OpenAI Introduces The 'o3-Pro' That Uses Even More Computing Power To Reason And Think

From its earliest days as a pioneering startup with big ideals, OpenAI has been the vanguard in transforming large language models (LLMs) from academic curiosities into everyday tools.

When it announced GPT-3.5-powered ChatGPT, it literally blew everyone away. Then, the release of GPT-4 and subsequent “o-series” models only accelerated this trajectory—sparking a race among tech giants and startups alike to outpace each other in AI prowess and capabilities.

In that feverish arms race, the world has seen iterative leaps.

From GPT-4’s fluent interactions, Anthropic’s safety-first Claude, Google’s Gemini with its Google Search power, and open-source challengers, they're all racing to keep up—all vying to push the frontier of machine reasoning.

Into that vibrant ecosystem stepped OpenAI’s o3, a “reasoning model” designed not to respond instantly but to think through problems with an internal chain of thought—solving complex tasks in math, science, code, and beyond.

And because the o3 simply isn't enough, OpenAI introduced 'o3-pro,' which steps up the game.

When launched in June 10, 2025, the o3‑Pro is OpenAI's most potent reasoning model yet.

Replacing o1‑Pro for ChatGPT Pro and Team users—with API access already live—it represents a deliberate “think longer, think deeper” shift.

To create this model, trained the o3-pro with reinforcement learning to think before they answer and perform complex reasoning, just like its predecessor reasoning model. But here, the o3-pro model uses more compute to think harder and provide consistently better answers.

OpenAI also taps into its "4/4 reliability" evaluation, where a model is considered successful only if it correctly answers a question in all four attempts, not just one.

Internal evaluations reveal consistent preference for o3‑Pro across challenging domains like science, programming, business strategy, and writing—scoring higher in clarity, comprehensiveness, and accuracy than its predecessors.

"Like OpenAI o1-pro, OpenAI o3-pro excels at math, science, and coding as shown in academic evaluations," said OpenAI.

Under the hood, o3‑Pro inherits the full range of capabilities introduced with the o3 model—this includes web browsing, code execution via Python, advanced document and image analysis, and memory features that allow for context-rich, intelligent responses.

However, since the o3-pro goes a step beyond the normal o3, its enhanced reasoning power comes with compromises. Response times are noticeably slower, and several features—such as temporary chats, image generation, and Canvas—remain unavailable for now.

As noted by OpenAI on a dedicated webpage on its website, o3‑Pro leads all current models in reasoning performance, yet it is also the slowest in terms of response speed.

"Since o3-pro is designed to tackle tough problems, some requests may take several minutes to finish. To avoid timeouts, try using background mode," explained OpenAI.

The o3‑pro model is currently available exclusively through the Responses API, enabling support for multi-turn model interactions before generating a final response—paving the way for more advanced API features in the near future.

It's priced at $20 per million input tokens and $80 per million output tokens when accessed via the API. Input tokens refer to the data sent to the model, while output tokens are the responses generated by it. For context, one million input tokens is approximately equal to 750,000 words.