{{Short description|Large language model}} {{Use mdy dates|date=January 2025}} {{Infobox software | name = o3 | logo = OpenAI o3 icon.png | developer = [[OpenAI]] | released = {{Indented plainlist| * o3-mini: {{Start date|2025|01|31}} * o3: {{Start date| 2025|04|16}}}} o3-pro: {{Start date|2025|06|10}} | latest release version = | latest release date = | genre = {{indented plainlist| * [[Generative pre-trained transformer]] * [[Reasoning language model|Reasoning large language model]] }} | replaces = [[OpenAI o1]] | replaced_by = {{Indented plainlist| * [[OpenAI o4-mini]] * [[GPT-5]] }} | license = }} {{OpenAI series}} '''OpenAI o3''' is a [[Reflection (artificial intelligence)|reflective]] [[generative pre-trained transformer]] (GPT) model developed by [[OpenAI]] as a successor to [[OpenAI o1]] for [[ChatGPT]]. It is designed to devote additional deliberation time when addressing questions that require [[Chain of thought prompting|step-by-step]] logical reasoning.<ref name="auto">{{Cite magazine |last=Knight |first=Will |date=December 20, 2024 |title=OpenAI Upgrades Its Smartest AI Model With Improved Reasoning Skills |url=https://www.wired.com/story/openai-o3-reasoning-model-google-gemini/ |magazine=[[Wired (magazine)|Wired]] |via=}}</ref><ref>{{Cite web |last=Metz |first=Cade |date=2024-12-20 |title=OpenAI Unveils New A.l. That Can 'Reason' Through Math and Science Problems |url=https://www.nytimes.com/2024/12/20/technology/openai-new-ai-math-science.html |website=[[The New York Times]]}}</ref> On January 31, 2025, OpenAI released a smaller model, o3-mini,<ref name=":3" /> followed on April 16 by o3 and [[OpenAI o4-mini|o4-mini]].<ref name=":5">{{Cite web |title=Introducing OpenAI o3 and o4-mini |url=https://openai.com/index/introducing-o3-and-o4-mini/ |access-date=2025-04-16 |website=openai.com |language=en-US}}</ref> ==History== The OpenAI o3 model was announced on December 20, 2024. It was called "o3" rather than "o2" to avoid [[trademark]] conflict with the mobile carrier brand named [[O2 (brand)|O2]].<ref name="auto" /> OpenAI invited safety and security researchers to apply for early access of these models until January 10, 2025.<ref>{{Cite web |date=December 20, 2024 |title=Early access for safety testing |url=https://openai.com/index/early-access-for-safety-testing/ |website=OpenAI}}</ref> Similarly to o1, there are two different models: o3 and o3-mini.<ref name=":3">{{Cite web |last=Franzen |first=Carl |date=2025-01-31 |title=It's here: OpenAI's o3-mini advanced reasoning model arrives to counter DeepSeek's rise |url=https://venturebeat.com/ai/its-here-openais-o3-mini-advanced-reasoning-model-arrives-to-counter-deepseeks-rise/ |access-date=2025-02-01 |website=VentureBeat |language=en-US}}</ref>
On January 31, 2025, OpenAI released o3-mini to all [[ChatGPT]] users (including free-tier) and some [[API]] users. OpenAI describes o3-mini as a "specialized alternative" to o1 for "technical domains requiring precision and speed".<ref name=":4">{{Cite web |date=February 13, 2025 |title=Introducing OpenAI O3 Mini |url=https://openai.com/index/openai-o3-mini/ |access-date=February 13, 2025 |website=OpenAI}}</ref> o3-mini features three reasoning effort levels: low, medium and high. The free version uses medium. The variant using more compute is called o3-mini-high, and is available to paid subscribers.<ref name=":3" /><ref>{{Cite web |date=January 31, 2025 |title=OpenAI o3-mini |url=https://openai.com/index/openai-o3-mini/ |access-date=2025-02-03 |website=OpenAI |language=en-US}}</ref> Subscribers to ChatGPT's Pro tier have unlimited access to both o3-mini and o3-mini-high.<ref name=":4" />
On February 2, OpenAI launched [[OpenAI Deep Research]], a ChatGPT service using a version of o3 that makes comprehensive reports within 5 to 30 minutes, based on [[Web search|web searches]].<ref>{{Cite web |last=Ha |first=Anthony |date=2025-02-03 |title=OpenAI unveils a new ChatGPT agent for 'deep research' |url=https://techcrunch.com/2025/02/02/openai-unveils-a-new-chatgpt-agent-for-deep-research/ |access-date=2025-02-04 |website=TechCrunch |language=en-US}}</ref>
On February 6, in response to pressure from rivals like [[DeepSeek]] R1, OpenAI announced an update aimed at enhancing the transparency of the thought process in its o3-mini model.<ref>{{Cite web |last=Wiggers |first=Kyle |date=2025-02-06 |title=OpenAI now reveals more of its o3-mini model's thought process |url=https://techcrunch.com/2025/02/06/openai-now-reveals-more-of-its-o3-mini-models-thought-process/ |access-date=2025-02-07 |website=TechCrunch |language=en-US}}</ref>
On February 12, OpenAI further increased rate limits for o3-mini-high to 50 requests per day (from 50 requests per week) for ChatGPT Plus subscribers, and implemented file/image upload support.<ref name="openai-tweet">{{Cite tweet |number=1889822643676913977 |user=OpenAI |title=Two updates you'll like— OpenAI o1 and o3-mini now support both file & image uploads in ChatGPT. We raised o3-mini-high limits by 7x for Plus users to up to 50 per day. |date=February 13, 2025 |access-date=February 13, 2025 |url=https://x.com/OpenAI/status/1889822643676913977}}</ref>
On April 16, 2025, OpenAI released o3 and [[OpenAI o4-mini|o4-mini]], a successor of o3-mini.<ref name=":5" />
On June 10, OpenAI released o3-pro, which the company claims is its most capable model yet.<ref>{{Cite web |last=Wiggers |first=Kyle |date=2025-06-10 |title=OpenAI releases o3-pro, a souped-up version of its o3 AI reasoning model |url=https://techcrunch.com/2025/06/10/openai-releases-o3-pro-a-souped-up-version-of-its-o3-ai-reasoning-model/ |access-date=2025-06-11 |website=TechCrunch |language=en-US}}</ref> OpenAI stated: "We recommend using it for challenging questions where reliability matters more than speed, and waiting a few minutes is worth the tradeoff".<ref>{{Cite web |last=Washenko |first=Anna |date=2025-06-06 |title=OpenAI adds the o3-pro model to ChatGPT today |url=https://www.engadget.com/ai/openai-adds-the-o3-pro-model-to-chatgpt-today-212126136.html |access-date=2025-06-11 |website=Engadget |language=en-US}}</ref>
On May 28, 2026, OpenAI announced that o3 would be retired from ChatGPT on August 26, 2026, following a 90-day sunset period. The company stated that the change applied only to ChatGPT and did not affect the API.<ref>{{cite web |title=ChatGPT — Release Notes |url=https://help.openai.com/en/articles/6825453-chatgpt-release-notes |website=OpenAI Help Center |publisher=OpenAI |date=May 28, 2026 |access-date=May 30, 2026}}</ref>
==Capabilities== [[Reinforcement learning]] was used to teach o3 to "think" before generating answers, using what OpenAI refers to as a "private [[Chain of thought prompting|chain of thought]]".<ref name="o3-mini-card">{{Cite report |url=https://cdn.openai.com/o3-mini-system-card-feb10.pdf |title=OpenAI O3 Mini System Card |date=January 13, 2025 |publisher=OpenAI |access-date=February 13, 2025 }}</ref> This approach enables the model to plan ahead and reason through tasks, performing a series of intermediate reasoning steps to assist in solving the problem, at the cost of additional computing power and increased [[Latency (engineering)|latency]] of responses.<ref name=":1">{{Cite web |last1=Zeff |first1=Maxwell |last2=Wiggers |first2=Kyle |date=2024-12-20 |title=OpenAI announces new o3 models |url=https://techcrunch.com/2024/12/20/openai-announces-new-o3-model/ |access-date=2024-12-22 |website=[[TechCrunch]] |language=en-US}}</ref>
o3 demonstrates significantly better performance than o1 on complex tasks, including [[computer programming|coding]], [[mathematics]], and [[science]].<ref name="auto" /> OpenAI reported that o3 achieved a score of 87.7% on the GPQA Diamond benchmark, which contains expert-level science questions not publicly available online.<ref name=":2">{{Cite web |last1=Franzen |first1=Carl |last2=David |first2=Emilia |date=2024-12-20 |title=OpenAI confirms new frontier models o3 and o3-mini |url=https://venturebeat.com/ai/openai-confirms-new-frontier-models-o3-and-o3-mini/ |access-date=2024-12-26 |website=[[VentureBeat]] |language=en-US}}</ref>
On SWE-bench Verified, a [[software engineering]] [[List of language model benchmarks|benchmark]] assessing the ability to solve real [[GitHub]] issues, o3 scored 71.7%, compared to 48.9% for o1. On [[Codeforces]], o3 reached an [[Elo rating system|Elo]] score of 2727, whereas o1 scored 1891.<ref name=":2" />
On the Abstraction and Reasoning Corpus for Artificial General Intelligence (ARC-AGI) benchmark, which evaluates an AI's ability to handle new logical and skill acquisition problems, o3 attained three times the accuracy of o1.<ref name="auto" /><ref name=":0">{{Cite web |last=Hsu |first=Jeremy |date=20 December 2024 |title=OpenAI's o3 model aced a test of AI reasoning – but it's still not AGI |url=https://www.newscientist.com/article/2462000-openais-o3-model-aced-a-test-of-ai-reasoning-but-its-still-not-agi/ |access-date=2024-12-22 |website=[[New Scientist]] |language=en-US}}</ref>
=== o3-mini performance === According to OpenAI's January 2025 report on o3-mini, adjusting "reasoning effort" significantly affects performance, especially for [[STEM]] tasks. Moving from low to high reasoning effort raises accuracy on OpenAI's [[American Invitational Mathematics Examination|AIME]] 2024 (different from the MathArena AIME benchmark), GPQA Diamond, and [[Codeforces]], typically by 10–30%. With high effort, o3-mini (high) achieved 87.3% on AIME 2024, 79.7% on GPQA Diamond, 2130 Elo on Codeforces, and 49.3 on SWE-bench Verified.<ref name=":4"/>
==See also== * [[Reasoning model]] * [[List of large language models]]
==References== {{Reflist}}
==External links== *[https://openai.com/index/introducing-o3-and-o4-mini/ ''Introducing OpenAI o3 and o4-mini''] *[https://community.openai.com/t/o3-is-80-cheaper-and-introducing-o3-pro/1284925 ''O3 is 80% cheaper and introducing o3-pro'']
{{OpenAI}}
[[Category:Large language models]] [[Category:2024 software]] [[Category:Generative pre-trained transformers]] [[Category:OpenAI]] [[Category:ChatGPT]] [[Category:2025 in artificial intelligence]]