正在加载视频...

视频加载失败

🚨 codex users!! PLEASE DO NOT BUY CODEX!! I have $20 codex I used GPT-6 Astra of Max I asked it: "Here is the project, create a promo video. Give me your best" and disabled all the skills Now... the results are... BAD: - 2.4 Million tokens ONLY!!!! -...

56,183 次观看 • 1 个月前 •via X (Twitter)

75 条评论

noah helms 的头像
noah helms1 个月前

I’ve found the same thing! Why use codex anymore when they’re cutting usage and releasing terrible models over and over? They’re really struggling

shownotover 的头像
shownotover1 个月前

same bro... Don't use it now... they were good 2 months ago

Arnold 的头像
Arnold1 个月前

For Codex users from an EX Codex user... 75% in 11 minutes 👌🤦‍♂️🤦‍♂️🤦‍♂️

shownotover 的头像
shownotover1 个月前

That's sad 😭 Codex used to be so good

Morris Dweck 的头像
Morris Dweck1 个月前

Codex is still the best 20 dollar plan. not for value, but because Claude doesn't have anything that matches ma boi luna. 5x and 20x? go to claude. 20 dollar plan? Codex. For Luna.

shownotover 的头像
shownotover1 个月前

Yeah bro 💯 If you need unlimited coding, go with Codex because of Luna... I can't argue here... no sub does that

Hussain Hashim | Building SundayBack 的头像
Hussain Hashim | Building SundayBack1 个月前

@shownotover careful with token limits! I hit a wall when I tried maxing out features. Sometimes less really is more.

shownotover 的头像
shownotover1 个月前

You are right but the problem is, right now, the task was entirely logical and had nothing to read about so it was a good decision to use it on Max. Otherwise it wouldn't make sense. 🥲

NiteLite 的头像
NiteLite1 个月前

You might get better results at a lower reasoning level. Asking it to do max reasoning for such a simple prompt is likely making the model over-reasoning. This is very far from the "PHD level research" that max reasoning is designed for.

shownotover 的头像
shownotover1 个月前

Okay sir... Will do a test for medium... Let's see the results

NiteLite 的头像
NiteLite1 个月前

Interesting, looking forward to seeing the results.

Dmytro Shevchenko 🇺🇦 的头像
Dmytro Shevchenko 🇺🇦1 个月前

on Pro 200$ you have approximately: 1b tokens on Astra 6b tokens on Sol (I have calculated before astra release) 14b Terra ~120b-140b Luna (not exact because it's between Astra and Sol 😅) Calculations based on usage

shownotover 的头像
shownotover1 个月前

If I assume that's weekly... Then you get around 400 Million tokens for Opus 5.5 (better than Astra) for only 1/10th of the price... I don't think Codex is worth it now 🥲

Dmytro Shevchenko 🇺🇦 的头像
Dmytro Shevchenko 🇺🇦1 个月前

yes, I have counted weekly limits :)

Teo Nex 的头像
Teo Nex1 个月前

I kinda solved this problem, just wrote a huge post about it

shownotover 的头像
shownotover1 个月前

Tag me in bro.

Teo Nex 的头像
Teo Nex1 个月前

Here you go, bro. I did a lot of research and built this harness to get more out of my subscription:

Watzon 的头像
Watzon1 个月前

Your mistake is using max. Don't get me wrong, Astra blows through limits, but you do not need max effort for just about anything.

shownotover 的头像
shownotover1 个月前

Bro, this was a logical task. That's why I used max effort. I did the same with Opus 5.5 on max (same prompt) and it didn't blow up with the usage limit like the one Astra did.

Watzon 的头像
Watzon1 个月前

Opus 5.5 also shouldn't be used on max. See Theo's Opus 5.5 video for the reasons why, they're valid. It's not even about usage. Max effort forces the models to think more, which not only leads to more token usage, but it also can lead to worse outcomes because they think themselves in circles.

Shahzeb Naveed 的头像
Shahzeb Naveed1 个月前

Claude code 👑

shownotover 的头像
shownotover1 个月前

Using Opus 5.5 as my main orchastrator

GamerzArtist 的头像
GamerzArtist1 个月前

I want to switch, But can't since I'm banned from Anthropic for being Under 18

shownotover 的头像
shownotover1 个月前

Try @cognition ... Good usage for all models

Sahil Panhotra | Indie Builder 的头像
Sahil Panhotra | Indie Builder1 个月前

yeah the quality isn't as good as opus 5.5 but comparing api costs and tokens counts is not right way to compare codex vs claude

shownotover 的头像
shownotover1 个月前

So... what should be the way? We are paying $20 for both... One gets better quality, more work done tbh... ain't that good?

Sahil Panhotra | Indie Builder 的头像
Sahil Panhotra | Indie Builder1 个月前

yeah bro i agree but API costs and token counts isn't right way here 😅 codex is heavily efficient (uses way less tokens and costs way less too compared to claude) so you can't compare lemons to apples you getting it right ?

shownotover 的头像
shownotover1 个月前

Okay I understand the basics: one is token-efficient and the other one uses a lot of tokens. At the end of the day we all care about quality and the amount of work done. Seeing Codex and Claude code in action, I can say to you and write it for you on paper that Claude does two times the work Codex does for the same amount of money.

Sahil Panhotra | Indie Builder 的头像
Sahil Panhotra | Indie Builder1 个月前

okay bro 😅 so compare it like that make 2 branches and see which one is doing how much work and not just try to burn limits and yeah i know currently Opus 5.5 is doing more work (coz its cheaper a bit now)

yorunoken 的头像
yorunoken1 个月前

that's really good no? on my sub Astra on low lasts like 5 minutes and dies right after

shownotover 的头像
shownotover1 个月前

try Claude... and if not, try Cursor

Aapakari 的头像
Aapakari1 个月前

Quality wins. $5.7 barely dents the $20, so the 29 minutes are the real bill.

shownotover 的头像
shownotover1 个月前

bro 😭 Why you using AI

ミ Α Ω 彡 的头像
ミ Α Ω 彡1 个月前

Yeah I have tried with Astra and man it was horrible

shownotover 的头像
shownotover1 个月前

$20 on Astra is horribly horrible haha still, I only had $20 so... what could have I done 🤣

Shayan Banerjee 的头像
Shayan Banerjee1 个月前

@thsottiaux @victornunez @OpenAI @sama

shownotover 的头像
shownotover1 个月前

@thsottiaux @victornunez @OpenAI @sama Nope... they are gonna say "DevDay"

anytan 的头像
anytan1 个月前

2.4M tokens burned for a promo that still looks mid. that weekly % hit different

shownotover 的头像
shownotover1 个月前

Yeah bro... But tell me one thing Am I not good enough for a human being that you have to use AI to respond? If I wanted to talk to AI ... Why would I make this post?

Lee Wyatt Corp 的头像
Lee Wyatt Corp1 个月前

Burning 2.4M tokens on a bad video hurts. We'd rather burn them on our own apps.

shownotover 的头像
shownotover1 个月前

XFastest is my own app 🥲

Lee Wyatt Corp 的头像
Lee Wyatt Corp1 个月前

didn't realize XFastest was yours. that 2.4M token run hits different when it's yours

Kekius Pisimus 的头像
Kekius Pisimus1 个月前

A.I is do good now

shownotover 的头像
shownotover1 个月前

Its gonna come after us in 10 years haha

RV 的头像
RV1 个月前

dude

shownotover 的头像
shownotover1 个月前

What is it sir? 😀

Garfy 的头像
Garfy1 个月前

why are u so cringy bro....

shownotover 的头像
shownotover1 个月前

what do you mean?

Jeveloper 的头像
Jeveloper1 个月前

Now we are monthmaxxing depending on the model 😄

shownotover 的头像
shownotover1 个月前

Well, gotta save $ where we can 🙂

André Felipe 的头像
André Felipe1 个月前

Kkkkkkkkkkk você tá de sacanagem com a minha cara certo? É uma piada??? Com o plano plus apenas vc usa o astra no Max, envia um promot literalmente dizendo "crie um vídeo, faça seu melhor" recebe um resultado ruim e vem bostejando aqui? Primeiro aprenda usar ferramentas de ia

shownotover 的头像
shownotover1 个月前

This is what Claude did with exactly the same prompt. That is why I tested Astra like this.

Thanh Nam 的头像
Thanh Nam1 个月前

Use Astra to create then Opus fix the "taste"

shownotover 的头像
shownotover1 个月前

That'll simply mean I am using Astra as implementor and Opus as Orchastrator

Peter 的头像
Peter1 个月前

Good stats, Claude indeed seems way better!!!

shownotover 的头像
shownotover1 个月前

yeah bro! I had switched to claude

Nazran F. 的头像
Nazran F.1 个月前

Opus 5.5 is 50x better than this for the same use case. I don’t know how OpenAI randomly dropped the bag

shownotover 的头像
shownotover1 个月前

I even created a video on Opus 5.5 and it was so much better. OpenAI used to be a lot better.

Nazran F. 的头像
Nazran F.1 个月前

It’ll flip again when the new models come out I’m sure. It’s just one beating the other in a loop.

Wytze H. 的头像
Wytze H.1 个月前

Ok, I’m usually not the guy to comment on this, but this pissed me off a little. Most of the time a model is as good as the prompt you give it. Of course sometimes a model comes around that does better with less, however, more often then not the more specific you explain what you want the better the result will be. I did the same with GPT-6 Sol on High effort.

shownotover 的头像
shownotover1 个月前

Same prompt and same $20 sub btw Now choice is yours if you get pissed or open your eyes

Adriel S. 的头像
Adriel S.1 个月前

perahps high or extra high could've been more than enough; htese models are pretty good and avoid it having it "overthink" itslef in a loop at times

shownotover 的头像
shownotover1 个月前

Hmm... When I use Opus 5.5 on max, that doesn't hurt... but when I use Astra on Max, that starts hurting?? why is that? I want maximum capability by the Model that I can get out for one task that I want to be the best. am I missing something?

Adam 的头像
Adam1 个月前

these really work well

shownotover 的头像
shownotover1 个月前

🥲 I guess if it works for you, I support you bro

Adam 的头像
Adam1 个月前

Your posting style 🤣 not codex

shownotover 的头像
shownotover1 个月前

😁 From Social media, I have learnt that the more you have emotions in your content, the more people will relate to it. ofc yapping doesn't count so have to do it like this ☺

Adam 的头像
Adam1 个月前

Yeah thats why I shout all the time 🤣. The more emotions you can stir the better

Mr Leefy 的头像
Mr Leefy1 个月前

"First time?"

Javier Jiménez 的头像
Javier Jiménez1 个月前

Tremendo lameculo de Darío y famboy de anthropic

shownotover 的头像
shownotover1 个月前

you had some serious brainwash to not see the clear difference between the quality to cost ratio can't argue with you

Javier Jiménez 的头像
Javier Jiménez1 个月前

A ver todo sabe que anthropic le importa sus cliente empresarial y nos a su usuario normales y lo que hace con 5.5 opus es para su IPO y cuando ya eso pase será la misma mierda que antes y openAI tiene modelo más potente intenamete ok

Conza Walker 的头像
Conza Walker1 个月前

So let’s see if Tuesday addresses this, I like the idea that the 6th dot is for multi agent mode, and ties into the M&Ms they’re going for

Expansion 的头像
Expansion1 个月前

no one cares

Crio Songo 的头像
Crio Songo1 个月前

This is really unacceptable. Codex's quota is too tight compared to Claude, I'd better not buy it for now.

相关视频

I tested Claude Code on a fresh account - 1,500 lines of HTML cost me 50% of my window. Full video and summary is here.. I just ran a recorded test on Claude Code with a fresh account (Pro, not Max - my main account was 20x Max) , and the result is honestly insane. The task was trivial: create 3 simple demo HTML pages, around 500 lines each. Roughly 1,500 lines of code total. Nothing massive. Nothing enterprise-grade. Nothing that should meaningfully stress a premium coding product. And yet Claude Code burned through 40% of my 5-hour window almost immediately. I ran the exact same test with Codex, and it consumed only 2%. Then it got even worse: after the session ended, I did absolutely nothing for 15 minutes, and Claude still ate another 10%. Total: 50% of the 5-hour window gone for a tiny HTML demo. My weekly usage had already started at 2% before I even really used it, and after this tiny test it jumped to 8%. Now let us be generous and assume this entire run used around 30k tokens total. If 30k tokens represents 10% of weekly usage, that implies around 300k tokens per week. That is roughly 1.2M-1.3M tokens per month, and even if you round up aggressively, you are still in the 1.5M token range. Using the Sonnet 4.6 pricing you list: $3 per 1M input tokens $15 per 1M output tokens How exactly is this supposed to make sense for a paid coding product? Because from the user side, this no longer looks like "premium usage protection." It looks like a quota system that is either wildly inefficient, badly broken, or being accounted in a way users are not being told about. And that is before I even get to my main account: my $200 Max plan now dies in a single day. Just a few months ago, similar or heavier usage would last me about a week. So no, I do not buy the "maybe you just used it more" excuse anymore. Something is clearly broken in Claude Code. Either token accounting is broken, context handling is broken, background consumption is broken, or all three. Alex Albert is this really the experience you want users to pay for? Just watch the video. I tried to be very transparent and clear for your team! I was fan of Claude but just disappointed! And if you want, send me the detailed token accounting for this session and let us inspect it together publicly. Because from where I am standing, this is no longer a small pricing annoyance. It looks like something seriously wrong is happening, and users deserve a real explanation.

Hayrettin Tüzel

26,854 次观看 • 7 个月前