the bottom line is that a.i is getting more and more expensive and they have excuses for that.
just last week trae.ai reduced its plan credits from 400$ to 100$
Inference for the same level of intelligence is getting cheaper. Frontier intelligence keeps finding ways to get more expensive, yes.
But gpt-6-sol, for instance, dropped price for the same level of intelligence. That will keep happening, and someday we'll have basically-free sol. And hopefully by then, very cheap astra.
But some new wild thing will be very expensive.
I'm sure I'll figure out how to use the new one, but I would be extremely happy with near-free Astra.
when people say "if I was in situation x" there is an implication that they are not currently in situation x. it looks a bit odd when you go for the attack from uncertain footing man, I'd feel bad for attacking back knowing that you've interpreted my words differently to how others would.
maybe i’m just not crying about it because im an immigrant from a 3rd world hellhole so all i and my family knows is how to adapt to do better. a handful of us in a sea of billions.
i’m going to take advantage of technology that is freely available in many cases to improve my standard of living because change is the only constant in life.
100 years ago Market St was covered in horse shit.
today it’s human shit (sad but true, but not really) but more so that there are cars, lime scooters, BART, etc.
Coders didn’t exist 100 years ago, but writers, scientists, and artists all did, and their work is being stolen just as much as the coders. Tired take.
meh. i’m a self taught former swe. i went to school for business just after 2008 crash.
times changed.
i had to adapt to 3x my take home TC to get a job at a unicorn that ipo’d to like 10x or more my NW because i felt what it was like to work in an area where ppl were just laid off and i didn’t want to work 60 hours a week to impress some suit.
my relative is a phd research biochem scientist.
he had to adapt by taking a couple of years to switch sub-fields, taking on new roles and now in his latest research role he uses AI heavily to assist him to analyze data for patterns that humans had to previously brute force. he works on cancer research btw.
adapt.
crying is cheap and easy, for people who were going to be left behind regardless.
I wonder what's the ceiling for these plans' prices.
At some point, it becomes a non-negligible fraction of the salary for the employee, especially in non-US employment markets.
A lot depends on how good open source models do. I know I can buy a 20k rig or whatever to get model X. If someone wants to charge me 3k a month to write code I just buy the hardware.
I think there will be an employer backlash at some point. Paying an extra couple thousand in subscriptions so a junior can write copy-paste emails and Slack messages? Nah fam, write those yourself.
Or just use AI mode on Google's homepage. I asked it to find an article that compares a certain thing to another, which doesn't exist. So Google automatically wrote a 5 page article comparing the two things and built a proper article with cited references. Most people don't need a subscription to anything.
The google AI thing on their website gives me terrible results when I use it at work. Basically it's so unreliable I don't trust it. FWIW I don't log into a google account on my work devices (because we don't have Google enterprise services, and I'm not going to log into my personal Gmail account on my work computer) so that could be making a difference.
In my observation employers are willing to throw practically unlimited cash, probably because it’s directly funding a future where they don’t have to pay pesky engineers with their holidays and their pensions.
Of course a company doesn’t want to pay for someone to delegate all their work for an additional fee.
This is why some companies stopped hiring juniors. They would rather pay a mid-level or higher with the expectations that the LLM use will be a productivity booster in a more direct sense. So the bar also rises.
to be honest wasn't much surprised cause of the current situation like frontier models being so expensive and the models like opus consuming way much more tokens you just can't justify any of that. this token math is just so damned right now :(
Particularly in this situation, it cannot be unlimited. A command to one agent could scale into a whole datacenter very quickly.
It can feel unlimited for many use cases, probably. But I doubt it can, let's say, 3D model things in blender on 3 different computers for 24/7. And hey, why not 30 things? 300? If you're just watching them, it's easy to get ridiculous.
I understand your point; but I disagree - Unlimited is possible.
It gets into a "define unlimited" discussion. You could give "unlimited" usage, but still "limit" both concurrency and throttle speed. Even if you run 24/7, they still have knobs they can twist.
There is always a limit unless you start scaling horizontally in the sense that you have multiple accounts via subscription or API. There is always some limit to any single account, even via API.
I'm thinking the persistent agent is something that can't scale to multiple agents. Imagine only one agent you can invoke as many times as you want on a single account, but it cannot invoke subagents or scale itself. It's just always there ready to do work when you need it. But it behaves as "one".
This would be trivial to implement. Of course, if you have the money, you can have multiple $500 subscriptions and have multiple persistent agents.
Cloud agents in Codex are quite poor solution compared to Claude Code, you are forced to use your own computer, which is not the desired setup for anyone with a standard laptop not suitable for doing lots of work in parallel.
So I would still prefer to spend some dollars in Anthropic or even Cursor solutions, which work much better as cloud dev envs.
https://chatgpt.com/backend-anon/checkout_pricing_config/con...
I'm waiting for the ProMaxUltra+
That's the next iPhone.
Duo.
Or Xiaomi
the bottom line is that a.i is getting more and more expensive and they have excuses for that. just last week trae.ai reduced its plan credits from 400$ to 100$
Inference for the same level of intelligence is getting cheaper. Frontier intelligence keeps finding ways to get more expensive, yes.
But gpt-6-sol, for instance, dropped price for the same level of intelligence. That will keep happening, and someday we'll have basically-free sol. And hopefully by then, very cheap astra.
But some new wild thing will be very expensive.
I'm sure I'll figure out how to use the new one, but I would be extremely happy with near-free Astra.
Sol6 is clearly worse than 5.6. Have you tried using it? And Astra is more expensive.
Claude subs are great value again ever since Astra release moved lots of people to OpenAI subs. Even moreso since Opus 5.5 release.
We should measure cost per standard task. This is falling fast.
We stole all your work, just subscribe to get it back... now at $500/month.
tired take
I'd be tired too if my industry was being destroyed and scavenged by millionaires
your industry? lol
it didn't exist 100 years ago
adapt
when people say "if I was in situation x" there is an implication that they are not currently in situation x. it looks a bit odd when you go for the attack from uncertain footing man, I'd feel bad for attacking back knowing that you've interpreted my words differently to how others would.
nah. life is all about adapting.
maybe i’m just not crying about it because im an immigrant from a 3rd world hellhole so all i and my family knows is how to adapt to do better. a handful of us in a sea of billions.
i’m going to take advantage of technology that is freely available in many cases to improve my standard of living because change is the only constant in life.
100 years ago Market St was covered in horse shit.
today it’s human shit (sad but true, but not really) but more so that there are cars, lime scooters, BART, etc.
Coders didn’t exist 100 years ago, but writers, scientists, and artists all did, and their work is being stolen just as much as the coders. Tired take.
meh. i’m a self taught former swe. i went to school for business just after 2008 crash.
times changed.
i had to adapt to 3x my take home TC to get a job at a unicorn that ipo’d to like 10x or more my NW because i felt what it was like to work in an area where ppl were just laid off and i didn’t want to work 60 hours a week to impress some suit.
my relative is a phd research biochem scientist.
he had to adapt by taking a couple of years to switch sub-fields, taking on new roles and now in his latest research role he uses AI heavily to assist him to analyze data for patterns that humans had to previously brute force. he works on cancer research btw.
adapt.
crying is cheap and easy, for people who were going to be left behind regardless.
I wonder what's the ceiling for these plans' prices. At some point, it becomes a non-negligible fraction of the salary for the employee, especially in non-US employment markets.
Its cute you tgink youll be the one leveraging AI.
A token misfired
Sorry, asked copilotto think forme.
A lot depends on how good open source models do. I know I can buy a 20k rig or whatever to get model X. If someone wants to charge me 3k a month to write code I just buy the hardware.
I’m interested: how would you spend $20k to get close to performance parity (speed and ability) with a frontier model?
Once people become fully dependent the ceiling will be whatever they want it to be.
Not if there's competition.
I think there will be an employer backlash at some point. Paying an extra couple thousand in subscriptions so a junior can write copy-paste emails and Slack messages? Nah fam, write those yourself.
Or just use AI mode on Google's homepage. I asked it to find an article that compares a certain thing to another, which doesn't exist. So Google automatically wrote a 5 page article comparing the two things and built a proper article with cited references. Most people don't need a subscription to anything.
The google AI thing on their website gives me terrible results when I use it at work. Basically it's so unreliable I don't trust it. FWIW I don't log into a google account on my work devices (because we don't have Google enterprise services, and I'm not going to log into my personal Gmail account on my work computer) so that could be making a difference.
In my observation employers are willing to throw practically unlimited cash, probably because it’s directly funding a future where they don’t have to pay pesky engineers with their holidays and their pensions.
In my observation employers are quick to say “unlimited” and equally quick to say “no quite that unlimited.”
Of course a company doesn’t want to pay for someone to delegate all their work for an additional fee.
This is why some companies stopped hiring juniors. They would rather pay a mid-level or higher with the expectations that the LLM use will be a productivity booster in a more direct sense. So the bar also rises.
to be honest wasn't much surprised cause of the current situation like frontier models being so expensive and the models like opus consuming way much more tokens you just can't justify any of that. this token math is just so damned right now :(
If it's unlimited that would be tempting. But that will never happen and if it does "unlimited" would not be
Particularly in this situation, it cannot be unlimited. A command to one agent could scale into a whole datacenter very quickly.
It can feel unlimited for many use cases, probably. But I doubt it can, let's say, 3D model things in blender on 3 different computers for 24/7. And hey, why not 30 things? 300? If you're just watching them, it's easy to get ridiculous.
I understand your point; but I disagree - Unlimited is possible.
It gets into a "define unlimited" discussion. You could give "unlimited" usage, but still "limit" both concurrency and throttle speed. Even if you run 24/7, they still have knobs they can twist.
Fair enough, but you're kind of changing the parameters.
The limiting thing now is total tokens, so unlimited would mean infinite tokens.
And it's not hard to spin up to infinite tokens, one agent and one command can orchestrate it.
Is there really no limit on number of concurrent agents on either OpenAIs or Anthropic's plans?
There is always a limit unless you start scaling horizontally in the sense that you have multiple accounts via subscription or API. There is always some limit to any single account, even via API.
I'm thinking the persistent agent is something that can't scale to multiple agents. Imagine only one agent you can invoke as many times as you want on a single account, but it cannot invoke subagents or scale itself. It's just always there ready to do work when you need it. But it behaves as "one".
This would be trivial to implement. Of course, if you have the money, you can have multiple $500 subscriptions and have multiple persistent agents.
Cloud agents in Codex are quite poor solution compared to Claude Code, you are forced to use your own computer, which is not the desired setup for anyone with a standard laptop not suitable for doing lots of work in parallel.
So I would still prefer to spend some dollars in Anthropic or even Cursor solutions, which work much better as cloud dev envs.
Related ongoing thread:
Ask HN: How do you feel about the new $500 OpenAI subscription? - https://news.ycombinator.com/item?id=49838409 - Sept 2026 (8 comments)
I want it and I want it now.