A first look at OpenAI's newest models: o1-preview and o-1 mini

I've now had a few days to play around with OpenAI's latest models, o1-preview and o-1 mini and like most people, I'm pretty blown away by what a major step forward this is. A couple of weeks ago I thought that Claude 3.5 Sonnet was quickly becoming the go-to model for software engineers, but I can tell you, o-1 mini has instantly taken its place.

First, for everyone who might still be playing catch up, here's the TLDR in a few bullet points:

  • four days ago, on September 12th - OpenAI released o1, their latest models
  • everyone thought they would call this GPT-5, but OpenAI said since this is such a foundational change, they're starting back at 1
  • the key breakthrough with these models is they are "reasoning models" designed to solve hard problems
  • OpenAI split these into two models, o1-preview which is the big kahuna, and o1-mini which is a smaller model, 80% cheaper, and aimed at developers

Over the last few days I've played around with both of these models, put o1-mini to the test writing some Python and Node.js code, and used o1-preview to do some European travel planning. Both are absolutely far and beyond the best LLMs out there today and while I feel a bit bad saying this, I haven't used Claude 3.5 Sonnet since they were released. 

Cursor also pretty much immediately announced support for both models so you can now use them there just as you would Claude 3.5 Sonnet which has been by go-to for coding until now. 

If you haven't read OpenAI's official Read Me about the new models, don't sit here listening to me, they do it much better, you can read it all here - Introducing OpenAI o1-preview.

Lately, one of my favorite things to do when I wake up in the morning is hop on Twitter/X and see what people are building with AI. Given what a massive update o1-preview is, people have been building like crazy and releasing some pretty wild stuff. Below are three that I think are worth checking out:

1. Mckay's o1 AI Playground - https://github.com/mckaywrigley/o1-ai-playground

2. Full weather iPhone app in under ten minutes - https://x.com/ammaar/status/1834348042637521031

3. Flappy Bird built in 3-4 minutes - https://x.com/slow_developer/status/1834614755153350809

There's still a lot more to uncover since it hasn't even been a week since OpenAI released the o1 preview. Right now my goal is to start to cost out some of the stuff I want to building using o1-mini through the API and make sure the volume I'm planning on sending it isn't going to cost me a small fortune. Measure twice, cut once right? Except in 1's and 0's.