OpenAI is set to release a groundbreaking artificial intelligence model, known internally as “Strawberry,” aimed at performing advanced human-like reasoning tasks. This move comes as the company seeks to maintain its lead in an increasingly competitive AI market.
The new model, officially named “o1,” is designed to invest more time in computational analysis before delivering answers to user queries, according to a blog post from OpenAI on Thursday. This enables the model to solve complex multi-step problems, including intricate math and coding tasks.
“As an early version, o1 doesn’t yet possess many of the features that make ChatGPT widely useful, such as browsing the web or handling uploads of files and images,” OpenAI stated. “However, for complex reasoning tasks, this represents a significant leap in AI capabilities. We’ve reset the numbering to reflect this new level of advancement, and are introducing the OpenAI o1 series.”
A preview version of the o1 model will be available exclusively to paid subscribers of ChatGPT’s Plus and Team plans starting Thursday. Bloomberg had previously reported the anticipated launch could happen this week.
The release of this new model comes at a pivotal time for OpenAI, as the San Francisco-based company aims to secure billions in additional funding while facing increased competition in the AI space. Companies like Anthropic and Google are also pushing the boundaries of AI “reasoning” capabilities with their respective models.
OpenAI shared several examples of o1’s performance, including its ability to tackle coding challenges, answer complex English language queries, and solve math problems. In a series of posts on social platform X, OpenAI research scientist Noam Brown explained that releasing the model in preview will help the company gather feedback and identify areas for improvement.
The new AI model introduces a distinct experience for users familiar with ChatGPT. Before responding, o1 pauses briefly while it considers multiple related prompts, a process known as “chain of thought” prompting. This approach allows it to generate more thoughtful and accurate responses.
OpenAI has been working toward developing AI systems that can handle multi-step reasoning for some time. In May 2023, the company published a research paper detailing efforts to enhance AI’s ability to solve math problems by rewarding models for each correct step in problem-solving, not just for the final answer.
As OpenAI continues to push the boundaries of AI technology, the introduction of o1 signals a new chapter in the evolution of advanced reasoning capabilities.