Skip to content
BENIBIX Learn

OpenAI o-series

Also known as o1, OpenAI o1, o1-preview, o3, o3-mini, o4-mini

Made by OpenAI. They think it through first, then answer.

Think of it like

It is like a kid who works a sum out on paper first. Then they say the answer.

Example

Jack gives o3 a hard puzzle about train times. It thinks for half a minute, then gives the right answer and its steps.

Why it matters

It showed that more time to think, not just a bigger model, can make answers better.

Definition

The OpenAI o-series is a family of reasoning models made by OpenAI. Before each answer, the model writes a long chain of hidden working. Only then does it give its reply. The first one, o1-preview, came out in September 2024. Later ones include o1, o3-mini, o3 and o4-mini. They do best at maths, science and code, but cost more and reply slower than GPT models.
The o-series always takes the lower path. It spends tokens on working before it answers.

Common mistake

Thinking an o-series model is just a newer GPT. It is a separate line, built to think first, and it is slow on simple tasks.

Key takeaways

  1. 1
    Think first. Hidden working, then a short answer.
  2. 2
    Pay for thinking. The working is billed as output.
  3. 3
    Pick it for hard tasks. Maths, science and code.

Compare

Learn it properly