OpenAI o3-pro: The New Pinnacle of Artificial Intelligence

In June 2025, OpenAI introduced its most capable model to date, o3-pro, which it describes as a milestone in artificial intelligence focused on complex reasoning. The o3-pro model was officially launched on June 10, 2025, and is available to Pro users in ChatGPT and through the API. This model represents the latest advancement in OpenAI's series of "o-models," which specialize in higher levels of logical reasoning and consistency when solving complex tasks. It is an enhanced version of the base o3 model that uses more computational resources to achieve better results on demanding mathematical, scientific, and programming tasks.

Technical Specifications and Key Features

The o3-pro model brings significant technical improvements over its predecessors. It features a context window of 200,000 tokens, enabling it to process extensive conversations and retain deep context throughout long interactions. The maximum number of output tokens per response is 100,000, allowing it to generate highly detailed and comprehensive answers. The model is trained using reinforcement learning, enabling it to "think before answering" and handle advanced tasks requiring logical reasoning. The model's knowledge cutoff is May 31, 2024, ensuring that its information is current up to that date.

Pricing Strategy and Availability

The pricing structure of the o3-pro model reflects its computational demands and advanced capabilities. Input tokens cost $20 per million tokens, while output tokens cost $80 per million. This price is significantly higher than that of the standard o3 model, corresponding to the additional computational resources required to operate it. The model is available to Pro and Team users immediately upon launch, while Enterprise and Edu users will gain access the following week. OpenAI has also implemented special support for "reasoning tokens," which enable advanced processing of logical tasks.

Performance Benchmarks and Comparisons

The o3-pro model has demonstrated exceptional results in academic and real-world benchmark tests. In the 2024 and 2025 AIME (American Invitational Mathematics Examination), it achieved the best results of all available models, outperforming not only previous OpenAI models but also competing solutions such as Google Gemini 2.5 Pro. In mathematics, science, and programming, the model delivers significantly better results than its predecessors, particularly on tasks requiring multi-step reasoning and complex analysis. OpenAI conducted extensive testing with human evaluators, in which o3-pro was preferred over the standard o3 model in 64% of all tested tasks.

Detailed Analysis of User Preferences

Testing with human evaluators revealed o3-pro's consistent superiority across various application domains. In scientific analysis, o3-pro was preferred in 64.9% of cases; in personal writing, 66.7%; in programming, 62.7%; and in data analysis, 64.3%. These results confirm the model's versatility across a range of professional fields. The model demonstrates the ability to autonomously decide when and how to use various tools, such as code execution, data analysis, or web search, to provide detailed and actionable answers, often in less than one minute.

Agentic Reasoning and Advanced Capabilities

One of the o3-pro model's most advanced features is its agentic reasoning capability—its ability to reason autonomously and decide how to approach tasks. The model can independently determine which tools to use to solve a specific problem, including access to web search, data analysis, running Python scripts, image generation, and deep reasoning about visual inputs. This capability represents an important step toward autonomous AI agents that can address multifaceted and open-ended questions more effectively than previous generations of models.

Limitations and Technical Issues

Despite its advanced capabilities, the o3-pro model has certain limitations. It currently does not support image generation and is not compatible with OpenAI Canvas—the workspace feature. As of the launch date, OpenAI temporarily disabled chat with o3-pro in ChatGPT due to technical issues that are being gradually resolved. The model also requires substantial computational resources, which is reflected in its price and may limit availability for some users. However, these technical issues do not detract from the overall progress that the model represents in the field of artificial intelligence.