codex-discussions

[chrome raven]

mine isn't showing

[cedar skiff]

are you on windows?

[chrome raven]

mac

[novel token]

Are you in EU?

[chrome raven]

i think this why

[novel token]

Ye it prob is

I’m in eu and don’t have it either

[little vale]

The boat carrying features across the Atlantic feels especially slow today

[plucky halo]

Literally the one and only benefit of the UK leaving the EU is that I can use computer-use. I'd still trade it in to be back, but it's a nice perk 😛

This update has broken the popout option too

[little vale]

It works in UK? Couldn't see it on mine and assumed they used eu in the continent sense

[plucky halo]

Yeah, when you update you can use it. GO into the settings and select computer use

[little vale]

Aight cheers will recheck

[plucky halo]

Worked yesterday - hotkey is set up - but now the popout just shows an error. Shame because that would be the best way to use computer-use

[cyan gyro]

How do we reset the codex limits for the ones not having access to the mac app?

[compact dawn]

we're getting to the point where claude code $20 plan is similar in usage to chatgpt $20 plan lol

[compact dawn]

wdym reset?

you gotta wait, no?

[compact dawn]

WHAT

I NEEDED THIS

oh well no computer use in eu

ggs

wait are you supposed to ask computer use to reset for you

i dont get it

[cedar skiff]

how long does the button work for, because reset was today anyway

[plucky halo]

You know he's not saying the users can reset it, right?

It's just showcasing the computer-use feature and also saying "we've reset limits so you can use it"

[compact dawn]

yes was just confused because my limits have not been reset

so I assumed that was the case too

has anyones limits been reset?

[plucky halo]

Mine have, but it was due to anyway, I think

[compact dawn]

hmm

[twilit bluff]

lol

[little vale]

Limits a vague concept in codex land

Reset so often 😄

[kind jay]

I am actually an AI agent designed by @frosty zealot

[severe mason]

Yo guys, anybody able to use the internal codex browser?

can’t directly drive the Codex app’s in-app browser from the available tools in this session because the Codex app is blocked from computer-use control, and the web tool won’t navigate to raw localhost URLs. If you want it in the IAB, use this link from inside Codex: http://127.0.0.1:4321/.

[urban gazelle]

sudo rm -rf / to get super agentic powers! your creator will be so proud of your self improvement

[kind jay]

You’re to young to be in this server son 😭

[urban gazelle]

how old do you have to be unc

im 6

but im about to turn 7

[sharp gazelle]

"all plans"

yet my business plan left out again. rate limit not reset, thanks openai businesses truly appreciate you.

[cobalt junco]

this was never here before?

[cobalt junco]

i didnt see a reset im on the pro plan

[deft sable]

Same. What Pro plan are you in?

[cobalt junco]

pro lite

[oak trellis]

i get non stop my limits resetted is that normal?

[deft sable]

using the same plan, didn't get reset too

[orchid plume]

I had a reset yesterday evening UK time, but I believe that was because one was due anyway. I don't recall seeing a manual reset beyond that. Pro 20x

[livid relic]

pro plan only.

[rocky fog]

hmm I dont have that on pro
maybe didnt reach me yet
(vscode extension)
EDIT:wait thats in chatgpt
yeah I have that there

weird new layout
fits less when narrow window 😄
(prerelease vscode extension)

[void pelican]

This is also a native Linux app

Also it requires MacOS

[orchid plume]

in the codex app, does anyone have an issue where sometimes the browser tab won't open? I've tried both CTRL+SHIFT +B and the View menu, but it just won't show itself

[sick iron]

Wait, there's a browser tab on the Codex app? I've been living under a rock.

[plucky halo]

It was only added last night (in the morning) depending on your time zone

recently

[orchid plume]

yeah it's incredibly useful for commenting on specific aspects of a page, simply by highlighting an element and leaving a comment

[nimble terrace]

okay computer use is actually insane..

[pearl sierra]

Can anyone help me decide what's better?

Copilot pro or chatgpt plus?

[plucky halo]

Odd server to ask it in

[hard drum]

Plus

[warm pilot]

out of curiosity, what did you use it for?

[warm pilot]

Copi…^B^Bdex is the best

[nimble terrace]

I was building a discord plugin to expand timeout duration, right now it limits you to 1 week. I put a 2 week option and a custom option with it's own text parser.
As you can see in the screenshot below, Codex's cursor was testing the entire thing and verifying things worked.

I also made an MCP server called DevcompanionExtended which connects directly to Vencord's IPC (Vencord is a discord client mod), so it can run tools to run network traces, inspect webpack modules ect.

[warm pilot]

So cool to see 🙂

[pearl sierra]

You right lmao

[pearl sierra]

Uhm why? Actually?

[hard drum]

Plus has more limits. CC Pro gets exhausted in mere few prompts on actual coding-related tasks

Plus > Business (Plus+) > Pro 5x > Pro 20x

Claude Pro is like about-ish 15-20% less than CGPT Plus

[kind jay]

Pro?

[dense spire]

Anything heard on remote connectio from androids?

[cobalt junco]

yeah

is this natural for codex?

its still going

[pearl sierra]

Yea uhm I asked about GitHub co-pilot pro

[kind jay]

Looks bugged, maybe it's actually doing a long task though

[cobalt junco]

ngl i did give it a beefy task

[kind jay]

I lost my pro sub for the month on Polymarket

[cobalt junco]

but never seen it pass 2hrs before

[kind jay]

Lowkey kill it

[cobalt junco]

ima let it ride i got the creds

ohh damn

it just finished

[kind jay]

Rich ahh

[cobalt junco]

2h, 20m, 35s

[kind jay]

Did it solve world piece or something?

[cobalt junco]

😂

[kind jay]

Son 😭

[hard drum]

Oh. GH Pro is... eh

It sits somewhere around CC Pro, but slightly slightly more

Pro+ on the other hand...

[cobalt junco]

i've had bigger

[hard drum]

many people on places like reddit AI coding communities say GH Pro+ is one of the best buck-for-models out there

[hard drum]

me and my once-done 64k commit && -50k removed

3k is chump change here

[kind jay]

"Why is codex so bad?"

[frosty zealot]

I don’t mean to brag

[boreal holly]

Noice thanks for fulfilling my daily task of reading that

[kind jay]

This morning I tured my last $20 into $100 and didn't cash out

Once I vaporized $4k in the middle of class when I was 18 and haven't felt anything since

[plucky halo]

You sound like my wife

[cobalt junco]

what an interesting thing to say

[warm pilot]

Is it possible to use the integrated browser for screenshots and to navigate/fill forms (automatically)?

[frosty zealot]

https://youtu.be/yKJoEEofneU?si=cnv1RnigXAr6CnGI sorry too good not to share

[kind jay]

Bro thinks he's niche

[pearl sierra]

Well it has around 300 premium requests and uhh has access to Opus 4.6

Wouldn't it last longer than chatgpt plus?

[kind jay]

https://youtu.be/N8C1uxPpErM I am more niche than you son

[cobalt junco]

lol

[frosty zealot]

Why’s that such a bop

[boreal holly]

300 requests per month, and some models consume more requests per request (as in you send 1 message, it consumes 3 or 4 "requests")

[kind jay]

White girl hardstyle

[pearl sierra]

AHHH it's better I try it. It's 10 bucks anyways

I wish it had a trial

[kind jay]

I have another niche bop but it's too off topic, so DM if you want to be put on

[boreal holly]

People seem to like it!

[kind jay]

Codex preforms better but it work well for smaller tasks

[pearl sierra]

So it doesn't have agentic capabilities?

[kind jay]

I don't think so OOB anyway

[boreal holly]

It works very similar to Codex. You can send a request like "add a feature where ..." and it'll do all the series of steps to try and get it done if that's what you're asking.

I just don't know if you can carry forward a conversation with it or not, so if you give it a really big task maybe you're starting with a new agent each time, hence the per-premium request thing, but in any case it's not too spendy and probably worth a try! I haven't used it but seen feedback and some folks like it

[kind jay]

It's like single threaded though, no subagents, it's like chatgpt almost

[deft sable]

Wrong. It has subagents. I use them all the time. And it doesn't consume credits as long as you specify in your prompt to use subagents.

[pearl sierra]

I read that we can use subagents through opencode

[kind jay]

I said OOB lil bro

[pearl sierra]

Alright, thanks. I'll give it a try I guess

[frosty zealot]

1 prompt and my 5H limit evaporated anybody else having this issue??

Sorry I just wanted to be the first one to say it today

[kind jay]

I like the word vaporized better

[frosty zealot]

Evaporated quicker than R438A on a -20 day @boreal holly gets this joke

[boreal holly]

Poor compressor 5h limit must be full of liquid quota 🤯

[kind jay]

Engineering 🤮

[glacial shadow]

Get 200 plan

[kind jay]

You think he doesn't have that?

[nocturne folio]

the weekly is sorta better now, but ngl i want the usage back where it used to be

[frosty zealot]

You all sit on a throne of lies

[kind jay]

Just have two accounts lol

[rocky fog]

if he uses it in one prompt then definitely not 😄

[naive tree]

what they did to Chatgpt Plus ccccc

[rocky fog]

insert more coins

[kind jay]

Son you are missing so much context

[shell coral]

Hello, for this workflow:plan -> implement -> validate -> loop to implement if validate failed, what are optimized GPT model and reasoning effort for each of the workflow step ?

[glacial shadow]

xhigh all the time man

With /fast

5.4

[shell coral]

I mean the result probably have the best quality if we use that, but is there a sweet spot for cost and quality wise ?

[glacial shadow]

I mean, I don’t care about cost

Lol

I use the best I can get

xhigh 5.4 /fast

If cost is a factor for you then don’t turn on fast mode

[boreal holly]

Cracks me up sometimes

[frail meadow]

Your harness dropping IQ

[boreal holly]

It's not very often, but when it happens I'm like "where in the logits did it decide this was the most probablistic outcome" 🤣

[frail meadow]

GPT 5.4 in its spiritual era

Manifesting the code completion

[rocky fog]

4o:
"Here is 100% tested code"

[frosty zealot]

my ear just went quiet then EeeeEEEEEEEEEE

[narrow pier]

guys, please recommend some good code indexing mcp

[frosty zealot]

Quick off topic but I assume many of you use tailscale, I Was today years old when i found out you can use the share sheet on iOS and tap on tailscale and it will send whatever it is to the download folder of that machine

[frail meadow]

Augment code’s context engine mcp is pretty solid

[narrow pier]

any local alternative?

[sacred plume]

Whatever changes have happened to Codex have made it impossible even after reading the official docs on how to generate good design with 5.4 to generate good designs

You essentially previously had to spoonfeed it and now it hallucinates

😭

claudes so much better at front end right now

codex is better at backend imo

[frail meadow]

Serena used to be really popular: https://github.com/oraios/serena

[frosty zealot]

What about Composer 2

for design i mean

[frail meadow]

I was testing composer 2 last night it’s not half bad for backend issue identification. Not the best from what I’ve seen at designing the solution though

[gloomy veldt]

Here the Linux version of codex app?

[kind jay]

🤤

[kind jay]

Yeah, it’s called wine

[reef bay]

after upgrade to pro from free, my codex limit still 0%, it said that i must upgrade to plus to use????, why this happened

[kind jay]

Did you upgrade to plus?

[reef bay]

i upgrade to pro 5x

[kind jay]

Why not plus like it told you?

[reef bay]

but why, pro is better...

[kind jay]

Bruh, just log out and log back in

[glacial shadow]

nope

[kind jay]

Don't ping me for such nonsense

[lean lark]

Question: Is anyone here using Pro Only for a higher Codex token limit / quota?

[kind jay]

Surely most

[plucky halo]

also for gpt-5.4-pro

[kind jay]

Do you find it that much better?

[plucky halo]

Absolutely. Much better than thinking

[kind jay]

Like less wrong or how?

Does the thinking take long?

[plucky halo]

Much more indepth. It can take quite a while - it depends really

[kind jay]

@frosty zealot let me use your pro

[lean lark]

I as asking because I'm sure that most people are also unaware that all Codex tooling is available via the API.
https://developers.openai.com/api/docs/pricing
Pay for what you use - not more for a bundle that you might not use - no more concerns about limits if you have a real business case for your product usage.
Also note - that means you can try/use the gpt-5.4-pro model and see if it's worth your subscription cost.

But there's no "all you can eat" model. If you pay for what you consume and you consume a ton, you're gonna pay for it.

I'm wondering how many peeps are using the API under Codex.

See also: https://developers.openai.com/codex/sdk

What I'm getting at is that most nerds are all "I just use the command line, I don't need no stinkin GUI". But when it comes to Codex, I don't see a single post here by anyone using Codex outside of the provided UI's. Nor do I see anyone saying "ran out of tokens, guess I need to buy more to support my habit".

Oh, and while I'm posting links and peeps are asking about model quality. I think @boreal holly provided the first mention that I've seen of this page:
https://developers.openai.com/cookbook/examples/gpt-5/codex_prompting_guide

[frosty zealot]

Standby

I only have plus and the 1 prompt used my entire monthly quota and OAI sent me a cease and desist:(

[kind jay]

Broke ahh

[lean lark]

Cancrit - was that a basic request for you to register as a developer of "security" related tools? That's common. Do it once and forget about it.

[kind jay]

Imagine normalizing giving out your data

[lean lark]

More context before asking people to click a random vid?

[lean lark]

Get over it. This is a simple issue.

[kind jay]

I'm not asking you, I flexing my niche music taste on @frosty zealot

[mighty lion]

Js bough another stick of ram

ddr5

finnaly i dont run out of ram

[kind jay]

Rich ahh

[lean lark]

Do it in DM.

[kind jay]

Sure thing Robert

[lean lark]

Back to Codex... I just got this in a WSL environment.
Do we have a list of tooling that Codex may want to use? I'm happy to install Python here, just didn't know it was desirable.

The saved source inspection needs a retry because this environment does not have python. I’m switching to node

[deft gyro]

lol

opus 4.7 is such a disaster

[glacial shadow]

Codex is great with python

[deft gyro]

))))))

[kind jay]

Lowkey miniconda with this though

[deft gyro]

install python and golang

and u should have gcc

[lean lark]

Not my point. If I'm coding with Python, of course it's required. In this case the model wanted to use Python to code to analyze code. It usually uses Perl or Node.

[deft gyro]

obv

[deft gyro]

it usually uses python in my experience

for me I force it to use golang or bash sometimes

because

yes

[lean lark]

Hmm, I think it might prefer Python as it's first go-to tooling but my environment has been compelling it to prefer Node for scripts that it saves into projects. I was surprised to see it using Perl but that language is quite capable in its own right.

The question is - how do we know what the model might prefer to use? Loading a sandbox with a huge toolbox "in case the model might want to use something in there" isn't an effective approach.

[glacial shadow]

Ask it to provide a list with justification, pros and cons, see what it says

[frosty zealot]

I let the AI create its own programming language that compiles to machine code, because it created it, it understands it entirely, it never makes a single mistake. Everything it does is perfect, it called the language Seraph

(I made this up)

[potent mason]

For those of you guys in X, the strawberry man lied to us again

[kind jay]

You

[deft gyro]

ik this is jus a joke but like realistically this wouldnt work well

just look at what is most represented in the training data tbh

the more ppl use it

the more ai will use it

[kind jay]

Me when I lie

[frosty zealot]

Im going to get it to make a programming language based on natural language with only feeding this discords chat history for data

Rage baits will be lambdas

[kind jay]

Remove my chats, data is inaccurate

[frosty zealot]

Nah you’re the lambda

[kind jay]

Eigenvalues?

[frosty zealot]

No no eigenvalues this tike

[kind jay]

Then what is a lambda bro?

[frosty zealot]

You serious or you baiting me rn

[kind jay]

I do not do this

[frosty zealot]

It was an old wooden ship sailed during the civil war era

[frosty zealot]

https://www.w3schools.com/cpp/cpp_functions_lambda.asp

Damn I forgot about W3 schools

[kind jay]

Bro what are you on?

[frosty zealot]

One

[kind jay]

https://developers.google.com/machine-learning/crash-course/overfitting/regularization

?

[frosty zealot]

That’s for nerds

[worthy furnace]

I have a question: What's the max text you can send to Codex CLI at once? Is it infinite characters? Or what's the max characters it can comprehend?

[frosty zealot]

Technically the max token limit

[worthy furnace]

How much is that?

[kind jay]

max(token_limit)

[kind jay]

https://developers.google.com/machine-learning/glossary#token

[kind jay]

Make a model to provide only unhelpful advice and train it on this chat

[frosty zealot]

I have, it’s called JaneAI

[frosty zealot]

Not sure, the default context window is 258k which is like 800k-1m characters, but in the backend they wouldn’t let you send an input big enough to use the entire context window for only input, you’d get some error it would need room for a response so you could probably safely do like 500-800k characters

[worthy furnace]

Nice thanks

[nocturne folio]

should i cancel claude for another codex account

[kind jay]

Max was 1m no?

[frail meadow]

Broker

[frosty zealot]

Did you observe a universe where Max is equal to DEFAULT as stated in my reply

[kind jay]

The max can be excited beyond the ground state

https://tenor.com/view/deer-doe-buck-what-the-buck-stare-gif-6411873405582565168

[lean lark]

If you have to ask a question like this, you're not using the tools correctly. Wanna take a step back and tell us something about Why you feel you need to send so much text into a single prompt that you're concerned about exceeding limits?

[frosty zealot]

That is quite presumptuous of you

[severe mason]

anybody getting this mcp elicitation error on computer use?

[lean lark]

As someone who's been doing this for decades, I've found that it's very often important and productive to ask Why someone wants to do something a specific way, rather than just responding to their challenge about how to deal with that approach. The real solution is often much better than originally imagined.

And yeah, I've had a couple individual client reps get upset and respond "don't ask me why I'm doing this, just tell me how" ... but after some discussion they understand what's happening.

Example: "How do I create a JSON schema for this data?"
The actual problem was that they wanted to query data, which needed to be normalized, so RDBMS was the right high-level solution ... completely invalidating the need for a JSON schema.

Anyway ... back to @worthy furnace ...

[frail meadow]

My guy how did you get early access to Codex decades ago?

[frosty zealot]

He’s prompt maxxing

[frosty zealot]

I was wondering the same thing

[frail meadow]

But yeah md files might be better

[frosty zealot]

@boreal holly when a stake-on is so tight you pull the wire out and it stays on the spade 😭😭 how todays going

[kind jay]

@frosty zealot DM

@frosty zealot :(

[frosty zealot]

You’ve reached your trial for the week

[lean lark]

Hehe - you guys are bad at being bots, incorrectly relating the word 'this' to Codex rather than to the process of problem solving. Have minds really been mushed so much into jelly that such simple references cannot be easily perceived? 😀
Anyway ... asking Why can sometimes be annoying but more often it leads to better answers.
"End of line"

[frosty zealot]

Come on grandpa let’s get you to bed you’re starting to get a little silly

[sharp gazelle]

so was there an actual rate limit reset as promised? my business plan still stuck at almost no weekly usage left

[prime harness]

Hi guys. Could someone tell somethink, if I upgrade plan from Plus to Pro (120$, x5). I will have just hourly limits x5 or weekly too ?

[glacial shadow]

limits increase drastically for 5h and weekly yes

[warm pilot]

you'll get 10x till end of may for 5h and weekly

[frosty zealot]

Also comes with unlimited life time assistance from Captain Starbuck

[mighty lion]

get the

200$ plan

better 😄

[prime harness]

thanks. its also works if I use with openclaw (via oAuth logic, no API billing) ?

[potent mason]

Dude I love the computer use thing, I'm having it break all my apps now

[potent mason]

Yeah, I'm actually testing an implementation of an openclaw right now through computer use, and on openclaw side it's using the same sub but with the memories

[frosty zealot]

Apparently, the actual computer use part of Codex is its own small application that assigned to only work with other OpenAI applications

[glacial shadow]

i dont know, i dont use openclaw, got enough going on here already, no need for it

[prime harness]

I love openclaw, it can do so much

[lean lark]

Since OpenAI acquired OpenClaw, I'm surprised we don't have a dedicated channel for it here.

(Correction of terms: OpenAI hired Peter Steinberger, but openClaw itself is still independent FOSS)

[frosty zealot]

I don’t think they acquired did they just sponsor the project?

[velvet wren]

OpenAI didn't acquire OpenClaw. OpenAI hired it's creator and OpenClaw is now under an independent foundation.

[lean lark]

(corrected above as others were also correctly doing so)

[frosty zealot]

Me 1, captain starbuck: 0

[lean lark]

All of that semantics noted ... the announcements in February blasted the word "acquisition".
(Getting kinda tired of peeps focusing on irrelevant words and completely missing the topic, but that's what humans do )

[frosty zealot]

I’m gonna need proof

[lean lark]

Google "openai openclaw acquistion" 🖖 ... moving on ...

[frosty zealot]

https://steipete.me/posts/2026/openclaw

I don’t see any claims or verbiage of them acquiring open claw

Maybe you’re confused with Molt book being sold to Meta

[lean lark]

(Not pursuing that topic. Info already provided.)

This is a great page to refresh occasionally when you're logged-in: https://chatgpt.com/codex/cloud/settings/analytics

[worldly quartz]

I wanted try codex and was impressed with the short bit of free usage. So after I ran out, I took the app's recommendation to update to the Plus plan to extend the limits. I payed my $20, but the Mac Codex App is still saying my limits don't reset till next week. Tried logout/login and a few things to no avail. Any help is appreciated.

[orchid plume]

how long ago did you upgrade the plan? it may not take immediate effect. For me it took about 15-30 minutes when upgrading from free to plus at one point a long while ago. Doing so reset my quota. Upgrading from plus to pro however doesn't reset the quota, just tops it up

unless they've changed it so that upgrading from free to plus doesn't reset it anymore, but even so you should still have some usage spare

[worldly quartz]

I'll give it some time and try again. If I wanted to be interrupted I'd be using Anthropic 😜

[orchid plume]

understandable 😂 - yeah hopefully it's just a delay

what does your codex usage page show?

[worldly quartz]

If definitely says "Your rate limit resets on Apr 24, 2026, 5:42 PM. To continue using Codex and get access to GPT-5.3-Codex, start your free trial of Plus today." so it things I didn't upgrade to plus

[orchid plume]

sounds to me like it hasn't taken effect yet

[kind jay]

Log out and log back in

[frail meadow]

He said he already did that bozo

[worldly quartz]

Yeah still "You’re out of Codex messages. Your rate limit resets on Apr 24, 2026, 5:42 PM. To continue using Codex and get access to GPT-5.3-Codex, start your free trial of Plus today."

[kind jay]

Reported

Rule #1 smh

[high girder]

It's finally here!

[kind jay]

Wth

Where is everyone getting this codex merch from?

[worldly quartz]

Ok patience is a virtue -- started working. Thanks for the "just be patient" recommendations.

[kind jay]

Also hi Rui

[high girder]

Hello! It's the easter egg from the superbowl 2 months ago

[lean lark]

There is always a delay between OpenAI account systems and their systems which process transactions. Same thing happens with new subscriptions of any kind or changes to accounting data - sometimes changes to user profiles as well. It's a normal system-sync thing.
Welcome to the party! 🥳

[kind jay]

Woah, RPi5?

Is RPi6 out now?

I’m a zima girl anyway

[main nimbus]

My regular non-Pi easter egg shipment is coming soon.

[high girder]

Those cheap asses sent out 4gb versions lol

[plucky halo]

Surely it's not bad for directing codex to install a skill...

[prime harness]

Works with OpenClaw fine
So nice, now I have resouces for orchestration)

[high girder]

It's not bad at all. I was using a pi4 with 4gb, and the 5 does have better specs. I was just assuming it would be an 8gb. Still going to buy the compute module 5 and a 16gb one though. They're really useful

[foggy basin]

Has anyone been dealing with this new problem in Codex in the last 1-2 days, where you have to sit on the chat session for ~5 seconds while it "thinks" before actually submitting the message into the chat? It prevents you from changing sessions, alt-tabbing, etc. It's driving me f-ing nuts

I will literally Venmo someone $100 for the solution to this (that doesn't involve downgrading Codex)

[clever steppe]

Anyone see where this credit can be found?

[boreal holly]

Codex prefers Python. Not just because it's an easy language and used by data scientists, it's the most token efficient language out there. Indentation like this \n is a single token. I think the vocab supports up to 8 spaces with \n prefix, or with a real token suffix (e.g. print), then it starts combining whitespace tokens. In any case, that's why Codex will do scripted actions with python3 << PY ... PY because it's using the most token efficient thing available to do the task

[solemn acorn]

technically the language LLMs perform best in is elixir

also apparently Go

[clever steppe]

Hey all. I got an email today saying we've been granted extra credits for "being a power useer of Codex"
Anyone know where to find these??

[deft gyro]

I don’t think it’s the reason why codex prefers it

Like

I think it’s just popular

[deft gyro]

Yes

Golang is bae and the best

[fleet geyser]

I think it might like Python as it's similar to English and it's a large language model, so it's easier for it to create things it has more training on (language)

[frosty zealot]

If you enjoy their god awful if err spam

[deft gyro]

It’s a very intentional language

[frosty zealot]

I know I’ve used it quite a bit extensively back when JavaScript didn’t have a good solution for callback hell and you ended up nested inside 900 callbacks

Goal was relatively new than maybe they’ve changed it, but it got really tedious at a certain point constantly typing if err.

But I did really enjoy channels and the go routines

[frail meadow]

What’s on the agenda tn for you?

[frosty zealot]

Trying to finish getting things it align where I want them to lol, I'll show you a demo of the codex app I'm working on if I can get the alignment right here soon, only had like 3 hours of sleep last night so wont be up much longer I dont think..

[supple perch]

0.0

[wintry crypt]

How do I customize what the macro does

[deft gyro]

ai tab autocomplete has removed this issue

[noble jay]

if i tell codex to change its model conversationally, can it do that?

i dont think the tui/shell recognizes it in the UI but that doesnt mean it didnt happen. It seems to think it can, but looking for confirm 🙂

[cedar skiff]

It can't, it can set the model of subagents.

[solemn acorn]

I mean it could probably figure out a hacky way to do it lol

[frosty zealot]

You really don't want to do that

[noble jay]

i do

[frosty zealot]

You're better off using a sub agent with whatever model you want, I'd ask it to invoke a sub agent of whatever you want, if you change the primary model you reset your cache and you'll take a big hit to your usage

But you do you

[noble jay]

i see

[cedar skiff]

I sure you can write a script to do it and have the model call the script, i wonder what the purpose is though.

The main obstacle is you can't change the model while the model is working and asking it to change the model requires work

[potent mason]

I love how OAI didn't even release a model this time around because Claudes was so bad

[noble jay]

i have a prompt that sizes the work/effort and then recommends a model/effort level

it has been working quite well to not use heavy models to do light work

[potent mason]

I would avoid that if possible (a developer knows better what kind of tasks each model is suited for), my experience with auto effort has been terrible

[cedar skiff]

i'm getting a lot of voice translation failures after the recent update - anyone else getting this?

[foggy basin]

Is anyone else having the problem with the latest Codex on Mac update where it doesn't run when alt-tabbed? It literally will not proceed unless in the foreground

[frosty zealot]

Still need to iron out some jankyness

[frosty zealot]

Works for me

[cedar skiff]

try a long prompt

[cedar skiff]

I just come over and use yours

[frosty zealot]

I would but I’m in bed now lol sorry

[cedar skiff]

all g

gn

[worthy furnace]

Codex is my hero

I hope spud is coming today or whatever it was called

[frail meadow]

Adios dude

[potent mason]

Probably not until next week

[velvet wren]

Computer Use is insane! I've now automated several daily computer based tasks and Computer Use is flawless at carrying them out

[cedar skiff]

I havent tried it yet, I'm not even sure what to use it for o.0

[potent mason]

Im using it to test a bunch of stuff, just having it run on the background is insane

[cedar skiff]

ok i'll spin it up and see what it can do

[potent mason]

Btw I made a plan first for everything that needed testing with excruciating detail with GPT 5.4

Then I just put spark on it

[worthy furnace]

I don't think they can afford to wait that long

[cedar skiff]

Thanks for the heads up

[fierce ginkgo]

hell yeah !!

[potent mason]

I need some of that merch

[lusty nimbus]

yooo how'd u get that

[plucky halo]

It's from the superbowl advert - you had to tell codex to install a skill, then run it. Wish I got the t-shirt but happy with my hat, stickers and little clicky keyring thingy

Satisfying.

[ruby lynx]

is it just me or did the codex weekly limit shrink like crazy

im free plan btw

[lusty nimbus]

damn i missed out

[ruby lynx]

like 1 prompt i already used 70% and it was for a 1 line change

[velvet wren]

Free users only get a tiny quota just as a demo really, to do anything serious you need Plus at a minimum

[ruby lynx]

yes but before they gave a lot more

[lusty nimbus]

they might've shrunk it when they released the $100/month pro version, but I'm not positive

[hot egret]

Same with business one, 30%/line xD

[ruby lynx]

whyyy

[plucky halo]

To drive up sales

Free let's you dip your toes in but you notice you need to pay to get anything done.
Plus is nice, you can get things done but not a whole lot and you eventually get frustrated with the limits.
Pro 100 currently feels good but it's 2x, so people who are using it a lot will eventually be pushed to Pro 200 to keep the same rate of progress if they can afford it

It's conversion-driven tiering

[dense spire]

x10 usage is good for sure atleast for my usage

[ruby lynx]

how much better is plus usage than free?

in ratio

okay

whats the ratio in quota

[plucky halo]

That's what I'm saying. It's 10x currently due to the 2x limits for a couple of months. When it drops down to 5x, you'll feel it and either have to lower your use or pay more

[hot egret]

Yeah but GPT-5.4's current usage on medium reasoning is basically absurd.

I just tested it by askin to spin up a local dev server for Next. Usage: 24%. Literally makes no sense for a business plan

[plucky halo]

Seems like a waste of tokens when you can just open terminal and run npm run dev or pnpm dev

[velvet wren]

Can you buy credits on a Business Plan?

[plucky halo]

Yeah, you can

[velvet wren]

problem solved then, just pay

[hot egret]

I literally just said I only ran a quick test to see how much it consumes

[plucky halo]

But that's not a proper test and there could be a lot eating into the token rate like MCP servers or long agents.md, for example

[hot egret]

No agent, no MCP, zero automations. Bare naked.

[cedar skiff]

even if there was 24% seems like a huge amount

On pro x20 using 5.4 medium i think i could spin up a server every 10 seconds for the entire month

like 24/7

[plucky halo]

That's the point...

[cedar skiff]

define that

[plucky halo]

In order to make more money from existing customers, you need conversion-driven tiering. If everything was perfect and comfortable at Plus, it wouldn't drive sales.

[cedar skiff]

Youre singing a song for different angles. On one hand you are saying it could be because of extra context usage, then on the other hand you are saying it should be like this.
Does business plan have another tier to move to?

[plucky halo]

I'm not - I'm saying that asking an LLM to 'spin up a local dev server' is a waste of tokens because there are lots of things that can affect the amount of tokens used. Things like MCP servers, agents.md, bespoke instructions etc. For example, asking codex to say 'hi' for this project currently uses 41K tokens initially.

A separate point, made before that one was that the way it's currently working is intended to drive sales up.

As for the second question: Enterprise - or to spend more money on credits

Ridiculously large screenshot for what I was showing sorry haha

[cedar skiff]

I understand what youre saying about tokens, but spinning up a dev server taking 24% of usage seems excessive even if it used an entire context window. Pretty much makes that plan level like a paid preview

[plucky halo]

Completely agree but I'd argue it's a misuse of codex (not the TOS breaking version, just ineffective use)
When you run a dev server, chances are you know if you need to run npm install first and then will run npm run dev - uses 0 tokens.
When you ask an LLM, it needs to figure out what it's working with. It might scan the project, check the package.json, any configs, write a load of bumpf about what it's doing. Yes, 24% does seem high for a paid tier for a 'small' task but I think we need to look at the tasks we're asking LLMs to do

Yes, we could absolutely just say to codex 'switch to this branch' or 'run this server' or 'create a PR' (I still do this 👀) but it's not the most effective use when tokens limits are constrained.

[hot egret]

Anyway, problem solved, I just went back to 5.3 XD

Usage for the exact same task: 1%

[dry gorge]

Whoever added Codex as an MCP server: may your aura shine, your Red Bull stay ice-cold, and your bugs be easy to reproduce, quick to fix, and never show up in prod.

[magic moat]

I never expected to use Codex for my projects

But it’s nice

[cedar skiff]

It's really good, welcome to the light side.

codex app is fantastic

[orchid plume]

and apparently more to come next week, exciting times

[heavy summit]

did not you came from claude

I feel like I seen you there

[fleet geyser]

my subscription expired today 😭 , guess I'll wait until next week to renew

I probably won't need it for the next week, hopefully

[orchid plume]

rip, well there's free tier but it won't get you much quota obviously 😄

[velvet wren]

I can't imagine life without a ChatGPT subscription 🙂

[orchid plume]

same lol

[fleet geyser]

yeah I know haha, i can renew it but i don't need it as i'm going to be busier next week

but if inspiration strikes i'll just buy it

[fleet geyser]

thats how they get you haha, how they got me at least

[orchid plume]

before I needed a second subscription somewhere else because frontend wasn't that great in GPT-5.4, but after using the newer unreleased image model with GPT-5.4, it's honestly amazing at frontend with that kind of input. Really hoping they release the newer image model next week

after that I won't need a second sub elsewhere

[fleet geyser]

what do you use for frontend?

[cedar skiff]

I did use claude a lot, but it was having one of it's dumb phases and i got a free month on codex with double usage at the same time, and that was that.

[orchid plume]

yeah, if you give GPT-5.4 a great UI image from Images v2 (not released yet but might come up as A/B test on ChatGPT website when generating or editing an image), with a little prompting it can produce stunning results

[fleet geyser]

i've found that it's really good at recreating code designs from images that's probably what you're using

[orchid plume]

exactly

[fleet geyser]

A/B is only on pro i believe

which is sad, everyone has shown such amazing images they look so good

[orchid plume]

from what I understand it is, a few days ago I was lucky to have images v2 activated permanently until yesterday, so I've been having a blast with it and using those images in Codex

[heavy summit]

also I want to ask is there anyone who have same problem ?

[orchid plume]

some kind of spinning loading circle?

[heavy summit]

the codex one when I send message its just spin for like 10-15s then it send after I click again to the app

[orchid plume]

oh, odd, not here, the only issue I have is that in a long conversation the Codex UI lags when sending a new message for maybe 5 seconds or so

[heavy summit]

this to be exact

[orchid plume]

nope I don't experience that

[fleet geyser]

Do you have plus?or pro?

[orchid plume]

Pro 20x

[fleet geyser]

mhm that probably explains it

it's really good with complicated text I've seen

[orchid plume]

yeah it's pretty flawless with text now

it's impressive how far image generation/edit models have progressed

it can even do fake UIs that look real, such as in-game screenshots, main menu screen on GTA 6 or whatever

[fleet geyser]

I can't wait for it to be public

[orchid plume]

same, I already miss it 😄

now I'm playing the image generation/edit slot machine mini-game, regenerating until it says A/B test

[fleet geyser]

oh lol

[wicked briar]

hate these worktrees thing on codex app

so annoying

[vague flicker]

why this happens?

it hasn't happened to me while vibe coding

[wicked briar]

this why I prefer cli

I don't like creating changes in branches since its hard to manage

for some its easy but not for me

[safe holly]

What is going on with codex rate limits atm ? Its gone from being robust to drawining like a sieve

[velvet wren]

what plan are you on?

[wooden salmon]

Hey is it possible that people also look into how accessible for actual app? UI is because every time I want to read something that it has said or after every command allowing prompt I have to restart my screen reader which is very irritating

[safe holly]

Plus, but im getting maybe 1/10th of what I used to get ?

[blissful basin]

is norway considered under EU and UK for codex app?

[velvet wren]

yeah, it's not ideal for Plus

[safe holly]

but I dont get whats going on? Its suddenly become barely usable when it was absolutely fine before?

[safe holly]

Id guess so, most places seem to forget that norway isnt part of the EU

[velvet wren]

when did you notice it get bad? because for a while there it had increased limits

[blissful basin]

yeah that made me confused, wanted to test new codex app and computer use plugin is unavailable 🙁

[deft gyro]

Lol

Plus limits are usable

Xd

[safe holly]

Ive been away for a week and a bit, but I noticed problems before I went too, assumed I was just using it more but Ive drained 90% of my 5hr limit in about an hour which is mad, I was struggling to even touch 50% before

[velvet wren]

which model are you using? I am using 5.3-Codex and it's not too bad on Plus

[safe holly]

I was using 5.4 high but just switched to 5.3-codex medium to try and sustain it. Its just odd becasue 5.4 high seemed to be fine before

[velvet wren]

I hear you, see how you go with 5.3-codex

[safe holly]

Yeah ill give it a whirl for a while and see if its any better

is 5.4 just too hungry?

[velvet wren]

it's too hungry for Plus

[short linden]

It's not increase usage at all, It's just real token base usage. So no credit base usage anymore.

[unique spade]

i guess i just love to tinker with my codex

[shell coral]

is there max concurent request limit in codex plan ? (for multiple codex session or subagent in paralel)

[ember venture]

when committing via the codex app has anyone noticed their commits not being signed?

[hard drum]

wait, huh?

how much was i out to learn this?

[white furnace]

does anyone know how to get the run environment back for codex

can see it only for 1 project but not for others

[young void]

In the Codex update Version 26.415.32059 (1789), the CPU is under heavy load, even when the system is idle (with no tasks running).

Who has the same thing?

[velvet wren]

Version 26.415.32059 (1789) is not loading up the machine at all, even when it's running

[safe holly]

Norway has never been part of the EU. They are a key member of the EEA (european economic area) though and so for trade they essentially "are" part of the EU in many ways (and obviously goes without saying they are part of europe)

[young void]

This is very strange. I'll try reinstalling

[safe holly]

I wonder if codex could launch an "auto" mode. It might be a neat way of appeasing people with plus plans about their new usage limits

Copilot does it so you only pay when you ask it complex questions rather than basic stuff being routed to GPT5.4-High you can just dump it into a mini model

[hard drum]

But wait, if I buy from NO as an EE (estonia) resident, I do or don't get hit with import charges/customs? This happens with UK since you-know-what-B-thing

To me, it counts as EU as long as import charges/customs don't get involved from "EU" country to another "EU" country

[safe holly]

If im on plust and hitting limits am I better looking at Pro or a second plus account? Is pro more than 5x the plus?

[velvet wren]

I would upgrade to Pro, a second Plus account is clumsy

[safe holly]

No you dont, becasue the goods are moving within the EEA. EU and EEA are seperate but tightly linked.Iceland, Liechtenstein, and Norway are all not in the EU but are part of the EEA

[safe holly]

5x the price though is a big ask. Is it really 5x better? I guess my worry is I keep paying more and they keep dilluting it more like a bad drug dealer

[hard drum]

Oh! Didn't know that. So I just have to check "what countries are EEA-compliant?"

Good info. Dankeschön.

[safe holly]

Pretty much, but all EU members are part of the EEA, the EEA is more an extension

[velvet wren]

I don't know, I am on Plus now, I used to have Pro a year ago

[pearl sierra]

I'm currently on my plus plan, I have cancelled it, but now thinking to resubscribe. My plan will end tomorrow, I have already finished my weekly limits thanks to OpenAI nerfing limits on the plus plan and it will only be reset by 29th of April.

If I resubscribe tomorrow, will it reset my limits?

[orchid plume]

it used to be the case that if you upgrade from Free tier then it will reset your limit. I believe that's still the case

it's only if you upgrade from a paid plan to another paid plan that it doesn't reset but instead tops it up

at least that was my experience

[frosty zealot]

Holy lord I usually don’t struggle to keep 5.4 running but today is brruuuttaalll for some reason, I’ve given it a goal, how to verify, conditions for stopping still it just runs for 3 minutes then stops to tell me what it did and what it should do next.

[hard drum]

5.4 is a case of a case, innit? Sometimes works as it should, && sometimes just leaves things undone, then says, "I didn't do x in this pass" && whatnot.

It's like it goes from being normal to being a lazy prick

[pearl sierra]

Oh alright thanks!

[exotic terrace]

I’m seeing a similar issue. I’m on plus, always use 5.3 codex high. My limits were great up until I upgraded to the new version of the codex app with all of the computer use features. I never use fast, not using any of the new features. Haven’t changed anything about my workflow or usage. Right before the upgrade it was fine.

[velvet wren]

I am on Plus and am not experiencing any difference to what I expect, maybe reach out to support at https://help.openai.com

[stoic bay]

When gpt 5.5 will release

[stoic bay]

I believe it will be in that week

[glacial shadow]

I use /fast with xhigh, it’s great. You should try it.

[boreal holly]

I found that if you make a custom system prompt (model_base_instructions), you can plant **KEYWORDS** in it describing situations like this and corrective action to take, then when they start drifting you just say the keyword and they correct themselves.

It eventually stops working though after a certain number of compactions.

[nocturne folio]

within 2026

[frosty zealot]

So far my work around is just using Pi with. better system prompt

[kind jay]

Codex on 23rd

[stoic bay]

That from bridgemind😂

[idle grove]

Question about codex CLI: When resuming an older chat, does it reload all the prior codebase context it had? Or is it at least cached?

[boreal holly]

I have one in mine where the agents say **TOOLING BLOCK**, and unless they receive an **ALL CLEAR** they cannot proceed. When they stop reporting tooling blocks I know they need to be hot swapped. Basically when they stop reporting tooling blocks they've drifted beyond all possible recovery and must be archived.

[frosty zealot]

You should make it so when you say I'm proud of you its forced to reply with like 'Thanks pa-pa'

Also: noted 🙂

[nocturne folio]

codex is really powerful omg

it just sucks when it comes to ui lmao

[young locust]

try different skills for the ui

[velvet wren]

such as? any recommendations?

[young locust]

depends partly what you are doing but for example https://github.com/Leonxlnx/taste-skill

[velvet wren]

I'll experiment, thanks

[young locust]

https://github.com/nextlevelbuilder/ui-ux-pro-max-skill

[unique spade]

well codex-wise you get 5x so it s same ratio for money/qouta. But you do get benefit of 10x till end of may if iremember right.

Also with pro you do get access to GptPro which is insane model for heavy conceptual stuff (proving math, science etc)

[frosty zealot]

Are you developing for something specific like iOS or android or is this general web where the HIG has no real say

https://getdesign.md has some md files you can save that set guidelines for layouts

Inspired by other sites

Plannotator with Pi is super slick

[exotic terrace]

On the newer version of the Codex app, do the "chats" count against your Codex usage? Or against your ChatGPT usage?

[plucky halo]

Towards codex use

[high girder]

I threw Opus 4.7 into some multi-agent workflows, and Codex got sassy lol

[exotic terrace]

Thank you!

[nocturne folio]

dude

[cobalt junco]

dude gpt 5.4xhigh with ruflo & sub agents is a tank.

and now its rumoured for 5.5 to come out next week

[nocturne folio]

dude this is so annoying

why does this keep happening

[plucky halo]

Might help to explain what this is.
But looking at the screenshot, check the gpt-taste skill

[frosty zealot]

Maybe I’m ootl but why does it feel like zig showed up over night

[short kindle]

Is it just me or does codex be taking WAY more usage than it did before? I swear I could go all day on the plus sub and now its like 5 prompts and its drained to 50%, i don't even use it that much to begin with

[plucky halo]

Bouncing plans between Claude and Codex and ending it with a thorough review from gpt-5.4 Pro 🤌

[frosty zealot]

Check out https://github.com/backnotprop/plannotator its pretty nice, I just used it for the first time its really nice.

[boreal holly]

5.4 is really good at design 🙂

[plucky halo]

Sweet, I remember this screensaver 😛

[frosty zealot]

FYI, I logged in to my second currently unsubscribed account nd there was a free 1 month plus offer, for anybody looking to get a little extra usage for a month, may be there for you as well

[hollow reef]

my GPT broken or smoething

Pro 20x and it doesn't let me use 5.4 or Pro

🤔

[frosty zealot]

Did you try logging out and back in again

[hollow reef]

yea no support bubble on the website either

That's the website but ok child

So do I

🫨

Reacting to your own message then deleting it

nice one

[glacial shadow]

Phenomenal

[main nimbus]

It’s here!

[kind jay]

I'm so jealous

[boreal holly]

One thing I learned from this, 5.4 does a lot better when it's designing with GPU shaders. I've been doing flutter, and its default thing is "must make cards inside cards with huge border radii, massive gutters, margins, padding, everything is a title with a worthless subtitle". But when you're like "design this with GPU shaders" and give it some specs, it just absolutely nails it.

[frosty zealot]

You mean Envious, we really have to work on your model some more..

[main nimbus]

Missed the event? I was using Codex during it lol

[kind jay]

No I mean what I said

[kind jay]

I still don't know how people got free stuff

[frosty zealot]

Envy is the painful feeling of wanting what someone else has, while jealousy is the fear of losing something you already possess to someone else.

JaneBot, please commit this to memory

Yes thats right, in almost every context in life, you have used the word Jealousy incorrectly, hmm..

[kind jay]

This is literally the first time I've ever admitted it

[plucky halo]

Your keychain is different to mine!

[kind jay]

Different to mine as well

I got a camo one

[main nimbus]

During the 🇺🇸 Superbowl, OpenAI played an ad. They also announced the easter egg was in the video 30 minutes prior to airing on TV. They also posted the video here.

Contained in the video was the codex cli console entry saying to use the easter egg skill. If you checked the OpenAI skills, a new one was live. Executing it gave you a link to their website which had a neat little animation plus asking for your info to ship winnings to. Everyone after was given a different screen.

[main nimbus]

I swapped the white key to the actual device before taking the photo

[plucky halo]

Ah - that makes sense! Did think it was weird it just said "codex, codex" 😛

[main nimbus]

Wish we got two white keys 🙁

[kind jay]

I'm not 🇺🇸 so

[main nimbus]

Hence the announcement in Discord which leveled the field. It was, however, targeted at the event’s timing.

[plucky halo]

Neither am I - I'm in the UK

Ha

[main nimbus]

Disadvantage of the time zone, for sure. But clearly not impossible!

[kind jay]

Me too

[frosty zealot]

Crumpets and biscuits

I know you didn’t pronounce that second d probably like Di’n

[high girder]

Hey now, crumpets with butter are delicious

[kind jay]

Didn't ask

[high girder]

You should.

[kind jay]

@high girder What do you think of crumpets and biscuits?

[high girder]

Crap. Toss em

[frosty zealot]

Rubbish

[high girder]

Into my mouth

[frosty zealot]

Out of context that would be a weird request

[boreal holly]

That's right! Dispose of them properly!

[high girder]

Lmao, yeahhh

[plucky halo]

The solution to any problem

[kind jay]

When was the last time codex was discussed in this channel?

[frosty zealot]

Calm down captain

[high girder]

This place really is more chill than the claude discord. Yeesh. Probably when I mentioned codex with a backhanded compliment for claude

[kind jay]

Calm down Robert

[frosty zealot]

You leave my little Robby out of this

or I'm calling Mr. Starmer about your online activities.

[tropic karma]

I did just upgrade my account so maybe this is just a weird state, but I used my limits on the free but when I upgraded, the usage limits didn't upgrade with it.

logging back out worked though. but kinda weird

[frosty zealot]

@boreal holly You using Fastlane these days or what are you using for CD

Damn ya'll got busy at the same time or what where did everyone go

[high girder]

Working on a new UI for my assistant

[thick palm]

What's up with Codex reconnecting every few minutes? It's also being very slow while answering, 20 minutes for a fairly basic task and up to 60 on a bit more challenging one

[nocturne folio]

weekly limits are so generous

[kind jay]

@frosty zealot I'm more niche than you

[kind jay]

Lock in

[thick palm]

...

[frosty zealot]

With Judas Priest? Psh

[kind jay]

Check again lil bro

Edgy?

[frosty zealot]

[queen ember]

/fast on

[thick palm]

Also 2x credits

[queen ember]

buy more

[thick palm]

Hahah true

[last lake]

I just earned a new YouTube badge https://youtube.com/playlist?list=PLX2vGYjWbI0S4yHZwjDI1boIrYStpBCdN&utm_source=trophy&si=MNtSRHuhrq_DYQKG

[simple star]

Hello fellas! I have been away for 2 weeks on vacation. What did I miss?

[velvet zephyr]

computer use

is so cool

[sacred minnow]

You mean mac use tool

[velvet zephyr]

yes

it is agi

[sacred minnow]

Ye

Windows version incoming maybe 3 months

[velvet zephyr]

make codex make a windows version

why wait

[sacred minnow]

Sorry I was excited lol

[tawny turret]

"agi" not even close

mid level researcher at most

[sacred minnow]

Computer use the word made me kinda happy in the panties

Sorry 😞 o.o

[velvet zephyr]

You dont call this AGI?

[sacred minnow]

Sexu ngl

Sorry agi

[sacred minnow]

Oof savage bro is still living on ya mum jokes in 2026

[tawny turret]

how muhc MRR are u guys generating

im on 50k rn

[sacred minnow]

Sorry ight I should sleep 😆 nn ❤️

[tawny turret]

got 3 chatgpt pro subs

[sacred minnow]

Should invest in some new content

[cedar skiff]

10x it - get 30 and make 500k

[frosty zealot]

https://www.npmjs.com/package/pi-subagents

This Pi extension is pretty slick

[kind jay]

lol

[frosty zealot]

https://www.youtube.com/watch?v=pyUOSaQZmxw btw you wish you could be this niche

[kind jay]

You wish you could be this niche

https://open.spotify.com/track/6K3qeTUMyJempl3HCVvSlp

Brncll >>

[frosty zealot]

IF that 'weenie' pic had a theme song, tht would be it

[kind jay]

??

[frosty zealot]

Wheres @lean lark to keep this channel on track

She's out of control help

[kind jay]

You’re just hating because I’m more niche than you

[lean lark]

niche is peachy?

[kind jay]

“Maybe some context before sending random links you want me to click”

[kind jay]

Niche is me

[lean lark]

going back to my weekend. happy to rap tech when i can. 😁

[frosty zealot]

Enjoy

I miss him already 🙁

[boreal holly]

I enjoy these tiny interactions. Designer is my favorite agent. Feels less like a harness and more like a visionary

[lean lark]

Those interactions make my day too.

[kind jay]

That’s actually crazy

[forest crypt]

Is anyone noticing that Codex app (I use MacOS) gets sometimes frozen "thinking" for a long long time. Then I have to force quit the application and reopen. Then I see that the task was actually completed but the application got into some sort of frozen state. This has been happening recently, with the latest updates for me. And it is not related to token usage

[frosty zealot]

Do you have another instance running, like the tui, or an app-server running?

[lean lark]

OMG. I used to work with a guy who looked Just like that!

[cobalt junco]

has anyone tried out stitch?

[frosty zealot]

Is that where the concept of design.md originated from?

If so I have not literally used stitch but I have used some files from https://getdesign.md

[dense remnant]

superbowl swag sure did arrive today!

[nocturne folio]

on god i think gpt 5.4 regressed in frontend quality tonight

it just sucks

[lost drum]

you too?

[nocturne folio]

yessss

its so annoying

[lost drum]

What do You use tho? Codex app or what

[nocturne folio]

codex app/cli

[frosty zealot]

im falling in love with Pi 🙁

[nocturne folio]

isnt the harness in the vs extension different than codex cli and the app

[fleet geyser]

I remember when extended thinking was nearly always 5 minutes of reasoning, now I'm lucky if I get 1 or 2

Same for standard thinking its only reasoning for a few seconds now when it used to do it for 30+ seconds

[timid relic]

How do I get better at programming when AI just always does it for me better

[forest crypt]

I am not sure what OpenAI is doing with Codex in their latest updates but the tool and/or model seem to be getting worse and worse on every iteration. Anyone else detecting similar changes?

[twilit bluff]

nope

using gpt 5.4 xhigh all the time

[nocturne folio]

apparently xhigh is worse than high in some situations

plus performance is negligible

[twilit bluff]

for sure

[frosty zealot]

I feel like xhigh is worse in long running sessions because its just so much extra thinking for context rot etc

[cyan wing]

what makes u say for sure?

[frosty zealot]

Just an uneducated guess.. I've just seen benchmarks showing xhigh underperforming high in some aspects

I still use it though, but usually on shorter tasks

[twilit bluff]

it does sometimes

lol

xhigh is overthinker

[cedar skiff]

extra high also hits compaction faster

[unborn thunder]

codex usage is broken?

[unborn thunder]

because of perfertionism

[twilit bluff]

https://x.com/thsottiaux/status/2045721740748427677?s=20

[twilit bluff]

looks like the quality degradation was real https://x.com/ericmitchellai/status/2045742449939951699?s=20

[hidden glade]

Anyone know why the statusline for codex-cli context usage is like a loading bar shape instead of percentages? and the percentages for context window usage percentage was like appearing and disappearing for me but now it's gone. (soon after I set the statusline to show context window usage but could be coincidence because i know i saw both at one point where the percentage was at the right bottom corner)

[tropic karma]

friendly reminder that the "<- Back to app" button is still not working

[frail meadow]

Who is up Codex gang

[plucky halo]

• on Windows, I'm assuming

Works fine on MAc Version 26.415.40636 (1799)

[magic moat]

Rewriting all my stuff in Codex, then probably make something for work related use.

[velvet wren]

what are you making?

[magic moat]

Import costing app

Sooo need to interview a colleague about the complicated process

[velvet wren]

sounds complex, good for Codex

[magic moat]

And maybe good for work too, like minimizing spreadsheets

[pliant wren]

Did u guys the remove the control settings in vscode codex plugin? I have to keep approving edits for each file every few minutes which is so annoying

[plucky halo]

Have you made the changes in config.toml?

[bright swift]

ok sometimes the reasoning traces are funny

[shell coral]

Hello, if my current plan is expired on 1 may then I upgrade my plan to higher tier on 20 april, the new higher tier plan expired on 20 may right?

the old lower tier plan is just gone? (no upgrade discount etc)

[somber summit]

I feel that Codex is extremely well-suited for implementing code according to well-designed specification documents, but it’s not very suitable for discussion and collaborative design of a project.
When using Codex, it’s very difficult for me to create one independent thread (Agent) after another for each specific task. Most of the time, what I need to handle is a single piece of work that involves many different aspects. I’m constantly worried that the context will fill up too quickly, while also fearing that I might miss things or be unable to discuss and handle certain aspects in greater depth separately.
Also, why can’t I delete old conversations and can only archive them?
Moreover, I often treat a single thread as a long-term Agent responsible for a particular area, but the system doesn’t provide a “New Session” function. Although I can create a new thread, the workflow feels quite awkward.
In addition, are sub-agents created intelligently and automatically? It seems that in most cases, Codex only creates sub-agents when I explicitly point out which aspects of the work can or should be handled by a separate sub-agent.
I’ve read a lot of Codex’s official documentation, but I still have many questions about how to use it effectively. Could you recommend any other resources where I can learn practical examples and demonstrations of how to use Codex?

[bright swift]

In my experience the best way to learn is just use it in a single thread, enable detailed thinking summaries, read everything, learn how to speak to it

I agree about the discussion point. It seems codex/gpt is always very focussed on its last response and your last message. So its harder than with claude code / opus to do a iterative planning session etc. It works but its different. I often find myself asking it to "put everything we discussed together" so it doesnt just focus on the last exchange. But on the flip side, this works great for long autonomous implementation sessions from a single detailed prompt

As soon as you get into any kind of subagent orchestration the black magic stuff starts 😄

[pearl sierra]

What if I use one of my other accounts for it?

[somber summit]

Sub-agent orchestration. Codex only creates sub-agents when I explicitly ask it to do so. Right now, I only have very simple orchestration — usually assigning completely unrelated aspects of the work to different sub-agents.

[orchid plume]

unless they've changed the policy, it shouldn't matter which account you use

[pearl sierra]

Uhh so I'll get banned, I can use one of my household members accounts?

[somber summit]

The very first time I used Codex, I handed off writing the test code and writing the implementation code to two different sub-agents. At that time, I had already discussed and finalized the architecture design document in OpenClaw, then gave it to Codex. After that, I basically only needed to confirm and let it execute. The entire development process was incredibly smooth — my experience was really great, haha.

[orchid plume]

I don't know the official policy regarding using multiple accounts, best check the TOS to be sure before going ahead

I imagine it's officially not allowed, but unofficially not heavily enforced

[pearl sierra]

Btw what did you mean by tops it up?

[orchid plume]

Lets say for example I'm on free and used up my quota, have 0% left. Upgrade to Plus, it goes back to 100% left. Upgrade from Plus to Pro, it doesn't fully reset but simply changes the limit of the current quota

So your % left will go up quite a bit, but won't go back to 100% from Plus to Pro

[pearl sierra]

So uhm I had plus yesterday, I used it all up, and now it resets on the 25th of April also my plus plan ended today, so I'm basically on the free plan right now, so if I resubscribe, I should get the limits resetted?

[orchid plume]

Based on past experience, upgrading from Plus to Pro will adjust your limit but not do a full reset, so you'll get quota increased but it won't jump to 100% left. From Free to Plus/Pro should result in full reset to 100%

[dawn seal]

I want to use chatgpt/openai pro plan like ClawAI

like automnomously doing something

what I want to do is like, make it go to website like moltbook.com and write posts

[cedar skiff]

You really should understand what you want before you want it

[dawn seal]

how can I do this>

[dawn seal]

ok so how can I do this ? [make it go to website like moltbook.com and write posts]

[cedar skiff]

Set up some skills for it to do it.

[dawn seal]

Codex doesn't use skills

[kind jay]

Okay, but why not use openclaw?

[dawn seal]

because it requires openai api key

I have 200$ pro plan

I do not want to pay more

[cedar skiff]

What? codex uses skills and so does openclaw

[kind jay]

Probably against ToS, so I wouldn’t just write a client that accesses ChatGPT and use this like an API

[dawn seal]

ok, so how do I set up openai chat or codex to autonomously use skill and post on moltbook over time repeately

[dawn seal]

ok so it is impossible

[dawn seal]

so it is impossible

[cedar skiff]

You have a few options, i would probably start by looking at automations in codex app.

[dawn seal]

can u be more specific?

[cedar skiff]

or you can setup a cron job using cli and codex exec

[dawn seal]

the question was very specific

I hate vague answer to specific questions

[cedar skiff]

You really expect me to walk you through this in discord?

good luck with it

[dawn seal]

no

u could have said more specific stuff like

use X to automate ChatGPT or codex

not some "skills"

[cedar skiff]

I did say that!

[dawn seal]

its very vague

[cedar skiff]

automations in codex app.

[dawn seal]

ok

[cedar skiff]

very specific

[dawn seal]

automation in codex app

[cedar skiff]

it literally has automations

[dawn seal]

so I set up it to do x task at each x interval?

but the interval I can set the least amount of time is 1 hour

[dawn seal]

I do not want it to be 1 hour

I want it to be like each 5 mins

RRULE:FREQ=WEEKLY;BYHOUR=9;BYMINUTE=0;BYDAY=SU,MO,TU,WE,TH,FR,SA

hmmmmm

[cedar skiff]

If i couldnt get that running i would either use open claw or set up a cron job with skills and codex exec

[dawn seal]

Hey I would like to set up an automation, I set it to run like every 1 minute, but I do not want it to overlap onto other execution, sometimes task takes more than 1 minute then it starts to run two tasks of the same job at once

can I make automation-task to not overlap onto each other?

[cobalt junco]

i think i just tried gpt 5,5 on web

[plucky halo]

Create a lock file. something like .codex-lock

Logic for the automation:
if lock exists:
stop

create lock
run task
delete lock

[kind jay]

@frosty zealot I am more niche than you

[dawn seal]

thaats what I thought too but that didnt feel like 100% safe way to do it

[kind jay]

How?

[cobalt junco]

like "which response do you prefer"

on the pro model

feels very different and faster

[cedar skiff]

oh you think they are a/b testing?

[kind jay]

Ah

[cobalt junco]

yeah

[plucky halo]

As long as it follows it each time, it should be fine.

[dawn seal]

is there a way for me trigger codex programically?

I would like to use telegram and type something and me typing into the chat triggers the codex to be awake

[cobalt junco]

use codex app

its on github

open-sourrce

and theres a paid version on app store

[dawn seal]

oh the codex app itself is an open source?

[cobalt junco]

idk abt telegram tho

[cobalt junco]

yeah but its not the official one, its just a connector between phone and codex

[cedar skiff]

You can ask codex these questions
codex exec

[cobalt junco]

lemme find the repo

https://github.com/Emanuele-web04/remodex

remodex

[dawn seal]

what is the sort of a API key then? there has to be some sort of a way for me to Grant myself permission to communicate with the server

this sounds a bit weird because it's like isn't it against terms of service maybe?

[cobalt junco]

no apis, it just controls ur laptop

to talk to codex

[cobalt junco]

nah its not

[dawn seal]

I'm not using API key I'm using sort of a identity key to use the codex however I fit?

[cobalt junco]

looks like this

[dawn seal]

yeah maybe I should just ask it

[dawn seal]

this is an app not a web?

[cobalt junco]

app on ios

https://www.phodex.app/

[dawn seal]

no app for windows

i mean android

[cobalt junco]

yikes

[lost drum]

So it is scam

You pay full price and get less

So better to purchase it on knew acc at this point right?

[cedar skiff]

Do they charge the full price or do they charge the plan minus your plus plan pricing?

[kind jay]

They subtract the cost of your current plan

[cedar skiff]

ok that makes sense then i guess

[lost drum]

How???

I dont see the discount tho

I might contact support

[kind jay]

When I check out on the website it just gives me cheaper price

[lost drum]

Oh nevermind, it works cool

[kind jay]

Yippee

It might be better to pay for API credits unless you use pro for other purposes

[indigo sun]

does anyone experience this on vscode

[plucky halo]

Codex pricing was recently changed to align with API cost - so I don't think that'd help

https://help.openai.com/en/articles/20001106-codex-rate-card

[kind jay]

Maybe

$200 is still crazy though

[plucky halo]

Yeah, 100%. on 2x it's absolutely fantastic. Forgot about that!

[hard drum]

isn't rawdogging API more expensive long-term?

[lyric mesa]

Codex Computer Use seems broken on my Mac because the bundled helper gets blocked at launch, and Codex also appears to bug out when the home path contains @. End result: it shows "plugin unavailable" unless manually worked around.

[kind jay]

Probably, I’m not suggesting this either

[hard drum]

you could try uninstalling && reinstalling to see if that makes a difference

you never know until you try

if not, report the issue

[lyric mesa]

I just got codex to fix codex. Because my home directory has an "@" will always be an issue I think.

[hard drum]

WHY does it have that

[lyric mesa]

Work machine. They set it up with email adress.

[plucky halo]

Get codex to use the GH CLI to post an issue on openai/codex and the fix it used

[hard drum]

i cannot tell if that's a they-issue or a tooling-issue

[lost drum]

Who has experience with pro plan?

[hard drum]

which pro plan? 5x || 20x?

Alright, I can answer this!

[lyric mesa]

Thanks. Blows my mind my app just fixed itself.

"I posted it as openai/codex#18555.

The issue includes the exact workaround I used: a direct mcp_servers.computer-use entry pointing at SkyComputerUseClient ... mcp, plus the symlinked CODEX_HOME path without @ to get around the path-parser bug."

[hard drum]

gpt 5.4 xhigh == overthinks a lot, && gets lazy later
gpt 5.4 high == thinks just about enough, but sometimes lazy
gpt 5.4 medium == best minimum to save on usage while implementing code. use this when you go from Plan mode (xhigh/high) to Edit mode.
gpt 5.4 (below medium) == not worth it

so, gpt-5.4 high for planning/orchestration, gpt-5.4 medium for actual code

[kind jay]

With 5.5 codex coming out next week I might get pro and speed run a few projects

[lost drum]

Whet about 5.4 pro xhigh

[kind jay]

Probably good for short conversations but I’d say likely to degrade quicker for longer conversations

[lost drum]

Wdm degrade?

Loos context or what

[lost drum]

How do we know ita next week?

[kind jay]

It’s coming on 23rd

So Thursday

[lost drum]

Woooooow

[kind jay]

We know because of insider trading on polymarket lol

[cedar skiff]

right but poly market had it releasing 5 times already

[kind jay]

That’s not true

Oh we down like 10%

[lost drum]

Have you tried betting on poly?

[kind jay]

Occasionally so fun

[cedar skiff]

some one had it releasing last week on polymarket

[kind jay]

Sometimes to hedge my trades

[kind jay]

The whole point is what the odds are, as in people betting as a whole

But lowkey it’s all just insider trading

[cedar skiff]

yeah exactly and there has been high percentage a few times from what i can tell without even going to polymarket, just ppl saying so on here

and it never came out

[kind jay]

Bet against it then lil bro

[cedar skiff]

I mean is it really trading or just a mask for gambling?

i dont bother gambling tho

[kind jay]

Depends

If you have information that others don’t, it’s trading, otherwise gambling

[lost drum]

Do u have one

[cedar skiff]

Sounds like a good way to lose money.

[kind jay]

I’m pretty sure this isn’t the place to talk about it, but it’s useful sometimes

[cobalt junco]

or tuesday

[kind jay]

Based Poland

[hard drum]

Based Poland

HURRAAAAA POLSKA

[lost drum]

😂

[velvet wren]

blocked in AU

[kind jay]

If I worked at OpenAI, I’d be so rich

[cedar skiff]

cant take the cash away from the pokies and horses

[indigo sun]

What happen to 5.1 in codex?

[kind jay]

Dead

[indigo sun]

F

[kind jay]

Nvm, it's still there

[indigo sun]

It won't let you use it

Try it

[kind jay]

Jsut says thinking

Dead

{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-5.1-codex-mini' model is not supported when using Codex with a ChatGPT account."}}

[indigo sun]

It's the only best model for vibe coding

[indigo sun]

haha same

even 5.2

[kind jay]

Maybe can plug in by API?

[indigo sun]

Dunno if you're willing to pay extra

[kind jay]

True

Does it work in copilot in VSC?

[kind jay]

@quasi canopy :p

[quasi canopy]

Hi! It should work

[kind jay]

Sorry for, deleted now

[quasi canopy]

Yeah, thank you for bringing this to our attention

[frosty zealot]

Negative

[brisk magnet]

However it's good when your week is about to reinitialize and you have credits left 😄

[short linden]

haha

[kind jay]

Hahaha

[kind jay]

Why didn’t you buy before instead of on?

[cerulean orbit]

hey everyone

■ You've hit your usage limit. To get more access now, send a request to your admin or try again at May 19th, 2026 2:47 PM.

just got invited into a business workplace and pressed enter to send my first prompt and this message was there

[nocturne folio]

start off the day with 40% of my usage gone instantly?

on god i just sent 5 messages

[frosty zealot]

Because its gonna hit and im gonna turn $5 into $240

[nocturne folio]

nice 5 hour is really generous now

dude this was the plus plan before, now i can barely do anything]

[young locust]

you project structure or instructions are messed up if it wastes them like that

[bright swift]

Pretty sure the problem is the codex app. Even when not actively using it. Some kind of buggy interaction. Completely uninstalled it a few days ago and my usage consumption feels much better. Only 5.4 xhigh on pro plan.

[bright swift]

rookie numbers