OpenAI Briefly Exposes Full o1 Model via Simple URL Change

Compiled by the editorial desk with reference to official statements from OpenAI and user reports as cited in Tom’s Guide and Futurism.

OpenAI accidentally gave some users access to the full version of its o1 artificial intelligence model on Friday, only to pull the plug within two hours. The exposure, which came through a simple modification to a web address, was confirmed by the company in a statement to Futurism.

According to a report from Tom’s Guide, users on X (formerly Twitter) discovered that tweaking a URL allowed them to reach the complete o1 model, which had been available only in a limited “preview” form to paying subscribers since its initial announcement in September.

An OpenAI spokesperson acknowledged the incident, saying: “We were preparing limited external access to the OpenAI o1 model and ran into an issue. This has now been fixed.” The company did not provide further details about how many users accessed the model or what specific capabilities were exposed.

Despite the brief window, several users shared their experiences with the full model on social media and forums. One user reported that the model successfully solved a complex math problem and an image puzzle. Another claimed it generated a detailed “full Chain of Thought” when asked to analyze a photo of a recent SpaceX launch. A Reddit user noted that the model handled a large JSON file that the preview version could not process due to token limitations.

Why the Full Release Matters

The o1 model, which was introduced in September, has already drawn attention for its improved performance on standardized tests and its chain-of-thought reasoning capabilities. However, OpenAI has also acknowledged that even its best models still provide incorrect or “hallucinated” answers a significant portion of the time, raising questions about how much the full release will improve reliability.

OpenAI has not announced a specific release date for the full o1 model. Some observers, including Tom’s Guide, speculate that the company may be waiting until after the U.S. presidential election this week to avoid any potential disruption. However, OpenAI has not confirmed this reasoning.

The accidental leak highlights the challenges AI companies face in managing access to powerful models while maintaining security and control. For now, users will have to wait for the official rollout to see whether the full version lives up to the early impressions.