Artificial Intelligence (AI)

OpenAI launches O3-pro, an upgraded O3 reasoning model

Ken Metral

Ken Metral

— 2 min read
Cosmico - OpenAI launches O3-pro, an upgraded O3 reasoning model
Credit: OpenAI, Inc.

OpenAI has officially launched O3-pro, the latest and most capable version of its AI reasoning model line. Designed to push the boundaries of reliability in complex domains like science, math, and programming, O3-pro is now available to ChatGPT Pro and Team users, with Enterprise and Education customers gaining access next week. The model is also live in OpenAI’s developer API.

What Sets O3-Pro Apart?

Unlike conventional AI models, O3-pro is part of OpenAI’s “reasoning model” family, following the earlier release of the base O3 model. Reasoning models are engineered to approach problems step by step, allowing for more consistent performance in logic-heavy tasks. According to OpenAI, this new release excels particularly in domains such as science, education, business, writing, and coding.

In internal expert evaluations, O3-pro consistently outperformed both its predecessor (O3) and the previous default model (O1-pro), especially in clarity, accuracy, instruction-following, and comprehensiveness.

Availability and Pricing

O3-pro replaces O1-pro as the default model for paying ChatGPT users. It is priced through the API at:

  • $20 per million input tokens
  • $80 per million output tokens

To put that into context, one million input tokens equals roughly 750,000 words—longer than Tolstoy’s War and Peace.

Tools and Capabilities

OpenAI has equipped O3-pro with a broad range of features, including:

  • Web browsing
  • File analysis
  • Visual input reasoning
  • Python code execution
  • Personalization via memory

This makes the model significantly more versatile than earlier versions. However, OpenAI notes that these expanded capabilities come with a trade-off: longer response times compared to O1-pro.

Current Limitations

Despite its strengths, O3-pro is not without limitations:

  • No image generation support
  • Not compatible with OpenAI’s Canvas workspace
  • Temporary chats disabled in ChatGPT due to a technical issue (pending resolution)

These limitations may affect specific workflows but are expected to be addressed in future updates.

Benchmark Performance

In internal benchmarking, O3-pro has achieved notable victories:

  • Outperformed Google’s Gemini 2.5 Pro on the AIME 2024 test (a benchmark for mathematical reasoning)
  • Surpassed Claude 4 Opus by Anthropic on the GPQA Diamond benchmark, which assesses PhD-level science knowledge

A Step Forward in AI Reasoning

O3-pro represents a meaningful evolution in OpenAI’s strategy: moving beyond language generation toward true multi-domain reasoning. With competitive pricing, broad accessibility, and superior performance in expert evaluations, O3-pro is poised to become a key tool for professionals and developers seeking deeper, more accurate AI-driven insights.

As the AI landscape continues to evolve, O3-pro may well redefine what users expect from advanced language models—both in terms of power and precision.

Read more