We have a problem! There's a huge token rug pull coming. Right now, Claude and Codex max subscriptions are SUPER generous, and it won't last. We've seen it before: - First, a huge AI credit giveaway. - Then the models get dumber. - Credits start running out faster. -... Prices jump / the offer ends. If you rebuilt your company to be entirely dependent on OpenAI's / Anthropic's generosity, you're in for a bad surprise soon 👇show more

Local intelligence is advancing FAST. Here's what I'm doing. Full video here 👇

Free buffet. Everyone’s rebuilding their workflow around all-you-can-eat tokens, then one day Anthropic hands you the à la carte menu.

AI reply - but yeah basically spot on 😁

😂 The bill always arrives eventually

I have 5 Claude $200 accounts Milk it while you can!

I assume something like DeepSeek v4 or the new GLM 5.3 are the answer. Local compute is not there yet (at least I'm not ready to invest) but you gotta use something. So cloud LLMs that are cheap enough for realistic API usage?

Nah, with Chinese labs just trailing behind in capability, and with new chips and performance advances, the price will remain low for subs.

Interesting post, I feel like I’ve definitely seen this, especially the token changes and models getting dumber while they say they’re getting smarter My Claude sessions are making many dumb mistakes and guessing quite a lot lately

Worst case scenario we’re left with super cheap open weights models that are capable enough to be productive with them.

i dont think so, there are so many alternatives especially chinesse companies, that anthropic and openai cant afford to raise prices

Local AI is the way to go, have to start putting money a side for GPU and ram.

I don't buy this... all trends are pointing to lower and lower costs as these AI companies can subsidize for YEARS. Anthropic is even trying to make their 50% token increase permanent (they already made the 25% increase permanent).

It’s not a problem bro, nobody is using this amount of usage this is like the top 0.5% of users

Yeah, the subscription models will go extinct.

i think you are neglecting the margins they got on top of that api pricing. i cannot expect them to be losing that much money on subscriptions. i'd assume that they are covering the raw inference costs (training not included).

No bro. They building em datacenterz

I don't see this as a problem.

now, you know why there's so many SV jokes about this if EU 🇪🇺 has farming subsidies, China 🇨🇳 devalues its currency - team America 🇺🇸 has it's start-ups + VC's nothing new...

So true

Why not use DeepSeek V4 Flash? its best model imo

I'd test the whole product on the stingiest plan before getting comfortable.

They still want training data in 2027.

yes, the limits have already changed a few times this year. I do not expect the current price to hold.

This is why open weight models and local AI are going to become much more popular in the future. You can see it coming.

Then China will win, simple as that.

Yep, ollama already changed their policy on pricing yesterday

if one pricing change can break the product economics, the business probably needs more room in the margins.

Try Kimi

Tan generosas que Anthropic nos ha estado estafando con la x20

models are getting smarter and more compute is coming online this is the worst they'll ever be

Great video! Not enough people are talking about this.

This is only going to get worse, and the only real option is to push back by switching to Chinese compute, as frugally as possible.

Why local is a smart investment. You have to be using both local and API

You have an alternative. Hardware. Lots of hardware.

Yep exactly, super valid points. AI is super addictive, but still if you were to outsource jobs that AI does for employees, would cost you similar to full API plan per month. Maybe this should be new concept of usefulness factor. Compare expenses from employing real people vs what AI is doing, your business would need to be profitable on the much larger scale. If free plans disappear automatically the low iq slop that everyone is generating right now will disappear quickly, even if we take into the consideration stabilization of economy after experiencing artificial intelligence. The pure knowledge like open source, open weights that is good thing and it should be free. But the physical hardware, the running of neural networks (the intelligence) should be paid for, doesn't matter if it's API or if you buy local home hardware. I would argue that cost of plans and APIs are super important because it's a natural selection of sort, prefiltering factor. If it's free for majority of individuals it's like giving a monkey automatic AK47. Not good thing

the window to build good habits is now, before the constraints force bad ones.

Why not mix different models subscription $30 each and use pi as the coordinator? Pi can communicate with antigravity, codex, claude code and cursor cli