GPT-4o versus GPT-4

GPT-4o versus GPT-4

This is not another article showing you the amazing speech capabilities of GPT-4o, I assume you've already read dozens of these anyways.

Neither will I show you how I made it sing Christmas carols or reason with your college professor about physics. These features are all amazing - But for us these parts are the least interesting parts about GPT-4o. The most interesting parts for us is actually its speed of generating text, its cost, and its quality.

GPT-4o cost

The price of GPT-4o is 50% of GPT-4. This is a big deal for us, since we're paying a substantial amount of money to OpenAI each month. In totalt we're paying maybe $250 per month for tokens to drive all our AI chatbots and demos. This implies we're now saving $150 per month since GPT-4o is 50% of the cost of GPT-4.

For clients of us that have thousands of questions asked each month, this price reduction is substantial and will be noticed.

GPT-4o speed

GPT-4o is roughly twice as fast as GPT-4-turbo. This explains the reduced price, since you're basically paying for CPU and GPU time when you use an LLM. Hence, twice the speed translates to half the price, because you're using OpenAI's CPUs and GPUs half the time with GPT-4o compared to GPT-4-turbo.

GPT-4o quality

I've seen some people claim GPT-4o delivers reduced quality compared to GPT-4-turbo. For us quality seem to having been improved actually. It looks like it's more able to more accurately follow instructions and rules we provide to it, and it seems to be more "accurate" as we ask it to do things.

Others might have different experiences, but at least this is our conclusion. We do however have fairly complex system instructions, with a lot of rules - In addition to high quality context data stored as Markdown with URLs and image references.

Try GPT-4o

You can actually try GPT-4o here. In the bottom right corner is our chatbot, which as of from today is running on GPT-4o. If you want to try it on a more complex AI chatbot that's doing advanced support, you can also try GPT-4o in our Magic Documentation website found below.

We're rolling out support for GPT-4o for all our clients these days, and hopefully before the end of the week, all our clients will have GPT-4o support in their cloudlets. First we need to stabilise and test though, which we do on our own internal cloudlets, to avoid having negative side effects result in down time for our clients.

Thomas Hansen

Thomas Hansen I am the CEO and Founder of AINIRO.IO, Ltd. I am a software developer with more than 25 years of experience. I write about Machine Learning, AI, and how to help organizations adopt said technologies. You can follow me on LinkedIn if you want to read more of what I write.

Published 15. May 2024