ChatGPT's New 03 Mini: The Best Model I've Tested Yet!

ChatGPT has just released its first reasoning model, 03 Mini, and it's available to everyone, including free users. I've been testing it out, and it seems like a big step forward, especially for anyone working with science, math, or coding.

What is 03 Mini?

OpenAI pushed this release forward, likely due to competition from other companies releasing similar models. The big news is that 03 Mini is designed to be exceptional at STEM capabilities – think math, science, and coding. It's also supposed to be faster and cheaper for developers building apps on top of it. This new model will replace the older 01 Mini in the model dropdown.

For ChatGPT Plus, Team, and Pro users, 03 Mini is available now. What's really cool is that free users can also try it by selecting "reason" in the message composer. This is the first time a reasoning model has been available to free users, which is a pretty big deal. Many free users haven't even experienced the older reasoning models, so this opens up a lot of new possibilities.

OpenAI says that while their 01 model is still the general knowledge model, 03 Mini is a specialized option for technical tasks that need precision and speed. They also mentioned that paid users can choose a "high" intelligence version of 03 Mini for more complex tasks, though it might take a bit longer to generate responses. Pro users get unlimited access to both versions.

Key Takeaways

• Available to All: 03 Mini is accessible to free ChatGPT users, a first for a reasoning model.
• STEM Focused: It excels in math, science, and coding tasks.
• Faster & Cheaper: Offers lower latency and cost for developers.
• Multiple Versions: Includes standard and "high" intelligence options for paid users.
• Web Search Integration: Can now combine reasoning with web search capabilities.

Performance and Testing

OpenAI shared some benchmark numbers showing 03 Mini performing very well, especially in coding and software engineering tasks, often beating the older 01 model. While they didn't compare it to outside models like DeepSeek R1 or Gemini, I'm planning to do that in future videos.

I decided to test it with some questions that the previous 01 model struggled with. The first was a simple comparison: "Which one is bigger, 9.11 or 9.9?" The 03 Mini answered correctly in just four seconds, stating that 9.9 is bigger than 9.11. This is a pass, as the older model got it wrong.

Next, I tried a more complex word problem: "A horse costs 50, a chicken costs 20, and a goat costs 40. You bought four animals for a total of 140. Which combination did you buy?" The 03 Mini took a bit longer, showing its work, and provided two correct combinations: zero horses, one chicken, and three goats, or two horses, two chickens, and zero goats. This is another win, as the previous model only gave half the answer, and a competitor model got it right.

I then tested the "03 Mini High" version with a coding challenge: "Create a game of chess that I can run locally on my computer." It generated the code, and while the pieces weren't visual, the game logic worked perfectly. I could move the pieces correctly, and it even recognized Checkmate.

Combining Reasoning with Search

One of the most exciting features is the ability to combine the reasoning model with web search. I asked 03 Mini to find the pieces needed to make the chess game look more realistic. It found a link, identified the pieces, and instructed me to save them to a specific folder. After doing that, the chess game worked with actual chess pieces! This integration of search and reasoning is something I haven't seen work this well in any AI tool before.

I also tried some STEM questions that I couldn't solve myself, and the 03 Mini got them right, matching the answer key. So far, I'm really impressed with how well it's performing, even the version available to free users.

It's fast, it has web search integration, and it's accessible to everyone. I'll be doing more in-depth testing and comparing it against other models soon, so let me know in the comments if you have any challenging questions you'd like me to try!