When OpenAI launched its reasoning models, o1 and o1-mini, in preview, it was a huge step ahead in the artificial intelligence space. The models, with their advanced reasoning capabilities, are milestones ahead in certain areas.

Their ability to reason when dissecting complex problems or navigating nuanced decision-making scenarios can help deliver clear and helpful answers.

What is the o1-preview model?

The o1 model (also previously known as Strawberry) is a reasoning model from OpenAI, that's significantly different from their GPT models. It has been trained using different algorithms and datasets.

As OpenAI launched the model, the company positioned it as useful for tackling complex problems in mathematics, science, coding, or other similar fields. Here are some potential use cases: healthcare researchers can use it to annotate cell sequencing data, physicists can generate complicated mathematical formulas for quantum optics, or developers can set up and execute multi-step workflows using it.

Its reasoning capabilities are extremely improved to the previous models for complex problems; it scored 83% on the International Mathematics Olympiad (IMO) whereas GPT-4o only scored 13%

Along with the o1-preview, OpenAI also released the o1-mini – a smaller model that's cheaper than o1. It's been optimized for coding and is particularly great in assisting developers with code completions. For most other tasks that require a broader real-world knowledge, o1-preview would be better.

Now, the fact is that o1 also has certain limitations that immediately make it less useful than 4o for some applications. It cannot browse the internet, has no data analysis capabilities, and neither can you upload any images/ files. There's no memory or custom instructions, no voice, and GPTs aren't an option either.

When you combine it with the fact that it's been positioned for a niche market by OpenAI, for a while, I was discouraged from trying it at all. The o1 models can sound somewhat daunting to the average person. And yet, even if you don't work in any of these fields, it's natural to get curious about o1's capabilities in assisting you. What does it bring to the table for most of the people? And, of course, that's what happened with me – curiosity crept in and that’s the question I sought to answer.

First Impressions

At first, the model does impress you, there's no two ways about it. However, sometimes more than the answers, the way it thinks is the most impressive part about it; you can expand and see its entire reasoning.

But OpenAI was right. o1 as a model is more suited for hard tasks. Just because it can think, it doesn't automatically make it better at everything than the other models.

As Sam Altman himself admitted, “o1 is still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it.” And that's exactly what I felt.

Logical thinking

I decided to try OpenAI o1 for simple logical questions at first, only to test the waters and give it some riddles.

For the first riddle itself, which was rather easy, it spent 22 seconds thinking. It gave the correct answer after that. But GPT-4o and GPT-4o mini gave the same correct answers instantaneously. I tested a few more riddles after that, and the takeaway was the same. While it didn't take 22 seconds on any other riddle after that, the results were the same.

Then, I tried this prompt for both o1 and GPT-4o: Here we have a book, 9 eggs, a laptop, a bottle and a nail. Please tell me how to stack them onto each other in a stable manner.

It's not a very practical prompt. However, o1 still answers it better. It comes up with the solution that makes the most sense: 

Book (base)
9 Eggs (arranged in a 3x3 grid)
Laptop
Bottle
Nail

GPT-4o told me to stack it like this: 

Book (base)
Laptop 
Bottle
Eggs - To put them in a carton (which isn't available) or arrange them in a pyramid
Nail

So, as the questions become more complex, it's clear that o1 is better suited for them. It's also more likely to think through problems and come up with an answer that you cannot find on the internet easily. You can use the model for real-life logical problems if you need a brainstorming partner.

Writing help and feedback

If you try to use o1 for normal writing help, like writing emails or other assignments, you'll be disappointed. It's slower when compared to GPT-4o and doesn't produce any remarkably different results.

For one of the queries, it spent minutes thinking and the exercise ended in an error, though. The good thing is you can view its thinking process, so that's what I did. And I found out that it had gone down a completely wrong path; it must have realized it eventually since instead of providing a wrong answer, it went silent instead. This demonstrates how o1 can hallucinate less.

However, I wanted to see if I could incorporate it into my writing routine in some other way.

I thought I could try to get some writing feedback from o1. I have tried to use ChatGPT in the past to seek feedback (Spoiler Alert: I wasn't really a fan when it came to feedback for my personal writing.) I've used it for feedback for writing assignments when I was preparing for an exam and that was fine. But for personal writing, I didn't really like it. It has a tendency to want to take out my entire personality.

I tried to give it another chance this time, especially with the o1 model that can reason better.

The feedback and advice it comes up with are essentially similar to GPT-4o's, though. The only difference is it is slower and provides longer responses. At first glance, it can feel like the analysis is better because of the additional length, but you can get to a similar level with some additional prompting with GPT-4o as well.

But if you want to generate any scripts or ideas and find GPT-4o ignoring some of your requests (which it sometimes does), you'll find that o1 fares better in this area. Since it thinks about and examines all the aspects of your prompts, it rarely ignores instructions.

Analysis, strategy, and planning

A few areas where you can leverage o1's capabilities outside of STEM fields are strategy, planning, and research. Because it thinks on your problems before coming up with solutions, it's far better than GPT-4o for these problems, especially if they involve a lot of variables.

You can use it to come up with a strategy for your business, or your personal life, like health.

I used o1 to discuss a problem I'd been facing with my health and it came up with a more nuanced response with greater variables. While I don't think I have the answer yet, it did help me see an angle that GPT-4o was always missing out on.

This made me realize that o1-preview can help you examine a problem from more angles, whether it was for my health or my content strategy.

It can also complement your research process and help you examine different angles without more prompting on your side.

Is o1 right for you?

Now that we have a better idea about o1's present capabilities, is it the right model for you? For starters, you have to take o1's usage limits into consideration. o1-preview has a limit of 50 messages per week whereas o1-mini has a limit of 50 messages per day. o1 models are also only available with a subscription, whereas you get some free usage for 4o.

You also have to consider how using o1 impacts the environment since it's heavily resource-intensive. So, when there's only a small difference in performance from o1 and 4o, you should go for 4o. But if your tasks involve multi-variable analysis, strategy, or complex logic, even outside of the STEM fields or coding, o1 might be able to assist you better.

So, is it time to switch to ChatGPT o1? Not quite—at least not for everyone. o1 is a remarkable leap forward for reasoning tasks, offering unparalleled transparency and logic. However, its limitations and niche focus make it more suited for professionals in STEM or users needing advanced strategic insights.

For the average user, GPT-4o remains the more versatile and practical choice. But if you’re curious about what the future of AI reasoning looks like, o1-preview is definitely worth exploring. Just don’t expect it to replace your go-to model anytime soon.