Exploring the Hidden Potential of GPT-4o mini

GPT-4o mini: A Game Changer for AI Pricing

GPT-4o mini is a game changer when it comes to pricing. Take everything that you think you know about how much it's going to cost to generate text with an AI in your no code app and throw that out of the window. I've got a quick Bubble demo just to show you exactly the impact that this pricing change is going to have.

Comparing GPT-4o and GPT-4o mini Pricing

For comparison, we've got GPT-4o, which is the most powerful offering from OpenAI as of date. That is $5 per million tokens input and $15 per million tokens output. Look how much it is slashed down to with the GPT-4o mini.

Demonstrating the Power of GPT-4o mini

In practice, we'll go over into my Bubble app here. In fact, I'm just going to refresh this, and we'll see the power of it because what I've got is, I've got a prompt. It's not a short prompt. It is the transcript from a video that I did earlier on in the week. And I'm saying generate a blog post. Then I've inputted the pricing for each of the two different models for GPT-4o and GPT-4o mini and I'm going to run the AI now.

Running the AI and Analyzing Results

This is sending an API request over to Bubble. It's using both the GPT-4o model and it's using both the GPT-4o mini, and it's going to report back how many tokens have been used and the total cost of running it once. But then also if we were to multiply up by 100 and use it, imagining that we've got 100 different users, or each user in our app uses it 20 times, we've got five users, we would need to be considering what it takes to multiply up by 100.

Comparing Output and Cost

So here we go, we get our responses back and we can see that they are different. Now, this is to be expected because GPT-4o is still the flagship model and OpenAI have said that GPT-4o mini is intended for smaller tasks. We can see that perhaps there's an improvement here because it's adding rich text to the numbering. In fact, it's just laid it out in a different manner. We scroll down. Oh, we actually see that, in this case, the mini has generated a longer output.

Cost Analysis: GPT-4o vs GPT-4o mini

But here's where it really matters. This is our cost. So this cost us two cents to run through the GPT-4o model. And it would cost us $2.3 to run that 100 times. Now, simply because of how I've set this up with it using a currency field that's limited to two decimal places, it's not free. We can just dive in. I'm using the mathJS plugin here, and it's telling us that the result is actually $0.08154.

The Implications of Low-Cost AI

So effectively that means that once you multiply that up by 100, we could run this prompt 100 times and it would cost us $0.08. So, yeah, like I said at the start, you're throwing everything that you previously knew out of the window about how you should price AI for your users. I've already seen articles in response to GPT-4o mini effectively saying that the grand dream of LLMs available through APIs is that they are effectively free because the cost is so minimal.

Cost Considerations for Web Apps

Especially when you consider all the other things that go into running a web app. If you can run this 100 times and it costs you eight cents, then yeah, it's a seriously, seriously low cost AI. And you could begin to analyze and think which one of these is a better response. But, I mean, ultimately it comes down to, well, what if GPT-4o mini is just good enough for what you're trying to do, and you've just slashed the cost of running the AI for your users.

The Future of AI in Apps

So what do you think? Are we going to see AI basically become a de facto feature in many apps where it isn't a cost anymore? Yeah, there are many apps around nowadays I think about, like Raycast. I love Raycast. Raycast brings the OpenAI and the anthropic API right onto my desktop. But I pay extra to get access to some of the more expensive models. But they're gonna come down in price. Where are we gonna be in a year or two years time? What if the best AI models in a couple of years time are just like this, where when you run it a single time and you limit your cost to two decimal places on the dollar, it's actually basically nothing?

Conclusion and Call to Action

So please leave a comment down below. And if you want to learn how I built an app like this, then do head over to our website, which is a link down in the description planetnoco.com. Because we've got over 300 Bubble tutorial videos just like this one. We've got a No Code community, and we've got some new Bubble courses coming your way. So head over to the website, subscribe to the channel to find out more.