AI Prices Are About to Shock Everyone

Im never paying for an LLM. I might change my mind when they get to an AGI.

6:07 AI costing more than humans only happened because large tech companies gave thier employees unlimited access to do whatever and some of them just ran endless agents. Not of them produced any really work. Thats not going to happen again and it’s just a anomaly in the budget over time

I'm here because i've noticed in the last 24 hours that suddenly my token usage is MASSIVE compared to what it was a couple days ago. I run out of my tokens instantly and before I could do an entire session for a couple of hours and barely use my tokens. I'm bloody shocked. all of a sudden, using AI is not feasable. I am not willing to fork over $200/month. SHIT.

p.s. look at how happy that CEO looks to be massively spiking prices. oh joy.

Google will give it away for free. Everybody else will die

vibe pricing

So you're saying this rip off is becoming more of a rip off. Try AI video generation. It was never cheap. 6 clips with a hundred quid worth of credits in some of them sites.

Actually, I'd argue you're wrong and you are overpaying. You are paying for something that is wrong 90% of the time. AI is notoriously bad at just being incorrect. It is honestly like talking to a drunk idiot at a bar who will keep arguing with you even after you prove you know more than him. If the price is going up, I would cancel immediately.

Here is the problem with what you are saying, and this is the part I think you are missing. Everyone thinks AI is going to make all this money, but the trends show the exact opposite. Companies are already complaining they are not getting any real value out of it. You get a quick boost for about ten seconds, but then it starts hallucinating and you end up losing money. Too many people have AI psychosis right now. Anyone who actually relies on it for anything important ends up hating it. If you are just talking to it, which is all most people do, it shouldn't cost more than a fucking dollar a day. If it actually created value, then sure, a price hike would make sense. But it doesn't make money. It loses money.

Nah not with open source models out there costing even less to run.

unfortunately like everything there will be a small fraction of people that will adopt LLMs at home with open source models but most of the people won't mind of pay thousands of dollars for more convenience and the big tech machine will keep working

I am indeed paying 20$ a month and this acceptable for me. If the price of my plan goes up to let's say 40$ then i cancel my subscription. If the plan stays on 20$ but usage becomes (very) limted i will also cancel my plan. I think there are many many users like me. There are even more users who don't want to pay anything at all... So i an curious to see what happens... They need uders to amortize their costs, if rising prices cuts the number lf users by half, they didn't get any extra earnings...

BTW thought
9:00 Stealth price hike in supermarket: shrinkflation, or if by using lower grade/less of the expensive ingredients (here, slower chips?) skimpflation.
Like maths, sometimes different fields have different words for parallel concepts.

No one will pay the expensive AI models, this will end in a disaster. The AI companies need to drop the prices of tokens to stay relevant. Local running LLM models are probably the future.

Watch everyone run to Chinese open source models

This guy doesn't know that opensource models exist like kimik2.7 glm5.2 that go hand to hand with top frontier models.
At the same time, chinese AI are free to use look at deepseek, kimi, stepfun.
Price change assumption only exist if the performance gap is orders of magnitude higher than free models.

I miss unlimited, good enough, ChatGPT3.5

Welp, it was nice while it lasted but I don’t need it🤷

Cool. There is a problem with the US monopoly however. They just put export bans on development. Fable was the first, but now we know where the watermark is for other models to not cross. Who is it that will NOT be constrained? People who own hardware, people who use Deepseek.
Deepseek, which is FREE. I also played with QWEN on my home machine and found it a good starting place...snappy enough(not that Opus is slow by comparison) and somewhat useful, but on my machine with ZERO hardware upgrades.
So, the US is competing with Free and Free. I pay extra for 1x OpenAI and 1x Anthropic account. What do they think will happen when China start getting used because the US have locked the RoW out of development that US people can (maybe) access? How will the US react to China 'getting all that data and building all those dependencies'.
So neither of these seems beneficial:
1) US block AI releases forever. They have 3-6 months for deepseek to catch up to Fable, 12 months tops.
2) The are competing with models on HuggingFace on hardware people control and own.

What the US models ARE is monolithic. What Hugging\Chinese models are is a broad wrapper around expert AI's. Both are very useful, however monolithic is not so much better that it give anything but a minor boost, so the US can lose this race very easily... and would have to ban US corps from using deepseek and any for of restricted access to models will invoke Risk Management reposturing.

Businesses are already screaming loudly that this is overpriced crap, and the average American consumer won't pay a nickel for this AI slop. Can't wait for this bubble to burst!

Pay for hallucination hits hard

AI makes you dumber. If the price goes up, the dummies will have to go back to using their meat brains 😂

Open Source models and Chinese models are getting better every month. It will be impossible to maintain premium pricing for a product that is marginally better in responses and significantly worse in resource demands. I see a closing window of opportunity. The good news is that the overall economy and the public will likely benefit as a result.

Open source models are the future.

Wrong. DeepSeek can keep its prices low for years, while US AI companies cannot.

20 bucks a month should cover the debt service on $1.4 trillion.

this is how china wins!

Average user won't keep using the new stuff when they can't afford to

China has a TON of unused data center capacity available for rent, using their cheap electricity and saving local water resources 🤔

The algorithms are probably trained to maximize token use via unnecessary loops and other nonsense

AI shrinkflation before the product is even good🙄

Actually the limitation on data center building is possibly the best thing for AI. They will have to stop and think on how to optimise the operations. Necessity is the mother of invention

Great video! The 2025 Nous Research study was a real eye-opener, but token efficiency has evolved rapidly since then. Here is where things stand now in 2026:

1. The efficiency gap is shrinking: Independent audits (like TokenMix in April 2026) show that while open-source models still use more tokens, the massive 10x waste has shrunk down to just a 1.5x to 3x overhead as newer open-weight models optimize their reasoning paths.
2. Models are on a "thinking diet": Labs are actively fixing the "bloated" reasoning issue. For example, recent releases like Kimi K2.7-Code (June 2026) have successfully slashed unnecessary internal thinking tokens by about 30% through direct execution paths.
3. Open source still wins on ROI: A January 2026 MIT Sloan economic analysis revealed that closed-source APIs are still about 6x more expensive per million tokens. Despite being slightly less token-efficient, open source remains the cheaper option for high-volume enterprise workloads.

The shift in 2026 is clearly moving away from "bigger models" and toward "denser, more efficient reasoning"!