Bug Exposes OpenAI’s Secret Model o1, Now Accessible to All

As anticipated by OpenAI, the highly powerful o1 model, equipped with absolute reasoning capabilities, is set to be launched this year. However, a leak during the week of October 28 has suggested that Strawberry, Q, or o1—whatever name you prefer—has already begun operating, and we’ve unwittingly witnessed its performance.

About a month ago, OpenAI announced the introduction of a new AI model that was immediately integrated into ChatGPT.

This new addition includes both o1 mini and o1 preview, which are now available even to users with a free ChatGPT account, albeit with certain restrictions in place.

Following the integration of the model, particularly o1 mini, ChatGPT has experienced a significant boost in performance, with notable improvements in precision when compared to GPT-4.

Free Access to o1 Following a Technical Incident

According to statements from OpenAI’s CEO, the o1 model is markedly different from earlier versions of GPT, including GPT-4o used by Advanced Voice.

Furthermore, documents from a corporate briefing indicate that the final version of o1 is expected to feature substantial advantages over its test version, a claim that seems to be supported by recent developments.

Although most users have not had access to the beta version of o1, a technical incident last Friday provided some users with approximately two hours of access to what may be the final version of o1.

This new system is anticipated to incorporate features such as image analysis, web browsing, and data processing.

In response to the incident, a spokesperson from OpenAI mentioned that a limited testing phase for the o1 model was underway when the technical problem arose; however, they were able to resolve the issue swiftly.

True Capabilities of the o1 Model Unveiled by a Leak

Since the preliminary release of the o1 model, those close to OpenAI have been eager to highlight its capabilities, especially after the “preview” label was removed.

Some experts speculate that the test version was trained on an earlier iteration of the GPT models, while the final version has undergone extensive training.

In a recent instance, an individual successfully solved an image puzzle using what appears to be the complete version of the o1 model.

Remarkably, it took o1 just two minutes to resolve this image puzzle, demonstrating its considerable potential for processing input data, particularly in areas such as images, multimedia files, and documents.

One user showcased the model’s capability by exploring an image of a recent SpaceX rocket launch in detail.

The o1 model accurately analyzed the colors and movements captured in the photograph, showcasing its proficiency in image analysis.

Despite these breakthroughs, it remains uncertain when OpenAI will officially unveil the finalized version of the o1 model.

Experts believe that many companies in the industry are likely waiting for the outcome of the upcoming U.S. presidential elections before announcing any significant advancements.

Our blog thrives on reader engagement. When you make purchases through links on our site, we may earn an affiliate commission.