Are OpenAI's o1-mini and o1-preview Truly Undetectable? An In-Depth Review

Since its inception, StealthGPT has understood that the most optimized form of AI content is humanized, undetectable AI content. Detectable content has no utility as an output than informing the user.

For the longest time, this undetectability is what was lacking from OpenAI’s ChatGPT and other mainstream llms. All the easily decipherable watermarks of bad AI writing were present in what ChatGPT users were delivered as outputs. If a student dared use ChatGPT for their schoolwork, any AI detector would immediately flag the text.

However, with new AI models, o1-mini and o1-preview, being released to the public today, the OG’s of AI content might’ve finally wised up about the future of AI and optimized their text to bypass AI Content detectors. That’s at least what we’re going to find out in this review.

Table of Contents

• What is OpenAI’s o1-mini?

• What is OpenAI’s o1-preview?

• What is Undetectable AI?

• Are OpenAI’s o1-mini and o1-preview Truly Undetectable?

• Conclusion

• FAQ

What is OpenAI’s o1-preview?

OpenAI’s o1-preview is the first of a slate of new AI models that take more time thinking so they can take on complex tasks and solve harder problems. New levels of reasoning have been input into the process of how these models generate their answers so that like the human thought process, o1-preview will try different strategies and recognize its own mistakes.

This makes for longer time generating outputs, but also more correct answers. o1-preview still does not have the capabilities of browsing the internet as GPT-4o, but for advanced reasoning, o1-preview touts itself as the new benchmark.

What is OpenAI’s o1-mini?

OpenAI released two new AI models today. The first of which, o1-mini is a cost-efficient reasoning model meant to take on all the demands of STEM, specially math and coding. For 80% cheaper than their o1-preview tier 5 users of OpenAI’s API can o1-mini as a lower latency of o1-preview. Testing for speed against o1-preview and GPT-4o, o1-mini was 3-5x faster in delivering outputs.

After undergoing human preference evaluations, like o1-preview, o1-mini was determined to outperform GPT-4o in reasoning-heavy domains but gets outperformed in language-focused domains, something that spells doom on the detectability front.

What is Undetectable AI?

OpenAI made these two models to address the STEM fields and made note of their capabilities in math, science and coding, however English or linguistics was largely amiss from their analysis and most likely from the new model’s machine learning.

Odds are it’s just as detectable as GTP-4o. That said, you may still be wondering what Undetectable AI is then. Undetectable AI is AI writing that has been made to feel like human writing by fine-tuning the language and removing the watermarks of AI writing such as low perplexity and burstiness. Those watermarks of AI writing are the exact elements that AI detection tools use to flag content as being AI generated.

AI tools like ChatGPT might have some use cases, but the only way to make it useful as an essay writer or blog writer is to run the text through an AI humanizer, the likes of which are usually available from tech startups like the best, StealthGPT.

Now that you understand what undetectable AI is and how to make ChatGPT undetectable, lets see if the new AI models are worth using or if they’ll be detected.

Are OpenAI’s o1-mini and o1-preview Truly Undetectable?

The only way we can determine whether these new Open AI models are undetectable is by passing them through AI detection tools like Turnitin, Originality, and GPTZero. First, lets see our outputs from each model.

Here is o1-preview’s essay about Los Angeles and San Francisco:

As you can see, the output now includes the duration of the AI’s thought process. When you click on that duration, an analysis of the AI’s thought process will drop down for you to see. More insight into this should open up new possibilities in terms of prompt engineering for more effective outputs.

At 7 seconds, we should expect o1-mini to be faster. Lets see:

o1-mini thought for only a second and didn’t provide a drop down menu of its process. With these two outputs, we’re gonna determine how detectable they really are.

vs Turnitin

Running o1-preview through the most popular and widely-used AI detector, Turnitin gave us this result:

Turnitin immediately detected this was 100% AI generated content with 10 instances of plagiarism found. Here’s the result for o1-mini:

Slightly worse, 100% AI generated content detected with 11 instanced of plagiarism found. Not good. These new AI models need to come with a disclaimer about its actually utility.:

“Only use this content it to inform yourself, not actually for anything useful.”

It should read.

vs Originality.AI

Originality.AI came up with the same conclusions:

o1-preview was detected as 100% AI-generated content.

o1-mini was also detected as 100% AI-generated content. This exercise has become a closed case of low expectations.

vs GPTZero

Lastly, before I move on to less obvious questions than the new ChatGPT’s detectability, lets run it through one last detector, GPTZero:

Same as the others. GPTZero detected o1-preview’s text to be 100% AI generated content.

And as for the new AI-model’s little brother, GPTZero detected o1-mini’s text to be 100% AI generated content.

Conclusion

Nothing is new under the sun. ChatGPT’s ambitions in automation never included using their content in professional or academic settings. Thus, their new models are as completely detectable as GPT-4 turbo.

If you’re a student, thinking about using o1-mini or o1-preview to write your essays, don’t. You will need an undetectable AI software if you don’t want to be accused of academic misconduct. You will need StealthGPT, the most undetectable AI writing tool in the game. You can visit our pricing here, where plans are tailored to your needs and discounts are abundant to make sure everyone has the opportunity to use the most optimized AI-content on Earth.

FAQ

Can OpenAI’s ChatGPT-4 Be Detected By Plagiarism Checkers?

Yes. Absolutely. Though we tested their new AI models, certainly their old one is just as unable to bypass AI detection tools as the new.

How to Avoid AI Detection

You will need to humanize your text by removing the watermarks of AI writing such as low perplexity and burstiness. Undetectable AI services like StealthGPT humanize text by making it feel like human writing with high complexity and randomness in the text’s word choice, sentence structure, and sentence length.

Is StealthGPT Undetectable?

Yes. We’ve proven it time and time again, against AI detectors large and small. No matter if it’s the largest tool on the market, or some free tool in some dark corner of the internet. We’ve tested our software against the best of the best to prove no AI detector or other undetectable service compares to StealthGPT.