OpenAI releases o3-mini

OpenAI released o3-mini, a smaller and cheaper successor to o1-mini offered ahead of the full o3 model announced the previous month. It came with three selectable reasoning-effort levels — low, medium and high — letting users trade response speed against the amount of step-by-step reasoning the model performed before answering, with ChatGPT defaulting to medium.

The release was notable chiefly for access and pricing rather than raw capability. OpenAI made o3-mini the first reasoning model available to free ChatGPT users, extending a capability previously limited to paying subscribers, and priced API access at $1.10 per million input tokens and $4.40 per million output tokens — which the company said was 63% cheaper than o1-mini, though still roughly double the price DeepSeek was charging for R1, the Chinese reasoning model that had unsettled markets ten days earlier. Reporting at the time framed the free rollout explicitly as a competitive response to R1’s rapid uptake.

Coverage also noted a safety trade-off that OpenAI’s own evaluations acknowledged: o3-mini scored higher than earlier non-reasoning models on tests of persuasion and on resistance to jailbreaking attempts, which OpenAI’s preparedness framework treated as a “medium” risk rating rather than the “high” threshold that would have triggered additional safeguards before release.

The model was superseded a few months later when OpenAI released the full-scale o3 alongside o4-mini, which added the ability to use tools such as web browsing and code execution mid-reasoning — a capability o3-mini itself lacked.