On the last day of the 12 days of ‘Shipmas’ of Openai, the company presented its latest models, O3 and O3-mini, which stand out in reasoning and even exceed O1 in a series of reference points, including mathematics and sciences. At the launch, the Operai CEO, Sam Altman, said that the O3 was scheduled to fall at the end of January, and today, the company fulfilled its promise.
O3-mini
Friday, OpenAi release Its O3-mini model, the most profitable model in the OpenAI reasoning series, to the public. Until now, that series has been composed of O1 and O1-mini. Like its predecessor, the model is particularly strong in science, mathematics and coding, according to the company.
Operai O3-mini is now available in Chatgpt and API.
Professional users will have unlimited access to U3-MINI and Plus & Workforce users will have triple tariff limits (compared to O1-mini).
Free users can try O3-MINI in chatgpt by selecting the reason button in the message composer.– OpenAi (@openai) January 31, 2025
When O3-MINI is selected, it will use a medium reasoning effort, which balances speed and precision. Although the O1 Unique model still has a broader normal knowledge than O3-mini, the main advantage of the new model is its fastest speed and its highest performance compared to O1-mini.
Reference performance
When comparing the O3-mini performance with O1-mini, expert testers found that O3-mini delivered more precise, reasoned and clearer answers than O1-mini. According to the Submit, 56% of the time preferred O3-MINI and observed a 39% reduction in important errors.
Beyond the evaluations of human preferences, at several STEM reference points, including competition mathematics (Aime 2024), doctoral level science questions (GPQA Diamond) and the Competition Code (Codeforces), O3-MINI with Average reasoning, which is what Chatgpt users will do by default: O1-mini overcome.
It is also remarkable that O3-mini, with a great reasoning effort at the reference points, approached the O1 yield, sometimes even overcome it, as seen in the previous 2024 Aime (verified by the SWE bank). The O3-mini model with medium reasoning effort coincided with the O1 performance at the codeforces reference point.
Security
Operai evaluated the security of O3-mini through public launch through Jailbreak and not allowed content evaluations. The company discovered that the model exceeds GPT-4O in evaluations. Operai published the evaluation results below and also launched an O3-mini system card, a 37 pdf pages That includes the detailed results of the evaluations.
How to access
All subscribers of OpenAI payments, including Chatgpt Plus, Workforce and Professional, can access OpenAi O3-mini as of today. In addition, team users now have the rate limit three times, which passes from 50 messages per day with O1-mini to 150 messages per day. Chatgpt Enterprise Entry will come in a week.
Also: the powerful feature of Copilot ‘Suppos Deer’ is free for all users: how it works
The O3-MINI model will replace O1-mini in the model selector, since it would be useful for the same tasks, except that the experience will now improve with a lower latency and faster limits. As a paid user, at the time of writing, I still did not have access to the O3-mini, and I am still seeing the O1-mini option.
If you do not have a subscription, do not worry: you can see if O3-mini is worth exaggerating your free account. All that chatgpt users have to do is click on “Reason” in the message text box or regenerate an answer. The Operai CEO, Sam Altman, confirmed free access in a Publish in x. So far, all reasoning models have remained behind a payment wall; Openai did not specify any limitation around the new model for free users.
