When to Use Each ChatGPT Model, According to OpenAI

ChatGPT isn't based on just one AI model. On the contrary, there are a number of different models you can choose to power the chatbot with—even more if you pay for it. While you can experiment with each on your own, OpenAI has its take on when you should be using each—the advice is just a bit hidden.

As spotted by Bleeping Computer, OpenAI now has official advice on the best times to use each of its AI models. It would have been easy to miss, as the advice in question is attached to a support document for ChatGPT Enterprise, the company's business-tier subscription service. And while there is some guidance that only applies to business users, much of the advice here applies to everyone.

When to use GPT-4o

OpenAI advises users to use GPT-4o for everyday tasks. The company says that includes brainstorming, summarizing text, drafting and workshopping emails, and creative content—basically, the things the average user might expect ChatGPT to be good for.

Some of the example prompts OpenAI suggest include summarizing meeting notes into "key action items," generating an email after launching a project, proofreading a report you wrote, or brainstorming a launch plan in "real time." For that last prompt, OpenAI encourages you to upload sketches or screenshots to aid the model in its response.

When to use GPT-4.5

OpenAI rolled out GPT-4.5 in preview earlier this year. In fact, you'll need to pay for ChatGPT in order to access it, and even then, it's a bit hidden in the "More models" section of the model selector.

If you do have access to 4.5, OpenAI thinks you should try it out when you want the chatbot to help with tasks that involve "emotional intelligence, clear communication, creativity, and a more collaborative, intuitive approach to brainstorming." At first glance, I don't really know what to make of that, since OpenAI also recommends GPT-4o for "creative content," but the company does offer sample prompts.

For example, OpenAI thinks you should choose GPT-4.5 if you want to make an "engaging" LinkedIn post concerning trends in AI. (The irony.) The company also recommends the model if you have a new feature launching on your product, and you want an AI bot to write you a product description, or if you want to come up with an apology letter to a customer with an "empathetic tone."

It seems the company wants you to use GPT-4.5 to replace most of your writing, including when you're supposed to be sharing a human moment with someone else. Personally, I'll keep writing my emails myself.

When to use ChatGPT's reasoning models

Reasoning models take a slightly different approach to generative AI than other models. The goal is to have the model break problems down step-by-step and "think" through each move, in order to end up with a more "thoughtful" and complete output.

The company's flagship reasoning model is o3. The company recommends o3 for complex or multi-step tasks, including detailed plans and analyses, "extensive" coding, advanced math, science, and visual reasoning. Based on the example prompts, OpenAI sees you asking o3 to come up with a risk analysis for market expansion, generating a business strategy based on data from competitors, or running detailed analysis on a spreadsheet, while forecasting the upcoming quarter.

I haven't personally put o3 through any of those paces. However, I did hop on the viral trend last month of using the model for geoguessing. I had mixed results, but, more interestingly, the results were about the same when using gpt-4o instead.

On the flip side, there are the o4 reasoning models: o4-mini and o4-mini-high. These, according to OpenAI, have more focused use cases. o4-mini, for example, is for "fast technical tasks," like extracting data points from a CSV file, summarizing a scientific article, or fixing an issue in Python.

o4-mini-high, on the other hand, is also good for technical tasks, but specifically ones that are more detailed—at least, according to OpenAI. That might include not only solving a hard math problem, but walking you through the steps, as well, or explaining a complex scientific concept in everyday language.

While all users can access o4-mini, only paid subscribers can use o4-mini-high and o3. Free users only have limited access to o4-mini, as well.

Despite the advice here, I still haven't found a regular use for any of ChatGPT's models. I don't code, so I can't speak to the bot's ability to help there, nor do I enjoy generating AI art. The thing chatbots are best arguably best at, generating text, I do myself (I like to call it "writing") and I don't have any interest in outsourcing that work to a bot. For now, the main thing I'll continue to use ChatGPT for is testing out new features to cover here.

Maybe a company like OpenAI will release a generative AI feature I do find regularly useful. That just hasn't happened yet.

View the full article