OpenAI’s o3-Mini - a new learning AI Model Competing With DeepSeek

Boyuan Qian

03 Feb 2025 — 2 min read

In response to the launch of DeepSeek's R1, OpenAI has introduced a more advanced and cost-effective model called o3-mini. This new release aims to match the excitement surrounding the open-source offerings from the Chinese AI startup.

OpenAI has been preparing the o3-mini for its debut on January 31. According to confidential sources, the company’s researchers have been diligently working to ensure its readiness for public use.

Teased in December, the o3-mini is a compact version of OpenAI's most sophisticated AI model, boasting exceptional reasoning capabilities. It is designed to deconstruct complex problems into manageable components, enabling effective problem-solving.

Availability and User Access

OpenAI will provide o3-mini to Plus, Team, and Pro users of ChatGPT. Free-tier users can also access the model, albeit with some limitations on query volume.

To enhance its capabilities, OpenAI has been engaging PhD students to assist in training the model. Recently, the company has been recruiting computer science PhD candidates at $100 per hour for a “research collaboration” focused on developing unreleased models. Additionally, OpenAI has sought expertise from PhD students in various fields through a staffing firm called Mercor, which has been tasked with creating challenging scientific coding questions to evaluate the performance of large language models in solving realistic research problems.

Industry Context and Competition

The emergence of DeepSeek’s R1 has intensified competition within the US tech sector. The availability of such a powerful model for free is putting pressure on companies like Google and Anthropic to reconsider their pricing strategies. OpenAI aims to reaffirm its leadership in AI development and commercialization amidst this competitive landscape.

DeepSeek's model has been noted for its efficiency in both training and deployment, reportedly requiring fewer resources compared to OpenAI and other US firms developing cutting-edge AI technologies. While specific details about DeepSeek's resource allocation remain undisclosed, there are indications that R1 may have utilized outputs from OpenAI's models during its training phase.

Features and Future Directions

Although OpenAI's o3-mini may not surpass R1 in affordability, it emphasizes the company's commitment to enhancing efficiency in its models. OpenAI notes that o3-mini excels particularly in mathematics, science, and coding.

The latest model will introduce several new features, including:

Integration with web searches
Capability to invoke functions from user code
Options to switch between different reasoning levels, balancing speed with problem-solving depth

DeepSeek's rapid ascent has also sparked discussions about the US government's approach to managing China's growth in AI technology. Over the past two administrations, the US has implemented sanctions aimed at limiting China's access to advanced Nvidia chips, which are essential for developing state-of-the-art AI models. While DeepSeek has referenced various Nvidia chips in its research, the specifics of their utilization remain unclear.

This evolving landscape showcases the competitive and strategic maneuvers within the AI industry as key players adapt to new challenges and innovations.