, etc

[tawdry jolt]

I hope this gets added soon tbh (nvm it did)

[tawdry jolt]

The in-app browser feature?

[turbid axle]

ya

it works now

[tawdry jolt]

Great i was sad it only worked on mac but glad they added it to Windows 2 now

[dawn seal]

I do not see agents mode in my codex

how do I enable agents mode

[turbid axle]

codex is agents mode

[dawn seal]

there is an actual agent mode thinging in the trailer

I want THAT

[turbid axle]

what trailer?

[dawn seal]

openai's new release youtube video

[turbid axle]

which one

[dawn seal]

the one that has agent mode in codex window

[turbid axle]

link?

[dawn seal]

too lazy to search/post it for you

diy

[turbid axle]

ok

@ivory zodiac what your blog btw 🙂

[cedar skiff]

hah the irony is real

[dawn seal]

lol

there is literally no menu after installataion

[cedar skiff]

Yeah but did you go read the docs and find out about it?

[dawn seal]

yes and it should be there

duh

[turbid axle]

my guy. these agents are for chatgpt, and they are for business/edu accounts only.

you can do the same in codex using automations

[cedar skiff]

ok then, you probably should go read it a bit closer and better

[turbid axle]

... yawn

[dawn seal]

https://openai.com/index/introducing-workspace-agents-in-chatgpt/

[signal tapir]

You talking about this thing?

[dawn seal]

yes

[carmine copper]

that's chatgpt in teams, not codex

[dawn seal]

thank god finally someone with knowledge!

[signal tapir]

Was just going to say, I'm using Teams

[turbid axle]

wow, this guy XD

[dawn seal]

so I need Teams to access this feature

[turbid axle]

https://tenor.com/view/im-out-we-out-peace-out-disappear-gif-15464202

lol

[cedar skiff]

If you actually read the doc about it you will see
duh

[cedar skiff]

the ironic part is you calling out someone else for asking for a link to the vid.
Then coming in here and asking..
Then claiming to have read the doc where it is literally in the first paragraph.

[signal tapir]

I recommend forgetting this conversation every took place. There are much more enjoyable things to focus your attention on. 😛

[bright swift]

wen gpt-5.6

[signal tapir]

Probably not going to happen. I'm betting on 6.0 before that

Although they'll probably change naming scheme again

[cedar skiff]

I'm scared for the next release

I think we are going to keep seeing the price squeeze and i like having all my usage

[bright swift]

gpt-5.5.1

[signal tapir]

The price for 5.5 will probably drop a bit when next version drops

[bright swift]

did model pricing ever drop?

[signal tapir]

I... am assuming?

[bright swift]

dont think so 😄

[turbid axle]

my guess is spuddi is probably a gigantic expensive model, it will be pointed at curing cancer and things like that. 5.5 models will keep upgrading as 'usable' models for the time being

[dawn seal]

you feel offended but its not my fault, I have described feature A and explained that it is posted on place B

do not ask me to search for it on place B and post it on here, I have provided enough context

[signal tapir]

They are incredibly usable though. 😛

[dawn seal]

you are being ridiculous @cedar skiff @turbid axle

[turbid axle]

for sure, plenty smarts, just make them faster and cheaper I say

[signal tapir]

I don't care about faster. Cheaper though

[obsidian prairie]

does anyone know how much usage u get if u pay 20$ for api compared to 20$ plan?

[turbid axle]

I mean, not faster? you want faster..

[bright swift]

dont feel like model intelligence is the bottleneck for me tight now

[signal tapir]

They are supposed to be the same

[bright swift]

more like workflow + costs

[turbid axle]

tight tight tight

[dawn seal]

I dont get people that complain about the pricing becuase

1. They can simply use older version
2. New thing being pricier is something natural, that's like how everything is

[exotic cave]

In a few years things today will look like a horse and a carriage vs a car. We move quite fast, faster than say PC updates along the years. besides if we keep with the 3-5% upticks in reasoning every 2 month's or even 4, in 2 years it'll be a different world indeed.

[bright swift]

pricing isnt that bad. just hurts on new projects where its not paying for itself yet

[signal tapir]

I'm still waiting for an offline model I can use as good as say 5.4-mini, on hardware costing less than 10kUSD

[dawn seal]

are you on the 100+ dollar spectrums?

(I am too)

[signal tapir]

Yea, like the game I've been working on for a year. 😂

Takes time to make back the money, if ever

[dawn seal]

I think your project paying itself is a very difficult challenge

making money is very difficult

[kind jay]

Lowkey Mac minis are the way to go

[dawn seal]

I do not think making AI should necessarily help you to achieve that task (making money with your products)

making money is more than just writing a good software

[signal tapir]

I've yet to find a model that can do it.

[turbid axle]

loki

[bright swift]

i have 2x pro right now, probably 3rd soon. according to ccusage/codex i couldnt justify API pricing yet though

[dawn seal]

all of these vibe coded start ups and stories of making millions bucks with AI, are often just bait to sell these "online lessons"

[signal tapir]

It's easy to make money if you don't care about your customers one bit.

[dawn seal]

what is the process of your usage like?

are you making a product?

[bright swift]

yea complex data analysis product, and building custom agent tooling for it

i have pretty strict workflow with post edit linting etc which tends to use more tokens

ironically i handcoded the business/product that pays for all of this and barely do anything in it with agents beyond some maintenance

[dawn seal]

I guess you are using it because it is still cheaper than using someone?

I think in that sense, you "made" money

[bright swift]

yea the stuff im working on i would have to raise VC to even think about hiring and attempting it

[signal tapir]

People are really expensive, but a bit more "self propelled" than AI. 😛