Opeenai’s O3-MINI think

Openai has released a new time for Amnesty International in time to face the rapid rise in its open source opponent Deepseek R1 – but would that be enough to achieve the success of the latter?

Today, after several days of rumors and increased anticipation between artificial intelligence users on social media, Openal for the first time O3-MiniThe second model in his new family of “logic”, models that take a little longer “thinking”, analyzing their own processes and thinking about their “thought chains” before responding to user inquiries and inputs with new outputs.

The result is a model that can lead at the level of a doctorate student or even a degree holder to answer difficult questions in mathematics, science, engineering and many other fields.

The O3-MINI model is now available on ChatGPT, including the free level, an Openai (API) application program, which is actually less expensive, faster and more performance than the previous high-end, Openai’s O1, the number of lower parameters, brother, O1-MINI.

Although it will be impartially compared to Deepseek R1, and the date of the version seen as a reaction, it is important to remember that O3 and O3-MINI were announced before the Deepseek R1 version in January, in December 2024-and the CEO of Openai Sam Al -Taman previously stated X Because of the reactions of developers and researchers, it will come to Chatgpt and Openai API at the same time.

Unlike Deepseek R1, O3-MINI will not be provided as an open source model-which means that the symbol cannot be taken and downloaded for use without connection, nor customized by the same amount, which may limit its attractiveness compared to Deepsek R1 for some applications.

Openai did not provide any other details about the old (supposed) O3 model that was announced again in December alongside O3-MINI. At that time, the Openai model from the O3 test drop -down menu stated that it will undergo “several weeks delay” before the third parties can test it.

Performance and features

Like O1, Openai O3-MINI has been improved to think about mathematics, coding and science.

Its performance is similar to Openai O1 when using an average thinking voltage, but it offers the following advantages:

• 24 % response times faster compared to O1-MINI (Openai has not provided a specific number here, but looking at the third-party evaluation group Artificial analysis tests, O1-MINI response time is 12.8 seconds To receive and remove 100 icons. So for the O3-MINI, 24 % velocity scattering will decrease the response time to 10.32 seconds.)
• Improved accuracy, with external tests prefer O3-MINI responses 56 % of time.
• 39 % less than the main errors of the complex real world questions.
• Better performance in coding tasks and scientific skills, especially when using a high thinking voltage.
• Three levels of thinking (low, medium and high), allowing users and developers to balance accuracy and speed.

It also includes impressive criteria, even O1 in some cases, according to the O3-MINI Openai system that was released online (which was published before announcing the availability of the official model).

Window Contest O3-MINI-the number of integrated symbols that can enter/direct in one reaction-is 200,000, with a maximum of 100,000 per output. This is the same as the full O1 model and It exceeds the window of Deepseek R1 From about 128,000/130,000 symbols. But much less than the new Google Gemini 2.0 Flash window window, which reaches a million icons.

While O3-MINI focuses on thinking capabilities, it has no vision capabilities yet. Developers and users who look forward to downloading pictures and files should continue to use O1 in the meantime.

The competition is heated

The O3-MINI Marks arrival for the first time Openai is available to free the thinking model for free Chatgpt users. The former O1 family was only available for Chatgpt Plus, Pro and other plans, as well as via Openai’s application programming interface.

As was the case with large language models (LLM)-Power by launching ChatgPT in November 2022, Openai was mainly established in the full category of thinking forms in September 2024 when it revealed O1 for the first time, which is a new category of models with new system training And architecture.

But Openai, in line with its modern history, did not make O Open Source, unlike his name and original founding mission. Instead, keep the royal model symbol.

Over the past two weeks, O1 has been overwhelmed by the Chinese company Deepseek AI, which launched the R1, which is a very competitive thinking model, and the source is largely open to a freedom to take, re -train, and allocate by anyone around the world, also using free of charge Deepseek on the web and mobile application – it is said that the model has been trained on a small part of the cost of the other Top Labs.

Deepsek R1’s Conditions of the Massachusetts Institute of TechnologyApplication/website free for consumers, and the decision to provide the R1 code is available for free to take and amend it to a real explosion for use in consumer markets and institutions – even Openai investor Microsoft and Antarbur Amazon supporter Hurry to add variables from it to their cloud markets. Confusion, the search company from artificial intelligence, quickly It has been added a variable to users.

Dibsic also He canceled the Chatgpt iOS application for the first place in the Apple App Store in the United StatesAnd it is noted for Openai’s superiority by connecting its R1 to search on the Internet in its application and on the web, something that Openai has not yet done for O1, which leads to the United States in creating artificial intelligence – even technology in general.

However, many researchers and scientists welcomed the artificial intelligence and the highest VCS such as Marc Andressen, to the appearance of Deepseek and its open sources in particular as accuracy that raises all boats in the artificial intelligence field, which increased the intelligence available to all while reducing costs.

Available in ChatGPT

The model is now launched in the world to users, in addition to the team and the professionals, with the arrival of institutions and education next week.

• Liberal users can first experience the O3-MINI for the first time by selecting the “Mind” button in the chat bar or renewing response.
• The 3x messages limits for Plus and Team users increased, up from 50 to 150 messages per day.
• Professional users get unlimited access to both O3-MINI and a new variable, up to the highest level, O3-MINI.

In addition, O3-MINI now supports research integration within ChatGPT, providing responses with relevant web links. This feature is still in its early stages as Openai improves research capabilities through thinking models.

API integration and pricing

For developers, O3-MINI is available via API API, API assistants and API. The model supports the summons of the job, organized outputs, and developers, which makes it easy to integrate into real world applications.

One of the most noticeable O3-MINI advantages is its cost efficiency: it is 63 % cheaper than Openai O1-MINI and 93 % cheaper than the full O1, at $ 1.10/$ 4.40 per million codes in/out (with 50 % discount from Temporary storage).

However, it continues to diminish compared to The ability to bear the costs from the official Deepseek applications interfaceR1 offer at $ 0.14/0.55 dollars per million icons in/out. But given that Deepseek is based in China and comes with geopolitical awareness and security concerns about user/institution data that flows to the model and abroad, Openai is likely to remain the preferred application programming interface for some customers and institutions that focus on safety in the United States and Europe.

Developers can also adjust the level of thinking voltage (low, medium, high) based on their application needs, allowing further control of cumin and accuracy bodies.

On safety, Openai says it uses something called “deliberative alignment” with O3-MINI. This means that the model was asked to think about the safety guidelines that were given the human being given, to understand more of its intention and the damage that was designed to prevent it, and to reach its own ways to ensure the prevention of those damage. Openai says it allows the model to be less sensitive when discussing sensitive topics while maintaining safety as well.

Openai says the model is outperforming GPT-4O in dealing with safety and crows challenges, and that he conducted a wide external test for safety before the issue today.

A The last report covered in Wireless (Where my wife works) I showed that Deepseek surrendered to every jailbreak router and an attempt among 50 security researchers, which may give Openai O3-MINI the edge on Deepseek R1 in cases where security and safety is very important.

What the next?

The launch of the O3-Mini is the broader Openai efforts to make advanced logic easier and costly in terms of more intense competition than ever before Deepseek’s R1 and others, such as Google, which recently released a free version of her own contemporary thinking model 2 flash thinking with The context of an expanded entry of up to one million symbols.

With its focus on thinking about STEM and the ability to withstand costs, Openai aims to expand the scope of solving the problems that AI drives in both consumer applications and developers.

But when the company becomes more ambitious than ever in its goals – it recently announced the $ 500 billion data infrastructure project called Stargate with support from SoftBank – the question remains whether its strategy will pay well enough to justify the multiple millions to it by that by that by that by that by that By deep investors such as Microsoft and other VCS.

Since the open source models are increasingly blocking the gap with Openai in performance and its superiority in the cost, the superior safety measures, strong capabilities, and the facade of the applications programming are easy to use and easy to use sufficient to maintain customers- especially in the institution- who may give cost priority and efficiency on These features? We will report developments because they are revealed.