OpenAI o3-mini Released

OpenAI releases o3-mini, a cost-efficient reasoning model for coding, math and science with three adjustable reasoning efforts.

OpenAI released o3-mini on January 31, 2025, in ChatGPT and the API; OpenAI's release notes record 56% expert preference over o1-mini^[1][2]
For the first time a version is free to ChatGPT users, with rate limits; paid users can pick o3-mini-high, and it supports search with linked web sources^[2]
API access is through Chat Completions, Assistants and Batch APIs^[2]
The same-day news brief also covers free o1 ("Think Deeper") in Microsoft Copilot, a National Laboratories partnership, and reported SoftBank and Stargate investment plans, which are separate from the o3-mini launch^[2]

OpenAI's first small reasoning model to support function calling, Structured Outputs and developer messages, with streaming and a choice of low, medium or high reasoning effort; it does not support vision^[3]
With medium reasoning effort it matches o1 on math, coding and science evaluations including AIME and GPQA; with high effort and a Python tool it solves over 32% of FrontierMath problems on the first attempt^[3]
Responses are 24 percent faster than o1-mini with better accuracy, strong in math, coding and science, and it shows its reasoning^[2]
In A/B testing responses averaged 7.7 seconds against 10.16 seconds for o1-mini, with time to first token 2,500 ms faster, and expert testers saw 39% fewer major errors on difficult real-world questions^[3]
It was trained with deliberative alignment, reasoning about human-written safety specifications before answering^[3]

参考资料 3可信度 92%

第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均；资料每比最新的一份旧 180 天，权重减半。

[1]95%
OpenAI Model Release Notes: Introducing OpenAI o3-mini (January 31, 2025)
help.openai.com· 发布于 2026年9月19日· 开始 2025年1月31日· 占评分 83%
OpenAI's own model release notes date this entry January 31, 2025
[2]70%
The OpenAI Daily Brief: o3-Mini Model, U.S. Energy Collaboration, and SoftBank's Stargate Investment
podcasts.apple.com· 发表于 2025年1月31日· 占评分 8%
The OpenAI^[1] Daily Brief's episode "o3-Mini Model, U.S. Energy Collaboration, and SoftBank's Stargate Investment" (2025-01-31) says: "AI reasoning faster and more accessible than ever!? Welcome to The OpenAI Weekly Brief, your go-to for the latest AI updates. Today is Friday, January 31st , 2025. Here’s what you need to know about OpenAI's groundbreaking launch of the o3-mini reasoning model. Let’s dive in. OpenAI has just unveiled its"
[3]88%
OpenAI o3-mini | OpenAI (Wayback Machine copy)
web.archive.org· 发表于 2025年1月31日· 占评分 8%
Archived copy of OpenAI's^[1] o3-mini announcement, which says it is "our first small reasoning model that supports ... function calling, Structured Outputs, and developer messages", offers low, medium and high reasoning effort, and averaged "7.7 seconds compared to 10.16 seconds" for o1-mini.

有遗漏或错误吗？用你自己的话说明：能佐证此图钉的链接、不同的开始或结束日期及理由，或缺失、有误的信息。AI 会对照此图钉的来源进行核实，搜索更好的来源，并添加任何支持你说法的页面。图钉自身的来源仍然最重要。