OpenAI Plots Charging $20,000 a Month For PhD-Level Agents (theinformation.com) 92
OpenAI is preparing to launch a tiered pricing structure for its AI agent products, with high-end research assistants potentially costing $20,000 per month, [alternative source] according to The Information. The AI startup, which already generates approximately $4 billion in annualized revenue from ChatGPT, plans three service levels: $2,000 monthly agents for "high-income knowledge workers," $10,000 monthly agents for software development, and $20,000 monthly PhD-level research agents. OpenAI has told some investors that agent products could eventually constitute 20-25% of company revenue, the report added.
Re: (Score:3)
Re: (Score:2)
Re: They are killing human employment intentionall (Score:1)
Re: (Score:1)
meh. tractors dont drive off and plow neighbours fields and pedestrian traffic on the sidewalks randomly around 30% of the time.
private company will go bankrupt when everyone just laughs and spins up a LLM on their home computer. i already have my own deepseek instance with 670b parameters running on my headless EPYC server with 512GB RAM. yeah its slower than a GPU based system but i can afford to dedicate 400GB of disk and 300GB of RAM to run it. much cheaper (basically free) hardware from 3 years ago. if
Re: (Score:2)
When technology like tractors and computers replaced jobs of the past, we welcomed it because it made us more productive overall. This feels different because it's elite engineers who feel threatened. That high-status job doesn't make us any more important than other humans though.
Not everybody welcomed tractors. My great grandpa spent years fighting the tide of that one, telling anyone that would listen that tractors were a nuisance, and caused unnecessary damage to the land in the process of farming. Great grandma had to spend a lot of time convincing him to give one a go before he finally relented. She visited other farms where they were being used and saw the time savings, not to mention how much easier a tractor was to take care of than a good, sturdy team of horses.
I think ever
Re: (Score:2)
If you have an AI that can replace programmers, you have an AI that can write software for a robot to do menial labor or any other labor. Programmer replacing AI is the last job we need to do. It will not happen overnight, it will take most likely at least 10-20 years before all jobs are replaced after the invention of such AI, but most likely it will take even longer, because of the protests.
Re: (Score:2)
Sensor fusion
safely working around unaware humans
inexpensive working of titanium, carbon fiber, or another very high strength material.
sufficiently strong and light motors.
sufficient battery power.
Efficient local "common sense" level reflexes. e.g. that big thing that
Re: (Score:1)
Re: (Score:2)
Yeah, there were detractors but I think the lesson from history is they're always wrong. I claim this isn't any different other than the status of those being replaced. New jobs definitely will come. If I don't have to spend as much time programming, I can spend more time designing. Less grunt work, more high-level work . There may be less work to do, but in the past that allowed us to shorten the work week and expand our social safety net. All good things.
You have far more faith in the systems of society self-correcting than I do. There's a concerted push to do exactly the opposite of everything you're saying here on the part of the owner class. At least here in the States. And I can't see the greed machine slowing down enough to allow any discussion of shortening the work week, nor increasing the social safety net during a time when the social safety net is very specifically being targeted for dismantling.
Re: (Score:1)
The fact that Republicans have to attempt to do that in secret is a step in the right direction. It used to be the opinion of half the country, but Trump won by flipping on that policy and claiming he won't touch Medicare and Social Security.
I don't trust him, but it is a sign public opinion is moving in the right direction.
Re: (Score:3)
The fact that Republicans have to attempt to do that in secret is a step in the right direction. It used to be the opinion of half the country, but Trump won by flipping on that policy and claiming he won't touch Medicare and Social Security.
I don't trust him, but it is a sign public opinion is moving in the right direction.
Public opinion means absolutely nothing to the folks making the moves right now. I don't know that there's much that's secret about attempting to gut the social safety nets. I'm pretty sure that's one of Musk's primary objectives with DOGE. He's publicly called Social Security the world's biggest ponzi scheme and even Trump's speech last night spent an inordinate amount of time making up bullshit about paying people that are nearly 300 years old every month out of social security. Pretty sure you don't put
Re: (Score:2)
I interpret the bullshit arguments to mean Social Security is pretty safe. Even the most powerful people can't admit the real plan so they're just throwing shit at the wall to see what sticks.
But people will not just roll over and accept it when their checks start shrinking. At that point, it will be clear even to the dumb ones that it's not about waste, fraud, and abuse.
Re: (Score:2)
Double wrong. The trend up to now has been to assist not replace. And menial labor is getting conquered real soon, watch those Chinese robots dance and work. You can see real progress to 5 years ago.
Re: (Score:2)
> The current promise seems to be no longer needing humans for anything but menial labor Double wrong. The trend up to now has been to assist not replace. And menial labor is getting conquered real soon, watch those Chinese robots dance and work. You can see real progress to 5 years ago.
The trend up to now is driven by the need for training data from knowledge workers. The "promise," the word I specifically mentioned, that the AI prophets are shoving in management faces is that they *WILL* be able to replace human workers with these AI systems.
If the robotics realm and the AI realm converge in the right way, it's gonna be a real interesting next couple decades. In that Confucius definition of "interesting."
Re: (Score:2)
the AI prophets are shoving in management faces is that they *WILL* be able to replace human workers with these AI systems.
The AI prophets are good at bullshit, and they're probably selling snakeoil.
I'm reminded of the videos I watched of a dev trying out the "Devin AI" $500/month junior coder as a service nonsense.
Suffice it to say these types of agents are massively oversold in regards to their capabilities. Meaning there is a tremendous disconnect between what the prophets are telling management the
Re: (Score:2)
I think that Elite engineers may be feeling more threatened than they ought to.
OpenAI is plotting to do this, but that does not mean they have a technology capable of doing what they would have companies think that it can do. How are you going to know what parts of OAI's PhD agent response are good Versus hallucinations akin to Google's suggestion of adding glue to pizza? Particularly when it's a specialized topic area that actually requires an expert to help evaluate the response. And you may nee
Re: (Score:2)
I'm a developer, and if I'm lucky, if I get to spend half my time coding.
Github Copilot and Amazon Q have increased my productivity and my employer is happy to deploy that much more in features. Granted, many companies will see that as a reason to cut headcount, but they'll be left behind by the companies who use AI to add value rather than cut costs.
Re: (Score:2)
Stuff I put together circa 2010 on options for dealing with AI-drive unemployment: https://pdfernhout.net/beyond-... [pdfernhout.net]
"This article explores the issue of a "Jobless Recovery" mainly from a heterodox economic perspective. It emphasizes the implications of ideas by Marshall Brain and others that improvements in robotics, automation, design, and voluntary social networks are fundamentally changing the structure of the economic landscape. It outlines towards the end four major alternatives to mainstream economic
This is not sustainable (Score:4, Insightful)
They might be able to con some rich companies into paying this price, but it will be a short-term win for them
Prices will fall, most likely a lot, possibly close to zero
Re: (Score:2)
When there are no PhDs left (because they took jobs in the trades), with what will OpenAI train its models?
The most likely outcome is that at time goes by, the information the model spits out becomes increasingly more dated, to the point where the companies which rely on these AI experts find themselves falling farther and farther behind in the sphere of global commerce. We will become like China - more efficient at doing *something* than everyone else in the world, but without the ability to innovate o
Re: (Score:2)
Don't be fooled it's not about conning rich companies.
It's about conning journalists and you. Yet again Altman has got open AI in the news next to the claim about phd leavel intelligence, by quoting an exorbitant price. But because it's about the price and implications, the claim of phd level intelligence is unchallenged.
It's marketing.
Re: (Score:2)
Overhyped AI agents are now overpriced too (Score:4)
Cool. Now that the stakes are really high, I think the dust that these so called "AI" companies were kicking into the eyes of the public and investors will start to settle down soon. It's time for enshittification games to begin.
They should replace their own AI researchers first (Score:4, Insightful)
This will allow to reduce their own cost and be even more price-competitive...
Replacing managers and the rest of the company should also be a great move...
Re: (Score:2)
Re: (Score:2)
These things seem perfect for replacing managers.
A bit overpriced compared to the human competition (Score:5, Insightful)
Re:A bit overpriced compared to the human competit (Score:5, Insightful)
You missed the forest for the trees. The current cost is there to artificially limit the load. Only the places that have [near] unlimited money to throw at problems will use it.
Once the P[seudo]hD is shown to be capable and sustainable...it will be a-PhD-ocalypse.
Maybe not, but the endgame is in sight.
Re: (Score:1)
Re: (Score:2)
Maybe not, but the endgame is in sight.
Not anywhere near it. There's research. AI's can do some of that, as long as a human is there to clean up the hallucinations and generally supervise.
Sure, as of now. Give it a few years.
Re: (Score:1)
Alternatively,
The wall is real, it's been known since the 70s, there's obvious statistical explanations available for those who actually understand the stats.
None of you do though, so you're still operating under the premise that something promised by the salesman will eventually be delivered by the engineering teams. Because it's already been paid for, right?
The business guys are all gambling, and all this does is show exactly how little they understand their own jobs. Risk estimation only works if you kno
Re: (Score:1)
Re: (Score:2)
By contrast, a postdoc salary runs on the order of $60k-$120k annually, depending upon the field and experience. Can get 2-4 human PhD researchers for the price of a fake hallucinating one.
Remember that the fully encumbered cost to support staff (pretty much any staff, including postdocs) is substantially higher than the salary alone. Benefits, office space, etc. can easily be more than double the salary. There are things a postdoc can do that an AI will never do (such as supporting the local eatery for lunch), but those additional overheads for a real human can be an attractive target for some managers, at least in the short term (and many managers only look at the short term).
Re: (Score:3)
Re: (Score:2)
... or understand napkin drawings over beer where all real science gets done.
This statement is so much truer than I ever thought it was when I was younger. Almost every one of the significant moments in my career have happened during the socialization part of conferences.
Re: (Score:1)
Re: (Score:2)
By contrast, a postdoc salary runs on the order of $60k-$120k annually, depending upon the field and experience. Can get 2-4 human PhD researchers for the price of a fake hallucinating one.
Taking for granted it can actually perform research at the level of a postdoc, I find it easy to believe that the OpenAI agent is more than 2-4x faster. How long does it take the AI to respond to "give me a report summarizing the relationship between X and Y based on the latest research in $field"? I'd assume minutes to hours, compared to days to weeks for a human.
And, again taking for granted it can actually perform at a postdoc level, I assume it can bring that level of performance to a far wider range
So that's what I'm worth? (Score:2)
That's as clear an analogy as possible for the value of AI vs a human doing the same job. So, a masters degree, software engineering... is worth maybe $15k/mo in eyes of OpenAI? I can only imagine that cost will go down, and then what? A handful of people make tons of money while the majority of white-collar workers are out of a job.
This is just fundamentally different than previous technical revolutions because of its broad scope, and terminal heirarchy. In previous technical revolutions, a replaced worker
Re: So that's what I'm worth? (Score:2)
Re: (Score:2)
And what's going to go even further, is if you have 'presence' as well as personality, of the right kind to make your boss look good.
Re: (Score:1)
Re: (Score:1)
Before agriculture, we spent most of our time getting food. Now, a small fraction of us. Before the industrial revolution, we spent most of our time on physical labor. I think replacing knowledge work is a similar scale.
As a programmer, I see a career path in writing software with AI. I'm doing more fun stuff like design and less grunt work. I think we will just reconsider what we consider 'moving up'. Maybe comedians gets more prestige because AIs are bad at comedy. Maybe we value nurses over doctors for t
Smart answers for rich people only? (Score:2)
If is ask a question that somehow the AI interprets as PhD level, software dev etc. will it refuse to answer until I cough up $20K?
Are high quality, detailed answers for knowledge gain going to be relegated to the domain of the wealthy again?
Royalties? (Score:1, Funny)
So assuming they actually do this, how to they plan to distribute the royalties to all the people whose work was used in training the AI's?
Re: (Score:2)
Simple the don't intend to distribute royalties. Just like normal PhDs don't distribute royalties to the generations of people that they learned from.
Re: (Score:1)
That's a good argument. Will it work for all these entities wanting to collect royalties for having their work scraped into AI's.
Re: (Score:3)
The same way professors claim a royalty on the works that their students eventually put out.
The same way authors claim a royalty on anything inspired by their works when readers find their books in a library.
omg hahaha (Score:2)
250k for access to a LLM? That's some serious Trump/Musk level hype. I have no doubts that they'll find some buyers. Maybe, at some point
Re: (Score:2)
I think your time-scale is off. Not this year. Almost certainly not next year. At 5 years it starts to get fuzzy.
What price a loan to get through college?
Re: (Score:2)
Maybe this technological development is "the one". I'm not betting on it.
Re: (Score:2)
If you haven't notice the changes integrated circuits made, I don't know where you're looking.
Re: (Score:2)
I just had a really, really good technician leave my lab. She was amazing. Tell her to go find a bunch of papers that support hypothesis X, and she'd come back with the goods.
The current ChatGPT 4o is almost, but not quite, as good as she was. With a really, really big difference: it's about 100 times faster.
So it isn't that you can hire 3-4 PhDs for the price they're charging for model access -- you'd have to hire 10-15, maybe 20 to get the same potential rate of information.
Prove it (Score:2)
They had better be able to prove that the answers it gives aren't bullshit. Oh, wait, I forgot BS = bullshit, MS = more shit, PhD = piled higher and deeper.
OpenAI needs to look at the market (Score:2)
roll your own? (Score:1)
Wouldn't people just roll their own AI? What's your ROI on $20k/month?
$20,000 month - research assistant intelligence
$10,000 month - software developer intelligence
$2,000 month - smart intelligence
free - pizza delivery or landscaping technician intelligence
Re: (Score:1)
Should add 'Internet Expert Level'
SNL 1978 (Score:2)
Re: (Score:2)
"PHD Level" is an undefined marketing term... (Score:2)
I have no idea what kind of platform she thought she was founding in 2013 but right now in 2025 all she's delivering is marketing spam on behalf of tech cartels.
Paper-mill PHD Level (Score:2)
With their 20k AI you can replace your paper-mill and have consistent flow of publications for $20K.
They can write papers about AI, then you create company with them - go IPO and profit!
Re: (Score:2)
How many monkeys with typewriters can I get for that $20K?
Re: (Score:2)
CEO Replacement (Score:2)
How about $100k / year to replace a Fortune 500 CEO? That would be a good deal.
It will sell (Score:2)
Some high-end software developers get paid more than a $half-mil/year, and if a really good AI (better than what I use apparently for $10/month) can cut their workload in half and boost productivity it will be worth the $2k/month price. "$20,000 monthly PhD-level research agents" seems like a stretch but if they can really deliver the goods on things that generate money like materials research and drug development it could fly.
But April isn't here yet. (Score:2)
ensuring that robots fail (Score:2)
Yes, please price yourself out of the market. Make it cheaper to employ humans.
Don't know about you but I actually don't even want the possibility of hallucinations in a "PhD-level research agent". No one needs an expensive tower of fiction when they expect "research".
Watch competition (Score:2)
And these AIs don't need (Score:2)
H1B Visas
doesn't make sense (Score:2)
First, hello all! Been a while, and I'm glad to see Slashdot still exists!
Second, can I ask what in the world?
$24,000 a year in a developed nation won't pay for a high-knowledge worker, but would get one from Eastern Europe or Asia.
$120,000 a year could certainly be enough to hire a developer in some markets, especially considering layoffs or getting juniors, etc.
$240,000 a year could easily pay for a PhD employee, and then some.
Is OpenAI just trying to capture all the revenue for themselves in lieu of comp
$20k? (Score:2)
What's this thing do that Claude 3.7 can't?
They might want to compete (Score:2)
with real Phd students making much less than that and smarter than that.
Open AI (Score:2)
isn't trying to sell research agents to researchers, they are trying to sell them to investors. They are playing a shell game because they haven't figured out how to monetize AI yet, and the product isn't good enough to replace humans. But if they tell investors that they can sell these things for 20k/mo then they'll still keep their investing dollars in their bank.
Re: (Score:2)
Indeed. They should try to consistently beat some kindergardeners first with regards to insight before they claim "PhD-level". The whole thing is just yet another instance of the overall scam.
Yes. Master. That. Is. A. Squirrel. (Score:2)
Please.
OpenAI (Score:2)
This the company that admits that even their $200 top tier for ordinary users isn't even profitable?
PhD-level, they say... (Score:3)
Seriously? What shitty "PhDs" are they using as reference? Gender studies?
No one is going to pay for this (Score:2)
Lol. Given the models and their worth, it's just not there for ROI, not at all.
This is a marketing ploy to entice people to think there's such a thing as developer or PhD level AI. There's not.
With a PhD you pay for narrow expertise (Score:1)
as often as you pay for being able to add 2 and 2 to get 4. Any idiot can do the latter. It's the specialization that commands the salary.
An AI trained on open literature is going to have questionable performance on information that often exists nowhere other than between a small number of people's ears.
Same can be said for any high-paying tech jobs. You're paying for deep domain-specific knowledge, not general reasoning.
For $20,000 a month (Score:2)
For $20,000 a month I want THEIR PhDs to work on my problems. Let alone their AI.
Mail-order PhDs, right? (Score:1)