OpenAI o3-pro

What o3-pro is

o3-pro is a version of OpenAI's o3 model, released on 10 June 2025^[1]. OpenAI described it as its most capable model at the time and recommended it for challenging questions where reliability matters more than speed^[1]. The release did not come with a separate announcement: it appeared as an update to the post that introduced o3 and o4-mini, where OpenAI described o3-pro as a version of o3 designed to think longer and give the most reliable responses^[6].

What the "pro" adds

The name points to a difference in behaviour rather than a new architecture. A reasoning model works through a problem step by step before it answers; a version built to think longer keeps working on that problem, spending more computation in the moment of the request. OpenAI's model page stated the trade-off directly: o3-pro's reasoning level was rated highest, and its speed slowest^[5].

The recommendation follows from the same choice. A model that spends longer on a task is worth using when a wrong answer costs more than the waiting time, which makes it a model to send hard questions to rather than one to put behind an interactive feature where users expect an immediate reply.

API specifications and price

Through the API, o3-pro accepted text and image input, with a context window of 200,000 tokens and up to 100,000 output tokens; its knowledge cutoff was 1 June 2024^[2]. The context window is the amount of material the model can hold in view at once, the knowledge cutoff marks the point after which it has no built-in information, and the output limit caps how long a single answer can be. Pricing was $20 per million input tokens and $80 per million output tokens^[2].

How it was offered

The model was available only in the Responses API^[5]. OpenAI said this choice supported multi-turn model interactions before responding^[5], so the model could work through an exchange rather than answer from a single prompt. The API snapshot carried the date in its name, o3-pro-2025-06-10, and the model was not offered to free-tier API accounts^[5].

Because answering could take a while, OpenAI warned that some requests might take several minutes and recommended background mode to avoid timeouts^[5]. In background mode a request is handled as a job whose result the client collects later, rather than as a connection held open until the answer is ready — the practical counterpart of a model that is slow by design.

Where it sits in the line

o3-pro was a step in a fast-moving sequence rather than a long-running product. OpenAI's GPT-5 system card names gpt-5-thinking-pro as its successor^[4], which places o3-pro as the reliability-first endpoint of the o3 generation, before the next generation took over that role.

For API users the timetable was explicit: the o3-pro API snapshot was deprecated on 11 June 2026, with shutdown set for 11 December 2026^[3]. Deprecation means the snapshot is no longer recommended for new work but keeps running until the shutdown date, giving existing applications a window to migrate. That schedule, rather than any change in the model itself, marks the end of o3-pro's life as a callable service.