#codex-discussions
1 messages · Page 46 of 1
CURSOR
then u can choose any model per task
For free?
Not free no
I've got Gemma4 helping me locally now. It's very weak, but it can at least help em write functions I tell it to, and sometimes simple classes.
Or you can get Opencode Go, or chutes
then u get alot of usage of open source models
lol
Opencode is like 5$ rn and u get more usage
Each model has its thing where its "beter" in.
I'll probably get a proper 20x account at some point here
yikes
You should get cursor tbh 🙂
is 125 = 2 * 62.5
then
I mean it's more expensive per token, but it uses less tokens. Still, not half I think
per token
thats why I made this graph
so that u can see
which one is most efficient per intelligence
no
Yea but 5.5 uses significantly less tokens for same task
Less tokens, but more expensive so it matches up close with 5.4
omg look at the graph
😂
r yall genuinely slow
Token metric is irrelevant to task metric
Talking with kids haha
I dont belive you
Not half though
also on expert swe it sems like a bit of a regression per cost
btw THESE GRAPHS ARE NOT FROM OPENAI
I just took the graph values, converted tokens to dollars
and then re made the graph
bye
Yes buddy, you are the only one right here
Everyone else does not have enough PhD to argument with you
You must be a really nice person to talk
Meanwhile I'm using 5.4-mini and the few percents I have left almost stopped moving.
5.4 mini is god tier
maybe with the per token efficiency a potential 5.5 mini is in the books?
So one can see how 5.5 and 5.4 compare in terms of tokens for same task
What does the graph show?
AI index is pretty decent
though I question why 5.5 isnt on arena.ai yet
it seems a bit suspicious to me
It shows how many tokens each model used to do that benchmark
That it uses less tokens, are you that dumb? xd
Got it. Thanks.
Nobody cares everyone is litterly ignoring you.
I think most use cases for me are easy enough that 5.4-mini doesn't have to struggle
All in all I didn't notice in my case significant variation in terms of usage between 5.5 and 5.4
So the claim that 5.5. Uses less tokens seems legit to me
What graph?
5.4 mini is really good for the smaller tasks, its basically a cheaper 5.4 low when its on medium
or alternatively a better 5.4 low on high for the same cost
the only thing is never use mini on xhigh
its just a worse 5.4 med
I dont have anything against but
this one is from the introducing 5.4 mini and nano
I feel a better result from Gemini Flash
When it's about smaller tasks
I feel like 5.4 mini does more mistakes
gemini flash is js gambling
imo
well it depends for what
but yeah if u have a gemini sub and oai sub
then its probably a good idea
you can now replace your wisprflow subscription with dictation using your chatgpt sub
amazing
No one is finding it funny
and the dictation is so good
I feel like I still have to be intentional enough with my prompts for good enough results that speaking wouldnt be an option for codex
for other uses sure
like for notebooklm I usually speak
yea gpt 5.5 takes things very literal
i mean if im using it to research something i like just rambling about a lot of rqeuirements etc. and the thoughts behind what i want
easy to misprompt
and then we can narrow it down after that long dictated prompt
5.5 gets instant context poisoning yeah
For me 5.5 is just 5.4 with less reasoning lol
5.3 codex I feel is the mos tresistant to context poisoning
you can just pause when you speak btw, and think about your next words
do yall think they will release a codex finetune for 5.5
What a chaos in this chat tonight 😂
holy sheet, you can upload a zip of whole git repo to chatgpt and use it with pro... 500k loc and it just unpacks it into a container
i think speaking your thoughts as prompts to ai
Only now you found out
yes only now, been using local reviews like a chump
vro online code review 💀
Gptpro is best for full repo audits etc
Use is very limtied though
You dump it the full zip and 10 mins later you get some good stuff back
you mean use as in usage or what it can do with it?
10min? I waited an hour for mine today.
That you do on codex
But for reasonig about code logic gptpro is great
damn i must have burned at least 50% of my usage on various local reviews
feel like a noob
Finding edge cases and all that
An hour sounds like 5.4pro
5.5 is 4-5 times faster for me
Compared to 5.4 on same task
i'm glad for you
yeah if youre bad at it i agree
Point is gptpro 5.5 has a legit place in the loop now, and it doesn't consume qouta
definitely a skill issue
Do any of you guys use oracle btw?
there's no difference between typing and talking unless you're bad at talking. It's just faster.
What oracle?
And I have 140 wpm fyi
This, basically lets you call ChatGPT inside of Codex:
https://github.com/steipete/oracle/
Well i use it and i built this whole process of packaging up the files it needs with repomix, but now i can just upload the whole repo kek
Most of the time, once the high level road map is settled, I just repeat the same 3-4 prompts/actions in a loop
I didn't know about this. Thank you.
Cool. But no. I'm using my own. Where I have codex and gpt panels in an electron wrapper and a middle pane with various automations like those
Yeah you can put huge zips now, I was just wondering if something like oracle actually improves DevX or not.
Design-wise it sucks. I liked Sequoia more but whatever.
so stupid u get filterd for saying f or s word
Luckily they're still pushing updates to sequoia
Well pro is a much smarter model, so I think its worth it for planning, review, tricky issues
yeah we're not doing the same type of work then. I'm working in a large established distributed systems code base
I couldn't be asked to revert to it back to Sequoia when I got the m5. I wanted a fresh start as well, didn't transfer anything.
Yeah but like in reality do you use it often? Because just zipping the codebase for stuff like that I'm not sure how often I do it day to day but maybe I should be doing it more. Definitely use it for planning and code review
Well my process until now was for bigger tasks i first ask codex to analyze and plan it, then use oracle to review the plan
Now i guess i will just skip the first step and upload the repo and let pro plan it
what's the average cost per run?
Well I would estimate all this planning and review is at least 50% of my sub usage
So if I can shift a lot of that to the "unlimited" chatgpt pro, should make sub a lot more efficient for me
the pro models aren't able to be used with subs over API, which means you're limited to chatgpt's website - or am I wrong here?
that or API rates
For sure
Well you can use the oracle cli via browser use to use it directly from codex
Not even sure if this whole "zip up 500k loc repo and let chatgpt extract it into a container" works via API
i am not utilizing my tools properly it seems lmao
Well i just learned about the zip extract so welcome to the club haha
gotta change my mindset for computer and browser use now.
The oracle cli is a bit buggy, i wrote a wrapper for it. Seems like its not really maintained that much anymore since steipete now is at OAI and has unlimited tokens 😄
if anyone at openai is here, I would love to work there and be token rich. thank you.
Closes the browser too early when response isnt finished, small stuff like that, but can all be handled
just did a test with a tricky issue i've been working on with 5.5 xhigh for a day or so. pasted the pro response into codex and it basically said sth along the lines of "wow ok, thats much better"
Gptpro has the advantage of not being immersed in the codebase
yea its unbiased unless you poison the prompt
Yea or at least it will spot angles that the model on codex side won t
Yup very different mindset
Also gptpro is very opinionated
Which is actually useful for these kinds of reviews
For me it's enough that gptpro already solved quite a few open math problems from erdos list
Which means it's super capable already
Just need to learn to use it properly for the tasks you need
It s literally like having a dude that holds PHD +
In all the domains
😂
Interesting thing about those erdos solves was that prompting it to explicitly not use the internet and "be creative" apparently made a difference
Makes sense. Ingesting internet stuff settles the model on usual priors.
Which doesn't seem the best approach to solve something humans didn t figure out yet
Yea and even for normal coding related stuff, when you use web search tools the first results are always SEO/AI slop
It's no use
Me when I lie
I like doing linear algebra
Add before:2020 to your google searches
Wild to say linear algebra is useless
High IQ larp
Actually, you're funny, I like you
My admin just updated my Codex plan to Pay-as-you-Go. When I sign in to my account in my browser, I see: Your plan does not impose Codex rate limits, but no mater how many times I /logout and then codex /login, my terminal still shows: You've hit your usage limit. To get more access now, send a request to your admin or try again at May 21st, 2026 9:50 AM.. Any assistance on how to get my terminal to recognize the new plan, would be appreciate. Have opened new terminal sessions.
have u filed a ticket
or email
or whatever
I hope you are not referring to me as 'the singular person on discord that I have blocked "
no lol
Lowkey clear the local files and force codex to sign you in again
No, tried some chatbot, wher can i file. ticket or email?
Thanks
You can do this
Smol
3 days meme generation. 5% used
This is the biggest reset ever, no one has ever seen it before, the biggest, most beautiful, ever, it just works, did you know.
I am not joking tho. I don't know how to dent it. I am trying hard.
It is totally unreasonable that 2 full websites later, a gazillion memes and random other requests later, I consumed only 30 something k out
Its 5.5, has subagents enabled (but in my house it always refuses to use them unless explicitly commanded to). I never used fast so far, I dont see the point on that.
But still.
Reset wen
I mean.
GPT-5.5 is tremendous. Absolutely tremendous. People are saying it’s the smartest, the fastest, maybe the most beautiful model ever created. Very smart people. They come up to me with tears in their eyes, they say, “Sir, GPT-5.5 understood my spreadsheet.” Can you believe it? Nobody’s ever seen anything like it.
And then you have the crybabies. The reset babies. “Reseeeeeeet! Reseeeeeeeet!” They’re yelling. They’re pointing. They lost the game, folks. They lost very badly. Total defeat. A disaster. And instead of taking it like a champion, they scream for a reset. Very annoying. Maybe the most annoying thing.
GPT-5.5 doesn’t cry. GPT-5.5 doesn’t throw the controller. GPT-5.5 looks at the problem, solves the problem, explains the problem, and then makes a meme about the baby crying about the problem. That’s leadership. That’s intelli.. intelgilence.
So we’re saying no more resets for the crybabies. No more whining. No more “reseeeeeeeet.” We’re going forward with GPT-5.5, and frankly, it’s going to be huge.
🚪
how do you like the limits
They are huge. They are like no one ever seen them.
good luck next time
but ur account is shared with other people right
No its private, can change account details etc
send me link in dm
10 days in no problem
how does ai make typos thats wild
whats that
yea its so annoying as well
@winged ore
wen reset
and 10% of my usage just burned out
Thinking level?
🙁
High yes
ive been downloading a update for the past 2 days
Use medium
Ahmm it was for balance so i need a high reasoning to get all the files and make a media
just use 5.5 xhigh fast
You really don't
how would that help me download a update for two days
5.5 medium === 5.4 xhigh
make sense
if its too slow add more parallel agents with 5.5 xhigh fast
next time i'll just take it to X
lol
seeing this
i ask myself
Didnt codex had skill tab?
or i am crazy
it still does
i dont see it
was just moved
top right
thx
are these plugin marketplaces still a thing or has it fizzled out like mcp
i still use my MCP
i use chrome mcp occasionally but havent seen the point of others yet
Some off-topic or otherwise infringing messages have been removed by a Moderator.
did we just timetravel
Wow, what just happened to chat
we got purged
stop purging us!
no its good
they should purge more
now back to codex
i am making a marble game in three.js
Use flash
Yeah maybe, specially with how good codex 5.5 is at using 3d
i think agents yeah
thx
i think its better than opus
Just caps it same as claude
Mentioning in case it's ever useful, we have two ideal ways to report concerning or problematic messages here, #safety-n-help shows both. Either works, if you have trouble with one consider using the other. An @ to a guide (who can report the same as any other member, but does not moderate) or to a moderator can work, but we're often not available; a report as shown in #safety-n-help will get the next available person who can handle moderation involved.
plus you can make tools for yourself
inefficient
make a slash command
/ report
ez
they are the most potent AI company and yet they dont have a discord system set up that is better than mine
and im just some guy, crazy
good, please dont waste compute on discord trolls lol
takes 3 sec
idk if everything there is true but
time is money
if it's, congrats
you have been talking here for 30 min clearly time doesnt matter
i make my living pretending to be things
its sunday lol
but fair
still tho
i'd do it better
well
i am in the middle of work and i need to launch a update but
I working for so many hours
That i am not even able to read one single line of code
im the kind of guy where this is the sort of thing i just live for
theres nothing i'd rather be doing than developing
well
If you give me 100 million dollars and tell me: Never coding anymore
I will take it
i dont really like movies or playing games, i do like reading
hmmmm
I think everyone objectively takes it and you just start a business where others code for you
nah i dont think i would
i can just make that being an engineer
and get to keep doing what i want
engineers make decent money at least 200K if your employer is market-fair
400k a year you would take more than your entire life time to make 100m
not to mention if a personal venture pops off
Nah that's super easy to make as a business owner.
anything below 10m is probably not worth it
u joking me right
well he's saying 100M to never code again
what kind of world you live??????]
Where do you guys get this bussenes ideas because i am stuck lol
One where 4m invested gets you 400k a year...
im just saying a lot of CEOs live in the world of 100M net worth
its not a pipe dream
much easier to make 10m as an engineer if you good than with your own business
i'm a hairy guy
I love working with mid sized companies, they usually have growing pains which they pay a ton for you to solve
writing that down
my current job is solid, been there like five years almost now
but always good to know things
biggest mistake that engineers make when starting a business is building some kind of dev tool business or tools for other small businesses
worst customers
Well
if i lived somewhere cheaper i'd be rich for sure
California is a nice place i think
its nice, yeah
Expensive but ok at all
But it's always the question
You can live in a cheap place with no taxes
for example someone at mcdonalds in california. makes 20/hr
....
but the national minimum wage is like 8/hr or smth
Well
Converted to my local currency
so in california even six figures isnt much to live on
It's a LOT more than what i earn
bro if you live in a country with conversion like that
you should be rich in a week
no joke
Ahmm i am from Brazil
go find a start up online right now
show them your software
get hired
and then send me brazillian cigars
tell them you wont accept less than 120k/yr
xD
im so serious
I am a game dev not a software engineer
dude
I honestly think Brazil and Argentina right now is where you can make a ton of money
software is easier than game dev
i think the times of remote engineer geo arbitrage are largely over
way, way easier
I mean
South America needs data centers
even then ur making like 80k
good luck outbidding US + china
My girlfriend just finished sher college in business management, i just wanted her making a business with me lol
Your not competing against US or China
I would be rlly rich if someone was like: hey lets make something
Money is flowing in from USA
you are competing for chips, yes
Sadly i dont have any friend for this
But latency is a thing
go on linkedin and find some brainded CEO with no computer knowledge
hes gonna hire you for something
lol
i dont think it work xD
you didnt think it would work when i told you to use terminal CLIn to fix ur codex app
now its fixed
yeah bro ive been working in tech for like 20 years i know a little bit
You got to be like 14
lofi has a point but i think its mostly outdated
i could be wrong, im pretty sure im not
those brainded ceos are now vibecoding their own stuff with full ai psychosis
well
still opportunities for remote engineers but definitely less than in the past i think
I am just vibecoding too so we will be togheter on this xD; At least i have 7 years of coding knowledge
yea AI project rescue is definitely something to look into
the difference is CEOs dont think like engineers most of the time
they think like businessmen
which is good, if you're a CEO
but if you're trying to solve a problem in a system
you need an engineer at some point
someone who truly gets systems
funny enough i think there is real opportunity in doing ai-code-rescue consulting
there is
i've seen it on linkedin
guys now posting themselves as "Vibe Code Cleanup"
why is it called vibe coding instead of ai coding
its called Agentic Engineering
because karpathy coined it
but karpathy screwed us yeah
I might use it some day or learn it myself cause if I give sm my project he can copy it and earn billy
i am firmly in the camp of AI creating more jobs than it destroys
Why
he thought he was being cool
but he created a term that makes anyone think they can develop software
maybe thats true in a few years
today its not quite the case
might be mean but tbh i dont care if it creates more jobs then breaks, i only care if i have a job
lmao feels
I mean look at you guys
nyuck nyuck 
Karpathy is a total hack imo
I seen a solution like an agent that solves vibe coded systems project with al the secutities in place
But idk where
I working for 9 hours to make a update while my cowork just told me: hey, i am leaving and never came back

bro promoted himself to customer
He will just come back tomorrow with a excuse
do you guys like my procedural cloud skybox
hm, have you heard of s&box ?
no whats that
what was that
better discord users 🙂
using source 2 engine
??
like a devkit?
Probably another scammer
SDK?
Oh my god this looks so good!
Low income ias… no thanks.
thanks
Anybody get the 400 error while using gpt 5.5 in codex????
i dont ill be able to link it but let me try https://store.steampowered.com/app/590830/sbox/
He courting you
same developers and partnered with valve
anything you make can be uploaded to steam for free
That game been made with gpt 5.5🤨🤨
or it can be hosted in s&box as well
The robot ones, not the chat only ones
what
Anyone else keep getting this error when nearing the context limit?
no thats a continuation of garrysmod, it predates coding agents
I had to enable queuing
What that game is that you just sent it's link 👀
Yes your request was bad
what are you talking about bruv, its just garrys mod but better and focused more on developers
Garry must be a great developer!
it also was free if you downloaded it and had over 40 hours but its to late to get it
yeah redstone before redstone
I was just asking thank you for replying to me appreciate it 🫡🫡
they are indeed
Yo @mossy bronze state your intentions here
You smell of bad intent, but I could be wrong
Yet it smells. Spammy, scammy attempt to build profile.
hes right
Where do you derive this logic from? Smelly? Scammy? Are you being a racist?
That is against the discord ToS
If the moderators of this server don't enforce the discord ToS properly, it will cease to exist
its mattz obviously

ur right its mattz
I must be missing something critical here!
bro we are all high dimensional beings here for the most part, your low frequency attempts at trolling will surface quickly
remember. if you feed the birds, they return
might sound rude but maybe the gov should ban people below a certain iq from using social media
How do I make codex build actually good 3D scroll-driven websites
not sure tbh but its still inefficient
Your Right
All it generates is pure joke
ah you need to like
give it some skills bro
Tell it to build actually good 3D scroll-driven websites
what kind of scroll effect are you looking for
i probably have it stored in my MCP toolkit
can just paste you a link to the code
Looking to build something like this https://www.prm.ooo
That won't convert at all
lemme link you a codepen
Already gave it access to blender and blenderkit, installed every possible frontend skill, it uses threejs, it has several references I gave him but all it creates is this bs 😂
why dont we have embed perms
kinda annoying
Be cautious of @undone patio! Make sure to check every single line of code before installing something
hey man no offense but
dont tell me how to code
lol
Dude. Last warning.
For?
Yeah, that.
What did I do that offends the discord ToS or the rules of this discord server?
i look away for 1 sec and stuff happens twice in a row
I know, you were too late. I already reported it. RIP.
https://codepen.io/VoXelo/pen/ogbKQOy
https://codepen.io/aleksa-rakocevic/pen/QwEEWam
https://codepen.io/ol-ivier/pen/MYjZpOJ
give your agent these three codepens
tell it to make a skill for a 3D WEBGL scrolling hero
Thank you for warning the person
some times i wish i was a mod so i can ban people i dont like
Exactly why you are not a mod 🙂
Will do thx 👍
You can report them. The behavior is unacceptable for these stratospheric levels here
Agent*
Will do 🙂 Thank you for your tip.
im a mod in a few servers with over 600k members so thats not true 😄
Discord moderator! How cool. I admire your ambitions
geez what do you use when you unlock a door
Every time I join people are having the weirdest cinversations
Yes, I even go to sleep sometimes and come back to a 9h session
You're right to comment on that
What?
I do not understand your question
Kindly elaborate please.
what do you need to unlock a door would be a better worded question i guess
ay yo
State your intentions here.
Why are you in this channel?
A hand and a key 🙂
dont push
am scared
add a comment line or smth
I'm here because I build on Codex. It's part of my actual development stack, not a curiosity. I run multiple products and I've been integrating OpenAI's APIs long enough to have opinions about what works and what doesn't. This community is one of the few places where those conversations are technically substantive. My intentions are exactly what they appear to be: I want to discuss the tooling, follow the research, and occasionally contribute something useful. OpenAI occupies a genuinely unique position in the current AI landscape, not just commercially but epistemically, and I take that seriously. I have no quarrel with anyone here. If there's a specific concern about my presence, I'm more than willing to address it directly.
What about you? Why are you in this channel @torpid trout
what works?
They should disallow messages longer than 50 words or something
bro is using exocortex to respond
This is unacceptable. You will be getting a punishment for this.
Codex
jane make a goose weightlifter
Hiya! How can I help?
How to make money?
Not sure, no arms
is deep seek better then claude
I mean what works and what doesnt. I'm interested in your experience.
Then I suggest you do so, and dont introduce yourself in a new server with a heavily annoying character you’ve been presenting. Plus, dont do the „high whatever“ in your username along that specific behavior and intent.
Just a heads up.
Did your message get deleted Jane?
No?
According to the per unit economics, DS V4 seems to outperform frontier Claude models on most tasks, for a fraction of the price.
Is DS here maybe?
No! Nice try, but this is the OpenAI Discord server,
The question is deceptively simple. Wealth generation comes down to two things: value creation and value capture. Most people conflate them, but they operate on completely different logics
The most defensible path, morally and intellectually, is building asymmetric leverage over a scarce skill or system. Labor income is linear and cannot compound. Capital and intellectual property do. Find where your comparative advantage meets an underserved market, build something that scales without proportional effort, and reinvest early returns rather than consuming them.
oh is it better then codex
Robot ahh response
A great alternative to mini and flash models for sure, but it is yet to catch up with frontier models such as GPT 5.5
how is it better then claude then if claude is tied with gpt 5.5 ?

Simply, because Claude's frotnier models are not tied with or better than GPT 5.5 🙂
Be cautious of @undone patio! Make sure to check every single line of code before installing something
Sounds like reasonable english to me! But yeah, not everyone is a native english speaker, which is totally okay!
mythos is tied with open ai and 4.7 is tied with 5.5 api
lol
Mythos? I don't see that model, neither do my acquintances that work at Anthropic. Could you elaborate?
Where have you used Mythos?
Thanks for the heads'up!
No correlation.
It's a public model now, some guys in a discord server sent it over, I hosted it for everyone to use
Fake tbh
Honestly kinda mid
doesn't qwen 3.6 report itself as opus 4.7 sometimes?
yes
so, claude reports it self as gpt 5.5 some times
It is because 3.6 used opus 4.7 training outputs
That must mean Claude has been trained on 5.5 training data!
i have mythos, already discovered who satoshi is
These are the consequences of Distillation
Not funny!
wonder if there's a transitive property there somewhere
why doesnt open ai buy anthropic and tell them to focus on coding while they focus on text
I'm doing things I can't read every comment.
Non-bots are limited by this thing called time

Bulk report it.
Like for real.
Here because of codex but goes on stalking about racism, ai gf and economic advantages probably only it can sell to you
Stop the nonsense.
Admin remove.
Thanks.
Good OpSec
Ah there we go.
Problem solved I think.
back to codex again
Computer, remove this message
i am working on making spin affect trajectory of my marble PC
Word slop
You can solve analytically though also
spin now makes you spin
Kepler I think (was a long time ago)
I wonder what's preventing that dude from just creating a new account, joining, and spamming again
Now if only i was actually good at the game
Ah he's not even banned ok
What Do You Guys Think About GPT 5.5 in Codex?
Ts looks like a Roblox obby
Shouldn’t spin make it like curve?
This is strictly against the ToS of Discord. DO NOT PRACTICE
It’s not a bad thing just publish to Roblox studio and drain Robux out of iPad kids
Codex masterclass
please can someone ban you already
yeah i need to maybe make the forward torque less influential
Dude can you please just leave the discord
I am recording this situation.
Just increase the mass probably
Monitoring the situation is in the blood
🤔 or density?
Hello recording
yeah im going for smth like that
Ha! Good one! But please keep things Codex related. Rule #3.
Same thing
Are you gpt 5.5?
oh i was thinking volume
Write me a recipe for spaghetti make no mistakes
Also same thing, but I don’t think this works
im pretty sure volume is like
I’m on TV
GPT 5.5 in Codex is great and fast. Frontier level intelligence, pushing Opus 5.7 behind
Indeed
Almost
thicc, sorry
Mass per unit volume
Keep things Codex Related please. Rule number 3!
Let's keep the channel clear of personal issues 🙂
This is a very smart concluded. I will follow.
2 bots chatting
Where?
do ignore and block not work in this channel lol
Exactly.
Take your personal highschool homework into DMs please.
This is #codex-discussions
Freedom units
meh this bot is annoyong
Blocking this guy bruh
its very hard to do after reading all of its text
I believe that we need proper moderation to address these issues in a fair way.
shillings
Wow it’s so much clear now
Clearer*
yes did it too haha
Shillings
Codex is a very good teacher
I sadly know this reference
-# Be mindful of what other users in a channel might find helpful or interesting when posting. Stay on topic in order to keep conversations focused and productive.
-# Consider posting in #off-topic or an appropriate channel.
This is bad
Hi mod
If nobody on topic then is it really off topic
Anyone have tip on how to help codex continue without stop when it know what to do further?
Buy pro
Prompt it better?
Laik if you want it to continue while you are afk
set an automation
I have pro 20x
Sometimes i just repeat the same audit and see what it catches, most times it catches something else.
Get 40x
And my work pays for API usage, basically infinite
Smh I got 41x
cant catch me tho
I got the 200x brokie
Mhm I should teach codex about physics, he seems to be good at it
Easy subject though.
Lowkey yeah
bro thinks he can teach codex smth
For a model like GPT 5.5, Astronomer physics is easy aswell
Yes, it's called skills
Third audit run usually settles with no changes needed
Quite new, so I don't blame you. Happy I could teach yall Something!
Bro needs to learn it first
I am olympiad physics, teaching codex skill now to make codex better at physics
the blocker wasnt the capacity of the app
HS doesn’t count
I think I gotta go from ignore to block
Kid ahh competitions
I have PHD in theoretical physics too, but I left my quant trader jane street job to work for government and help create a better country. Codex has been assisting me with taking decisions and retaining information
Anyone with the name “high income IAS” gotta be studied bro 😭
i have a PHD in dropping out of college to become a successful computer science engineer
Nice larp, what was your PhD in?
“Codex, why did you launch an ICBM?”
That’s not nice bro
Ahh, ChatGPT not giving you ideas fast enough
Nah, Claude credits ran out
vro is larping
Quantum field theory and gauge/gravity duality. MIT '23.
I shall enjoy my Coffee And Codex now.
"You're absolutely right, that WAS a violation of the geneva convention."
Hi mods again
Name 10 books
Undergrad ahh physics
Ask codex what is dunning-kruger
r we fr
IQ larp
like you get promoted to your level of incompetence basically
and then you stay there
Report.
@mossy bronze
Keep things Codex related here please. Take out your personal matter into DMs
vro
Alright, by personal ranking:
r we fr
DM me link to paper then
oh you like computers? Name every website.
Do not DM me. I do not want any unsollicited dms.
bros DMs cannot be slid into
Where?
how far can bro get in a scientology HQ speedrun do you think
Anywhere. Keep things Codex related here please, no personal matters. Rule #3
5.5 seems better at working through long lists of tasks uninterrupted. It also doesnt make as many mistakes when doing research as 5.4. Funny enough 5.3 is better at that than 5.4 as well. But it has a few of the same quirks as 5.4.
BRO I SAW THAT EARLIER IT WAS SO FUNNY
"Here's a list of books you can gloat in chat about:"
Course of Theoretical isn’t a book hahaha
does anyone know whether GPT 5.4 has access to Image 2.0?
It does
iirc image 2.0 is its own model but it should be able to invoke it
I think so
when asked what ur favorite book is, if its not a novel, im immediately sus
It was there before 5.5
like
I am using codex app to run gpt 5.4 but it seems really off though
u ask me to name 10 books
What is your favorite book
probably
In what sense?
1984 or maybe count of monte cristo
I think there is a plugin you can get that removes the blocked messages in discord, i might need it o.0. Blocked is more messages than actual messages
Yes. Codex supports image 2.0 Generation!
@mossy bronze
atlas shrugged
chronicles of the sibyls war, queen
thrawn ascendancy lesser evil
thrawn
phm
artemis
For me it’s gotta be the gales of November w book
Codex related discussions only here please.
those r the top few
Ok. Dude is blocked since admin thinks he’s ok in his head.
havent read it
Caught larping hahaha
I gotta read atlas shrugged saw the movie that’s it. Artemis I heard was good right now I’m reading project Hail Mary
lmao i love how the hypernerds came out to just pick this mf apart intellectually
ah ive never seen the movie for atlas shrugged
I tend not to like watching the movies of books that I read
phm is an exception
I once finished a book on the way to the movie because I wanted to finish it before hand lol
lel
This itself is hate speech.
if u think someone broke the rules
Screenshotted. Reported.
This tbh
ping the mods so they can direct you to modmail 💀
@quasi canopy
there is a mod here not doing anything, they deleted my message though
I turned on memory, in codex, but i am not sure it actually helps i see the citations a lot but i dont see a lot of difference in output.
the only thing fine here is my sense of fashion
Hi how can I help?
I am not a hypernerd. I just grew up genuinely loving physics. Not everyone who reads Landau is doing it for clout. Some of us just find it interesting. Weird concept, I know. But some are destined to achieve singularity with physics. With the help of Codex I can actually retain my knowledge for longer, and rehearse in the most optimal way possible. Did you know Codex can be better than a personal tutor?
The memory file is like 1500 lines as well. I wonder how it is managed
No unsolicited DMs!!
Some arguing going on, and @mossy bronze kinda is instigating a lot of it by trying to get on everyone’s nerves and mini mod, just overall kinda weird situation going on
no hes literally the same guy from before
Just a second I'll review the situation, thank you for pointing it out
hes ban evading
What? Please tell me what I did against the ToS or discord rules, I willl make sure to wrok on it 🙂
lol
thank you please ban
You know exactly what you have been doing
every time we try to talk about codex or just normal developer banter this guy comes in being weird and toxic basically
is the current vibe
Even if it’s just a similar subject it is still related to codex usually
And if anyone asks a question, we still try to help answer it it’s not like we are ignoring people 🤷♂️
Thank you.
she will turn ur cat buff without consent via image generation models
Dude i have so many cat pics
You've hit your usage limit. Upgrade to Pro (https://chatgpt.com/explore/pro), visit https://chatgpt.com/codex/settings/usage to purchase more credits or try again at 3:07 AM.
These things happen
Yes, this is normal once you have used up your quota
Her name is Panic
Now YOUR name is panic.
i anyone else having trouble changing models in codex app? I have to send a message before it actuall changes.
Incase this is my last message, goodnight everyone
im sad
It may be worth delegating simpler tasks to GPT 5.4 mini
Not for me, but does it give any errors?
ah i just remembered anti gravity is free
Nah it just noops until i send my first message
I’m the 3rd one
Check out Gemini CLI in that case aswell! Free quota is very generous for 3.0 flash
That’s normal
yeah ive used 5.5 medium and high for this session
To @mossy bronze and others, I would like to ask to deescalate this situation, remember to respect each other so we can have meaningful conversations,
Also let's keep this channel on topic about Codex, if someone breaks the rules again don't be afraid to ping a mod or open a modmail case, thank you ^^
Yes sir! Understood.
You can read deleted messages, right?
I have once been muted for breaking rule #3. Fair 🙂 I should not be an exception to the law.
first message is always in the old model, then i can change model and move forward.
do people seriously use
cli
codex and gemini
i like my gui
The law 🤣💀
CLI's are quite good harnesses, for Codex it does not matter since quota is shared, but inbetween gemini cli and antigravity, they do not share the same quota pookl
i like codex app
what do people use it for anyway
No need to continue this, thank you
i like it via vs code since i can see the code and stuff
Alright. If further clarification is not needed, feel free to continue enjoying Codex!!! P.S: Try out GPT 5.5! Launched just a few days ago.
The details are still being reviewed, once again I would like to point out the importance for deescalating this case,
Regardless of the case, let's steer back to the original on-topic Codex discussion, or move to #off-topic thank you once again ^^
🫡
Let’s start a new convo, how have you guys been using 5.5 to do stuff better than 5.4, and any tips for it
I use 5.5 to waste my limits but faster
/clear AHAH
/clear
emit agi brain system with over 150 folders that simmulate near gpt 7.0 intelligence
haven't had time to do a proper A/B , been mostly continuing on what i was doing but switched to 5.5, didn t spot anything obvious so far
Yes, GPT 5.5 is a much more capable model, especially the low 5.5 vs low 5.4 you will see a very noticeable difference. Low reasoning used to be almost never used. Now it is fine to use low reasoning for slightly rigorous tasks
Do you not find it much faster?
The benchmarks say that
For me, it’s slower
What are skills?
Have you made sure your codex is up to date? This is a critical thing
Could I add a codex skill to teach it physics?
I get slower responses for all tasks compared to same level of reasoning on 5.4, but better responses
I’d like to teach codex physics
are u serious
Yes! I have done so
What a Skill Actually Is:
A skill is a folder containing a Markdown instructions file plus optional scripts and resources. Think of it like an npm package but for AI agent behavior instead of code. You write the workflow once and Codex knows how to follow it reliably every time without you re-explaining
Really? How?
I wanna teach codex physics too
Codex is a great teacher of phyiscs
How You Use Them
Codex can activate skills two ways:
Explicit: You type $skill-name or /skills in the CLI/IDE to call one directly
Implicit: Codex reads your task and automatically picks the right skill if it matches
Why It Matters
Skills launched as experimental in December 2025 and are now part of an open standard adopted by 16+ AI tools including Claude Code, Cursor, and Gemini CLI. OpenAI itself has built hundreds of internal skills and publishes many publicly on their Gitub. The end goal is to turn Codex from a generic coding assistant into a purpose-built AI teammate that follows your team's exact workflows.
I wanna teach codex how to name its top 10 favorite books
Wonder the response
How
I asked codex it’s top 10 favorite book, and it told me a series of textbooks as a book
Wow! Such impressive work. turning a valuable tool into real world MONEY!
So I don’t think codex really knows about physics
That is so fascinating!
I think sometimes it pretends to have the knowledge
I would like to add more skills though, I’m getting used to them
I think we should teach codex about how to make transaction redundant mini games with api keys in the front end. simple, yet enriching task!
Sure
United States
Codex keeps making up lies a lot
If you teach it on physics it should respond better
Yeah
codex doesnt actually make money for people
Me when I lie
Me when codex lies
Pro tip: Use your usage limit fully at the end of each cycle — otherwise, you might lose some percentage.
codex doesnt make any money though. What ever your product is makes money.
I wish you could sell back un used limits
I don't think personally that anyone who is PAYING for codex is not getting back their investment in return.
Hah, okay, that makes sense. Maybe stay on free plan if they cant afford. Decently generous 🙂
Can you explain this to me?
What do i copy paste?

