[OtterLover]

I’ve been paying for the 20x ChatGPT Pro plan and using it. But when I think about it, I actually don’t use ChatGPT that much. I’ve never once used up the 5-hour limit or the weekly limit in Codex. At most, my weekly usage has only been around 10%, and my 5-hour limit has mostly never dropped below 20%. Would switching to the 5x plan be worth it?

[NoSuchAgency]

better to have too much than not enough.


If 200 dollars a month is reasonable for you to pay i'd keep going.


But i'd also reccomend you look into how you can use codex to its fullest.

I have 2 200 max anthropic accounts and a 200 chatgpt pro.

The 100 dollar plans for me have never been worth it.

I run autonomous engineering though basically 24/7, so for me it makes sense.

[7]

better to have too much than not enough.


If 200 dollars a month is reasonable for you to pay i'd keep going.


But i'd also reccomend you look into how you can use codex to its fullest.

I have 2 200 max anthropic accounts and a 200 chatgpt pro.

The 100 dollar plans for me have never been worth it.

I run autonomous engineering though basically 24/7, so for me it makes sense.

TBH I pay over 3k a month in AI credits. Im actually moving more onto local models now as its getting crazy like 3k additional charges on top of the subscriptions. and now all the limitations also. TBH as much as a I hate OSX on a 128gb ram macbook I can run massive models just like my KJP flow but twice as fast as the Strix Halo. I dont know I love API and the freedom but costs are high for me particularly but not just that the limitations now... constant downgrading models blocks ... pain in the ass

[NoSuchAgency]

TBH I pay over 3k a month in AI credits. Im actually moving more onto local models now as its getting crazy like 3k additional charges on top of the subscriptions. and now all the limitations also. TBH as much as a I hate OSX on a 128gb ram macbook I can run massive models just like my KJP flow but twice as fast as the Strix Halo. I dont know I love API and the freedom but costs are high for me particularly but not just that the limitations now... constant downgrading models blocks ... pain in the ass

honestly ive found its way cheaper to abuse the subsidized compute you get from the actual subscriptions themselves

my 600 ish a month gets me around 50k a month to burn with sota models.

and you can just stack multiple subscriptions with both anthropic and openai. things like cswap make it very easy. You can even automate it in your processes.

highly reccomend anyone who is paying api prices to consider also paying sub prices.

Spoiler (Click to View)

🔐 loading encrypted image…

AND local models are great for a lot now. But, there are times where a really good model is required. Offset costs with local for fuckin sure though

[7]

honestly ive found its way cheaper to abuse the subsidized compute you get from the actual subscriptions themselves

my 600 ish a month gets me around 50k a month to burn with sota models.

and you can just stack multiple subscriptions with both anthropic and openai. things like cswap make it very easy. You can even automate it in your processes.

highly reccomend anyone who is paying api prices to consider also paying sub prices.

Spoiler (Click to View)

🔐 loading encrypted image…

AND local models are great for a lot now. But, there are times where a really good model is required. Offset costs with local for fuckin sure though

Nice bit of information and yeah I use multiple API's also but for the majority of the work ill be using local models now mostly due to cyber restrictions more than costs but also it will save on the heavy lifting also. I will still ofc have API usage for certain things as local models especially as arbitrated models quickly become out of date and fresh data is needed. But also training local data sets is useful with certain providers too. One of the things I like about perplexity mostly actually for live data but final touches to stuff using either anthropic or whatever actually gemini has been far less restrictive.

But having unified ram and XDNA with the halo is useful especially under linux or if not vulkan or WSL. I can run massive models now locally. Also vast.ai is useful for things like cyberteil

[DramaKing]

I would second vast.ai and the Ryzen AI Halo with gpt-oss or another open-source model.

[Jire]

You can't buy ChatGPT 20x anymore, it's been disabled. But it is vastly superior to the 5x and I wish I could buy a second subscription for it.

[XauGold]

I’ve been paying for the 20x ChatGPT Pro plan and using it. But when I think about it, I actually don’t use ChatGPT that much. I’ve never once used up the 5-hour limit or the weekly limit in Codex. At most, my weekly usage has only been around 10%, and my 5-hour limit has mostly never dropped below 20%. Would switching to the 5x plan be worth it?

There is no 5 hour limit since months now for Pro 5-20X users, what are you smoking? Lol
Just get the 5X plan. Nothing competes with subscription plans, its not even close.
Anyone using API is a dumbass or rich.

honestly ive found its way cheaper to abuse the subsidized compute you get from the actual subscriptions themselves

my 600 ish a month gets me around 50k a month to burn with sota models.

and you can just stack multiple subscriptions with both anthropic and openai. things like cswap make it very easy. You can even automate it in your processes.

highly reccomend anyone who is paying api prices to consider also paying sub prices.

Spoiler (Click to View)

🔐 loading encrypted image…

AND local models are great for a lot now. But, there are times where a really good model is required. Offset costs with local for fuckin sure though

What in the F are you doing that you used 270B lifetime tokens?
Hopefully you are at least making 20K a month or something with that much usage, otherwise thanks for wasting resources I guess lmao.

[NoSuchAgency]

There is no 5 hour limit since months now for Pro 5-20X users, what are you smoking? Lol
Just get the 5X plan. Nothing competes with subscription plans, its not even close.
Anyone using API is a dumbass or rich.


What in the F are you doing that you used 270B lifetime tokens?
Hopefully you are at least making 20K a month or something with that much usage, otherwise thanks for wasting resources I guess lmao.

dipping nuggets you've never heard of into sauces you couldn't even comprehend.

[OtterLover]

There is no 5 hour limit since months now for Pro 5-20X users, what are you smoking? Lol
Just get the 5X plan. Nothing competes with subscription plans, its not even close.
Anyone using API is a dumbass or rich.


What in the F are you doing that you used 270B lifetime tokens?
Hopefully you are at least making 20K a month or something with that much usage, otherwise thanks for wasting resources I guess lmao.

There wasn’t a 5-hour limit? I remember there being a 5-hour cap for Pro accounts in the Codex app. Before Astra came out, for about six months. I was using the Pro plan. Now I’m using the Grok Heavy plan, but I’m debating whether to switch back to the ChatGPT Pro plan. If there really hasn’t been a 5-hour limit since a few months ago, then I may have seen something wrong.

---

You can't buy ChatGPT 20x anymore, it's been disabled. But it is vastly superior to the 5x and I wish I could buy a second subscription for it.

Yes, the 20x one has been discontinued for sale. Maybe that’s why I’m sad. When will it open again? I’m currently using Grok Heavy because Grok can generate images and videos.

[rodrigayz]

depends on your tasks.

the 200$/month (20x pro) used to be really good, it kinda felt unlimited for months.

however past 2 weeks, it hasn't been the case, there's no way to use their latest model (ASTRA) without burning the limits way too fast.

so i guess we're now waiting for the 500$ tier.


quota usage constantly changes from one provider to another so it ultimately depends on your use case.

for example, anthropic feels like more usage is given to you rn than openai, but thats probably temporary because alot of ppl moved to openai.

but now since openai is reallocating their compute to specific models, and they have alot of new users, its changing.


as for cyber tasks, i would never go to anthropic, its been months since theres no way to do reverse engineering w their ai models.

---

TBH I pay over 3k a month in AI credits. Im actually moving more onto local models now as its getting crazy like 3k additional charges on top of the subscriptions. and now all the limitations also. TBH as much as a I hate OSX on a 128gb ram macbook I can run massive models just like my KJP flow but twice as fast as the Strix Halo. I dont know I love API and the freedom but costs are high for me particularly but not just that the limitations now... constant downgrading models blocks ... pain in the ass

i wish there were actual comparable alternatives when it comes to self-hostable models but tbh i've not seen anything match daybreak blue SOL 5.6 from openAI when it comes to agentic reverse engineering. i tested so many options too, like deepseek v4.1 flash abliterated, glm 5.3, etc. nothing comparable yet...

also deepseek v4.1 flash, the max i could get on a 200k$+ USD multi-gpu setup is 90 token per second...was super slow imo and made tons of mistakes SOL 5.6 wouldn't of done.


i would say that i can only wish we get comparable models thats self-hostable in the next few months/year but i kinda am doubtful.
it wont be for economics reasons, strong models cost alot to host.