OpenAI Unveils o1-preview, a New AI Model Built for Complex Reasoning

Compiled by the editorial desk with reference to official statements, public filings and industry data.

OpenAI has officially released its long-anticipated AI model, previously known by the code name “Strawberry.” The new system, branded “OpenAI o1-preview,” marks a departure from the company’s naming convention and is designed to “spend more time thinking” before generating responses, according to an official company update. The model aims to handle “complex tasks” and “harder problems” than earlier versions, signaling a shift toward deeper reasoning capabilities.

The release has been the subject of speculation for months, with many in the AI community viewing it as a potential step toward artificial general intelligence (AGI) — the theoretical point where machines could outperform humans across a wide range of cognitive tasks. OpenAI’s focus with o1-preview is to give the model a form of “reasoning,” allowing it to tackle more challenging problems, such as advanced mathematics. The company claims that the model “performs similarly to PhD students on challenging benchmark tasks in physics, chemistry, and biology.”

Despite these capabilities, OpenAI is clear that o1-preview is still an early-stage product. “As an early model, it doesn’t yet have many of the features that make ChatGPT useful, like browsing the web for information and uploading files and images,” the company wrote. “For many common cases GPT-4o will be more capable in the near term.” This admission suggests that while o1-preview excels in specific domains, it may not replace existing models for everyday use.

Availability and Access

Starting now, OpenAI o1-preview is available to ChatGPT Plus and Team subscribers. The company also plans to release a more streamlined version, called o1-mini, to all free users of ChatGPT, though no specific timeline has been provided. This tiered rollout mirrors OpenAI’s previous strategy of offering premium features to paying customers before expanding access.

In terms of safety, OpenAI says it designed the new model with security as a priority. In one of its “hardest jailbreaking tests,” o1-preview scored 84 out of 100, compared to just 22 for GPT-4o, its predecessor. The company attributes this improvement to a “completely new optimization algorithm and a new training dataset specifically tailored for it,” according to Jerry Tworek, OpenAI’s research lead, speaking to The Verge.

How It Works and Potential Uses

The model employs a “chain of thought” process, evaluating multiple possible answers before selecting the best one. This method can be time-consuming — in a demo seen by The Verge, o1-preview took 30 seconds to solve a reasoning puzzle involving the ages of a princess and a prince. In contrast, standard ChatGPT responses are often nearly instantaneous. This trade-off between speed and depth is a key characteristic of the new model.

OpenAI suggests that o1-preview could be useful in professional settings, such as “healthcare researchers to annotate cell sequencing data, by physicists to generate complicated mathematical formulas needed for quantum optics, and by developers in all fields to build and execute multi-step workflows.” These examples highlight the model’s potential in specialized, high-stakes environments.

On the issue of hallucination — a persistent problem in AI chatbots where models generate false or fabricated information — OpenAI offered a measured response. “We have noticed that this model hallucinates less,” Tworek told The Verge, but “we can’t say we solved hallucinations.” This candid acknowledgment reflects the ongoing challenges in AI reliability.

OpenAI CEO Sam Altman also commented on the release, describing the model as “still flawed, still limited, and it still seems more impressive on first use than it does after you spend more time with it.” When asked directly whether o1-preview represents a breakthrough toward AGI, Altman responded simply, “No.”

The launch of o1-preview comes amid intense competition in the AI sector, with companies like Google and Anthropic also pushing the boundaries of model capabilities. While the model’s long-term impact remains to be seen, its focus on reasoning represents a notable evolution in how AI systems are trained and deployed.