$200/month open AI new pricing

[integral]

I think this is a sign of where things are going. The thing is, you can’t afford not paying for the best models as far as the future is concerned. because if you’re not paying for the best models, everyone else is and you won’t be able to compete with them.

The person with the best AI has a huge advantage over everyone else. So they will just keep introducing more and more expensive models.

There will become a point where Subscription fees will cost more than rent.

Edited December 6 by integral

[Leo Gura]

Seems quite the opposite. All these AIs will be indistinguishable and it won't matter which one you use for your purposes.

Whatever this $200 model gets me I could get from Claude for $20.

Edited December 6 by Leo Gura

[integral]

16 minutes ago, Leo Gura said:

Seems quite the opposite. All these AIs will be indistinguishable and it won't matter which one you use for your purposes.

Whatever this $200 model gets me I could get from Claude for $20.

The latest models are gonna take more and more computation power. They’re gonna be larger and more expensive to run. A model that’s 1000 times smarter than Claude could cost $1000 a month or more. 

Edited December 6 by integral

[Basman]

The trend of technology long-term is that it becomes more affordable over time. What you see now might be AI companies trying to make more profit after already having captured an audience. I suspect much of the price is paying for all the energy that it takes to run their servers.

AI is quiet competitive so there will be incentives to undercut themselves eventually.

 

[Bobby_2021]

We should run the tech companies dry while they are still at their phase of market acquisition and hope the compute will get cheap and OSS models will be good enough to match claude's capacity at the least. I am hopeful of that happening. The competition is getting stiff. 

[Basman]

Why make subscription fees more expensive than rent when you can raise both in peace and harmony? 😊🌼

[Bobby_2021]

The tech corruption is so soul crushing to see. They downright fucking parasitically drain society in broad daylight and we could do nothing.

Nvdia, Apple all could easily make their GPU, phones far better but hold off on such upgrades since they could sell more with incremental upgrades over a period of time. They are not producing the best equipment they could because there isn't any competition for them. Sick world. 

[AION]

Let the battle of the AI’s begin 

Edited December 6 by AION

[mmKay]

I wonder how home they hadn't done that sooner. It's common practice to have at least a single more expensive product to offer for the people that want to throw money at you 

[OBEler]

4 hours ago, Leo Gura said:

Seems quite the opposite. All these AIs will be indistinguishable and it won't matter which one you use for your purposes.

Whatever this $200 model gets me I could get from Claude for $20.

 

With o1 you can set how long it should "think". The longer it thinks the better the answer.

 

 

[Leo Gura]

4 hours ago, integral said:

The latest models are gonna take more and more computation power. They’re gonna be larger and more expensive to run. A model that’s 1000 times smarter than Claude could cost $1000 a month or more. 

Computation gets cheaper every year. So no.

[RightHand]

From what I'm seeing, the "Pro" plan only provides access to a more powerful version of o1.

Stick with the "Plus" plan, which is still $20/month.

Edited December 6 by RightHand

[Buck Edwards]

Don't invest so much in AI. 

It's gonna be a regular thing. 

[Fearey]

As @Leo Gura pointed out, for just $20 a month, you get access to what's arguably the most powerful general purpose model out there, Claude 3.5 Sonnet. Sure, you can pay an outrageous amount to be able to use OpenAI's extremely resource-heavy o1-preview and o1-mini models, but these are not designed for every day use, and you're not going to need them 99% of the time.

I've used these new models in work-related settings, and while they're extremely powerful if put to good use, they're also complete overkill most of the time. The reason the subscription to use these models is so expensive is because of the extreme amount of resources they consume.

I don't know if this is still the case, but with the $20 premium ChatGPT subscription, you used to get a limited amount of prompts with o1-preview and o1-mini each week. I don't think I ever reached either of those 20 or 30 prompt limits during my time with a premium subscription, because I never faced enough problems that was beyond the reach of their other models.

There was maybe 2 or 3 occasions where I had very deep bugs in an enterprise-level project I'm building, and where GPT 4o and Claude had no idea what was going on, or what was causing the error. When I gave the new o1-preview model the necessary information about that specific piece of logic, it managed to pinpoint exactly what was causing the bug and why, after thinking for about 1 minute.


To conclude, no, I don't think this is not a sign that AI is getting so expensive that normal everyday people won't be able to compete. These extremely powerful models were never intended for general use, and it would be a waste to use them for that. They were designed to tackle highly complex problems with a lot of moving parts, for the few occasions where that's what you really need to solve a problem.

For the very rare occasions you may actually need to use such a model, pay by a per-use basis. Don't buy a subscription for it. It'll be a lot more affordable. You don't need a tier 5 subscription with OpenAI to access these models freely without contraints. You can purchase usage from third-party services to accomodate your exact needs.

With the release of the full o1 model and o1 pro mode, I would assume these are designed to tackle tasks of an even higher level of complexity. I would struggle to think of any kind of problem I could ever run into where o1-preview or o1-mini wouldn't be powerful enough to help me come to a conclusion, but maybe that'll happen some day, and I'll buy a single prompt for one of these even more powerful models.

Edited December 7 by Fearey

[RightHand]

Cool video about the topic :

 

 

[Hello from Russia]

These AI chats fail to do simple addition for some reason, lol

[gambler]

26 minutes ago, Hello from Russia said:

These AI chats fail to do simple addition for some reason, lol

it's wild, isn't it????

[OBEler]

@Hello from Russia no Chatgpt4 o1 can do math quite well.

[Leo Gura]

11 hours ago, Hello from Russia said:

These AI chats fail to do simple addition for some reason, lol

That's just not its function.

These AI are probabilistic chaotic fuzzy machines, not formal logical systems. That's a feature, not a bug.

Edited December 9 by Leo Gura

[RightHand]

11 hours ago, Hello from Russia said:

These AI chats fail to do simple addition for some reason, lol

It's hard to do simple addition when high on acid