OpenAI’s Latest AI Model Enhances Reasoning Capabilities
In a bold step highlighting the growing rivalry among generative AI providers, OpenAI new AI features include an advanced reasoning model and an innovative agentic AI capability. This release follows the recent breakthrough by Chinese startup DeepSeek, which has gained attention in the U.S. tech industry with its widely recognized open-source reasoning model.
On January 31, OpenAI introduced o3-mini, an enhanced iteration of its proprietary o1 reasoning model launched in September. The o3-mini model offers advanced developer features, including function calling, structured output, and developer messages. Function calling enables large language models (LLMs) to interface with external tools or data sources by identifying predetermined functions. Structured output allows LLMs to generate responses in established formats like JSON, while developer messages provide specific instructions to the LLM from developers.
The o3-mini model is currently available to ChatGPT Plus, Team, and Pro users, with plans to extend access to enterprise users later this month. Free ChatGPT users can experience the model by selecting the ‘Reason’ option in the message composer.
Shortly after the release of o3-mini, OpenAI launched deep research, a new agentic mode within ChatGPT designed to conduct comprehensive internet-based research for complex tasks. This feature targets professionals in fields such as finance, science, policy, and engineering, enabling the AI agent to perform multi-step research processes that can take between five to thirty minutes. Currently, deep research is available to Pro users, with upcoming availability for Plus and Team users.
Bradley Shimmin, an analyst at Omdia, noted, “We’re seeing a lot more capabilities when we give models time to think, and also when we give them multimodal capabilities such as being able to see and type.”
The introduction of these features by OpenAI occurs amidst ongoing discussions about DeepSeek’s recent advancements. DeepSeek’s ability to train its new V3 model using time-test scaling—with reportedly less computational power and at a lower cost than established generative AI vendors—has led some to question OpenAI’s position as the leading innovator in the field. However, there is speculation regarding the transparency of DeepSeek’s claims about the computational resources and AI chips utilized in developing V3.
While OpenAI has been a pioneer in certain technologies, it has been slower to market in others. For instance, its deep research capability bears resemblance to Anthropic’s computer use tool, which was introduced in October of the previous year. Additionally, OpenAI appears to be catching up in offering agentic capabilities that operate on behalf of users. Despite this, Shimmin emphasized that OpenAI maintains a lead in other areas, citing the October release of Swarm, a system designed to enhance collaborative AI functionalities.
As the generative AI landscape evolves, OpenAI’s recent developments underscore its commitment to advancing AI reasoning and agentic capabilities, even as new competitors emerge and challenge the status quo.
