My mother was annoyed with ChatGPT the other day. She had asked it about a particular doctor, and it had cheerfully made things up. None of it was right.
Well, I said, you're on the free version.
It sounds like a cheap shot, but it is really what this column is about. How clever an AI you get to use depends more and more on how much you pay. And the price is set in American dollars.
RM100, RM400, RM800, RM2,000
Good AI now works like Netflix. You pay by the month, and there is more than one tier.
Take ChatGPT's Malaysian price list. The cheapest plan, Go, is RM24 a month. The mainstay, Plus, is RM99.90. Pro starts at RM420. Claude is much the same, with two Max plans above its hundred-ringgit Pro offering five and twenty times the usage, at a little over RM400 and RM800 once converted and taxed.
At the end of last month OpenAI went further still, with Pro 500 at US$500 a month. That is over RM2,000.
"Usage" here generally means a cap on how much computing you can burn every five hours and every week, and on some plans every month as well. Hit it, and you wait for the meter to reset.
You get what you pay for?
So why should the expensive ones be cleverer?
Regular readers will know that a large language model like ChatGPT is, at heart, a word-guessing machine. It reads what came before and predicts what comes next. The early ones did little else. That goes for the first two GPTs, academic curiosities from 2018 and 2019, and for GPT-3.5, the model that made ChatGPT an overnight sensation at the end of 2022. Ask one of them about the weather in Malaysia and it would most likely tell you it was sunny. It would have no idea that the haze is back and that Kuala Lumpur has turned, once again, into what locals call KLCannotC, for the days when even the towers at KLCC disappear.
Today's models are a different animal. Two columns ago I wrote about them cracking mathematical problems that had stood for decades. In my experience, if you pay for the basic hundred-ringgit tier, most of what you ask will come back more or less right. Someone still has to check it, of course.
The money goes on two things.
The first is size. AI is still largely a matter of brute force, and bigger brains are cleverer brains. The big labs mostly keep their model sizes to themselves, but Kimi K3, which the Chinese company Moonshot AI released for anyone to download in July, has 2.8 trillion parameters. You can think of those, loosely, as the connections between neurons. Smaller models exist, and they are noticeably less good.
The second is tools. A lot of what makes today's AI impressive is not raw intelligence but whether it knows how to use them. Ask a good one a medical question and it will search the web, pull up the research literature, take notes, think it over, and only then answer. Which is more or less how any of us would do our homework if we were doing it properly.
Both of these burn electricity. Bigger models burn more, and doing the reading before answering burns more still. On OpenAI's own price list, its top model, GPT-6 Astra, costs a hundred times as much per word as Luna, the cheapest model of the same generation. The free version of ChatGPT chats with a small model from the Luna class. What my mother met that day was a machine built to answer her as cheaply as possible.
Are we being fleeced?
Is this, then, the usual story of capital squeezing the customer? Not entirely.
In early 2025 Sam Altman, OpenAI's chief executive, admitted that the company was losing money on its US$200-a-month Pro plan, because "people use it much more than we expected." That July, Anthropic said one subscriber on its US$200 plan had gone through tens of thousands of dollars' worth of usage, and brought in weekly caps.
In other words, the companies are not necessarily making money on their heaviest users. What holds prices where they are is competition. OpenAI and Anthropic are going at each other hammer and tongs, and a pack of open-source models is close behind.
To the throat, not the lungs
The trouble is that a hundred ringgit is not small change either.
This is where Malaysia starts to feel the pinch. We are better off than many countries, but the exchange rate leaves us short against the rich ones.
Among my students and friends, the people who actually use AI for their work all end up on the RM400 tier at the very least. The hundred-ringgit allowance is gone after an hour or two of serious use. There is a Cantonese phrase for this. It reaches the throat but never the lungs. A taste, not a meal.
Do the arithmetic. The median full-time worker in the United States earns about US$5,400 a month. A US$100 subscription is a fiftieth of that, which stings, but no more than another internet bill. The median Malaysian wage earner makes under RM3,000. RM420 is a seventh of it.
Go out to the kampung, the villages, and it is worse again. The median rural household, the whole household, brings in about RM4,600 a month. The subscription would eat nearly a tenth of it.
Rice, rent, and AI
It used to be that whatever the exchange rate did, as long as you weren't travelling abroad, the essentials were priced locally. Food, clothes, housing, getting around. In Malaysia those are cheaper than in a long list of rich countries, so a salary of two or three thousand ringgit could buy a perfectly decent life. Imported luxuries were another matter.
AI is one of the few necessities priced in dollars. To be fair, OpenAI has given us a budget plan at RM24, which is less than the US$8 Americans pay for it. But the plans that do the real work, Plus and Pro, cost roughly what they cost in America. In the US, the twenty dollars for Plus is about two McDonald's meals. Here, the RM400 tier is about the monthly instalment on a Perodua Axia, the most basic car on the road.
Step back far enough, and in this one respect Malaysia itself may be the kampung.
I have never been an AI evangelist. But it is becoming a necessity for anyone who works in an office, and it may come to matter as much as where you live or how you get to work. I don't just mean coding. What matters more is learning to live alongside it. It is not going away, and worrying will not help. Better to work out where its limits currently are, which jobs you can safely hand over, and which still need a human watching.
You find the tricks by using it
One perk of working in science is that the problems are hard enough to stay interesting. Whenever a new model comes out, I give it the problems the last one couldn't solve. It tells me very directly which work is still mine to do and which has stopped being worth anything.
That kind of judgement may be the most valuable thing anyone can build over the next few years. Even for a clerical post, an employer would rather hire someone who knows AI well and knows where it tends to go wrong.
And the only way to build it is through a lot of use. A few columns ago, writing about data centres, I argued that this business comes down to trial and error, and a team that can afford ten attempts has better odds than one that can afford a single go. People are no different. The genuinely interesting uses rarely occur to you the first time. More often they arrive somewhere around the hundredth. A hundred-ringgit allowance does not stretch to many attempts.
That is why my PhD students, who are not exactly flush, grit their teeth and pay for the RM400 tier.
What other countries are doing
Plenty of governments have noticed.
In March, Singapore's Manpower Minister Tan See Leng told Parliament that "like learning a language, developing true fluency in AI comes from consistent use and building confidence through experimentation." Since September, Singaporean citizens aged 18 and over who complete an approved AI course get six months of a paid tool such as ChatGPT Plus or Google AI Pro, free. The government expects to reach about 200,000 people over three years.
Malta has been blunter still. Since May, under a deal with OpenAI, citizens who finish an online course called AI for All get a full year of ChatGPT Plus.
Malaysia is moving as well. At the end of July, Prime Minister Anwar Ibrahim announced AI Untuk Rakyat, AI for the People, run by the Digital Ministry. Malaysians aged 18 to 30 who finish its courses get three months of AI tools free, with 100,000 places in the first round. The 2027 Budget, tabled on 9 October, adds another 100,000, puts ChatGPT on the list, and lets AI subscriptions count towards the lifestyle tax relief.
The reasoning is the same everywhere. Even if the state takes a small loss, a population that knows how to use AI is now one of the core measures of a country's competitiveness. Microsoft reported last month that about 22.6% of Malaysians aged 15 to 64 are using generative AI. That is above the global average, and less than half of Singapore's 64.3%.
The good news is that it keeps getting cheaper
The more hopeful news is that most work does not need the very latest model. Even for learning, the frontier model of six months ago is more than enough.
AI prices follow two curves. A team at MIT estimates that running the most advanced models is getting three to eighteen times more expensive every year, because the models keep getting bigger and spending longer thinking. That goes some way to explaining a RM2,000 tier. But reaching any fixed level of performance is getting five to ten times cheaper every year. Capability that needed the RM400 tier a year ago is probably within reach of a plan costing tens of ringgit today.
Open-source models are closing in as well. The same Microsoft report found that they often catch up with the closed frontier within months, at a fraction of the price. And companies have got shrewder about it, saving the expensive model for the hard problems and handing everything else to something cheaper.
Getting in the water
So my advice is simple. Don't throw money around. But where the tuition is worth paying, pay it.
My mother's run-in with ChatGPT wasn't entirely the AI's fault. She had simply met the version designed to use the least electricity. In this era, the people who get left behind will not necessarily be the ones who cannot afford the most expensive plan. They will be the ones who never really got in the water.
That is true of a kampung, and it is true of a country. The tide is already in. The ones who wade in first are the ones who get to ride it.
About the author: Graduate of Chong Hwa Independent High School, Kuala Lumpur. Earned his PhD in Astrophysics from Harvard University in 2017, then held a NASA Hubble Fellowship at the Institute for Advanced Study (IAS) in Princeton. Former joint professor of Astronomy and Computer Science at the Australian National University, and currently Associate Professor of Astronomy at The Ohio State University, where his work focuses on using AI to accelerate discovery in astronomy and the fundamental sciences. Now on sabbatical in Malaysia, watching the local scene as an outsider, and welcomes letters from readers. Personal website: https://www.ysting.space