OpenAI Launches o3-Pro, Its Most Capable Reasoning Model Yet

OpenAI has released o3-pro, a new version of its reasoning-focused AI model, calling it the company’s “most capable yet.” The model is now available to ChatGPT Pro and Team users, with access for Enterprise and educational users expected to follow next week. o3-pro is also live in OpenAI’s developer API as of yesterday.

o3-pro builds on the foundation of the o3 model launched earlier this year, which was designed to reason through problems step by step—unlike traditional AI models that often produce answers more quickly but less reliably. This stepwise reasoning approach helps the model perform better in areas like mathematics, physics, and computer programming.

Availability and Pricing

As of Tuesday, o3-pro replaces the older o1-pro model in ChatGPT for paying users and is also available through OpenAI’s developer API.

API pricing is set at:

• $20 per million input tokens

• $80 per million output tokens

For context, one million input tokens equates to roughly 750,000 words—slightly longer than War and Peace.

Performance and Capabilities

In internal evaluations, OpenAI says o3-pro outperforms the base o3 model across all tested categories, with especially strong improvements in science, education, business tasks, programming, and writing support. According to OpenAI’s changelog, reviewers also gave o3-pro higher marks for “clarity, comprehensiveness, instruction-following, and accuracy.”

The model also features tool integration, allowing it to:

• Browse the web

• Analyze files

• Interpret visual inputs

• Run Python code

• Personalize responses using memory

However, OpenAI notes that these enhanced capabilities come with a tradeoff: o3-pro typically generates responses more slowly than o1-pro.

Current Limitations

Despite its advancements, o3-pro launches with several limitations:

• Temporary chats in ChatGPT are disabled due to a technical issue.

• It does not support image generation.

• OpenAI’s collaborative workspace feature, Canvas, is not yet compatible with the model.

Benchmark Performance

o3-pro shows strong results in popular AI evaluation benchmarks. On the AIME 2024 test, which assesses mathematical reasoning, o3-pro outperformed Google’s Gemini 2.5 Pro. It also surpassed Anthropic’s Claude 4 Opus on the GPQA Diamond benchmark, which tests PhD-level science knowledge.

What This Means

The release of o3-pro marks a clear step in OpenAI’s shift toward AI models that prioritize structured reasoning over speed or flash. This direction is especially significant for professional and technical fields—like science, education, and software development—where reliability and step-by-step logic matter more than quick or conversational responses.

By outperforming leading models from Google and Anthropic on benchmarks for math and scientific knowledge, o3-pro positions OpenAI to compete more directly in high-stakes, accuracy-driven applications. These include tutoring, research support, enterprise productivity, and software engineering.

At the same time, the model’s integration with tools like Python, web browsing, and file analysis shows OpenAI’s ambition to make AI not just a chatbot—but a hands-on assistant that can work across workflows. That direction could help embed AI more deeply into daily tasks, especially for knowledge workers who rely on precision and transparency.

While some features are still rolling out, o3-pro’s performance suggests a broader shift: from general-purpose AI toward more dependable systems built for real-world reasoning.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.