o1-mini launched on 12 September 2024, the same day as o1-preview, and answered the obvious objection to reasoning models: that paying for thinking tokens would make them unaffordable. It cost 1.10 dollars per million input tokens and 4.40 per million output — roughly a fourteenth of o1-preview's rate, at the same 128,000-token context window.
It was also, oddly, the more generous of the two on output: 65,536 maximum output tokens against o1-preview's 32,768. For a model whose reasoning is billed as output, that headroom mattered more than the label "small" suggests. Input was text only, with a knowledge cutoff of October 2023.
OpenAI's own model card eventually recommended against it, pointing developers to o3-mini, which offered higher intelligence at the same latency and the same price. The deprecation was announced on 28 April 2025 with a six-month notice period — twice what o1-preview received — and the model shut down on 27 October 2025, with o4-mini named as the successor.