Opnai’s O3-mini reasoning model comes to counteract deep consumption

Operai has launched a new patented model in time to counteract the rapid rise of the Deepseek-R1 open source rival, but will it be enough to avoid the success of the latter?

Today, after several days of rumors and a growing anticipation among the users of AI on social networks, Openal is debuting O3-miniThe second model in its new family of “Reasoners”, models that take a little longer to “think”, analyze their own processes and reflect on their own “chains of thought” before responding to user consultations and tickets with tickets with new outputs.

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

The O3-mini model is now available in Chatgpt, including the free level and in the OpenAi Application Programming Interface (API). And, in reality, it is less expensive, faster and more active than the previous high-end model, Openai’s O1, and its lower parameter counting brother faster, O1-mini.

While it will inevitably be compared to Depseek-R1, and the launch date is seen by some as a reaction, it is important That Operai CEO Sam Altman previously declared in X That due to the comments of developers and researchers, it would arrive at Chatgpt and the OpenAI API at the same time.

Unlike Deepseek-R1, O3-MINI will not be available as an open source model, which means that the code cannot be downloaded for out-of-line use, nor customized to the same extent, which can limit its attractiveness in Comparison with Deepseek-R1 for some applications.

Operai did not provide more details about the (alleged) largest O3 model announced in December along with O3-mini. At that time, the OpenAI option drop -down form to prove O3 declared that it would undergo a “several weeks late” before third parties could try it.

Performance and characteristics

Similar to O1, Operai O3-mini is optimized for reasoning in mathematics, coding and science.

Its performance is comparable to OpenAI O1 when an average reasoning effort is used, but offers the following advantages:

• 24% faster response times compared to O1-MINI (OpenAI did not provide a specific number here, but observing a third-party evaluation group Artificial analysis testsThe O1-mini response time is 12.8 seconds to receive and generate 100 tokens. Then, for O3-mini, an increase in speed of 24% would reduce the response time to 10.32 seconds).
• Improved precision, with external testers that prefer O3-MINI responses 56% of the time.
• 39% less important errors in complex questions of the real world.
• Better performance in coding and Stem tasks, particularly when a reasoning effort is used.
• Three levels of reasoning effort (low, medium, high), allowing users and developers to balance precision and speed.

The model also has impressive reference points, even surpassing O1 in some cases, according to the O3-mini openai system launched online (which was published before the official announcement of model availability).

The O3-mini context window, the number of combined tokens that can enter/exit into a single interaction, is 200,000, with a maximum of 100,000 at each output. That is the same as the complete O1 model and has a higher performance Veteran–R1 context window of around 128,000/130,000 tokens. But, it is well below Google Gemini 2.0 Flash Thinking, the new context window of up to 1 million tokens.

While O3-mini focuses on reasoning capabilities, he still has no vision capabilities. Developers and users seeking to load images and files must continue to use O1 in the meantime.

Competition is heated

The arrival of O3-mini brands The first time OpenAi is making a reasoning model available for free chatgpt users. The previous O1 models was only available to pay chatgpt plus, Pro and other plans subscribers, as well as through the OpenAi paid API.

As he did with the chatbots with the large language model (LLM) through the launch of ChatgPT in November 2022, OpenAI essentially created the entire category of reasoning models in September 2024 when he presented O1 for the first time, a class of a class of Models with a new training and architecture regime.

But Openai, according to its recent history, did not do O1 Open Source, contrary to its original name and foundation mission. Instead, he maintained the proprietary model code.

And, in the last two weeks, O1 has been eclipsed by the Chinese Deepseek startup, which launched R1, a rival open source reasoning model, highly efficient, to a large extent open, available for free to take again, return to train and Customize around the world, like this. As use for free on the website and the Deepseek mobile application, a model, according to reports, trained at a fraction of the cost of O1 and other LLM of the best laboratories.

Deepseek-R1 Permissive MIT license termsThe free website/website for consumers and the decision to make the R1 code base be available for free to take and modify has led it to a true explosion of use both in consumer and companies markets, even Operai Microsoft inverter and the anthropic sponsor Amazon running to add variants to their cloud markets. Perplexity, the AI ​​search company, also quickly A variant for users was added.

Deepseek also dethroned the chatgpt iOS application as number 1 in the US Apple App StoreAnd it is remarkable for overcoming Openai by connecting its R1 model to the web search in its application and on the web. This is something that Openai has not yet done for O1, which leads to greater technological anxiety among technological workers and others online that China is up to date or has surpassed the United States in the innovation of AI, or even in Technology in general.

However, many AI researchers, scientists and the main VC such as Marc Andreessen have welcomed the emergence of Deepseek, and their open supply in particular, as a tide that raises all ships in the field of AI, increasing the available intelligence for all while reducing costs.

Availability in chatgpt

O3 is now being implemented worldwide for Chatgpt Free, in addition to equipment and professionals, with access to the company and education next week.

• Free users can try O3-MINI for the first time by selecting the “Reason” button on the chat bar or regenerating an answer.

• Message limits have increased 3 times for more and team users, 50 to 150 messages per day.
• Professional users obtain unlimited access to both O3-MINI and a new variant, even of greater condition, O3-mini-High.

In addition, O3-Mini now admits search integration into chatgpt, providing responses with relevant web links. This feature is still in its early stages, since Openai refines search capabilities in their reasoning models.

Integration and API prices

For developers, O3-mini is available through the chat termination API, attendee API and Lot API. The model admits calls of functions, structured outputs and developer messages, which facilitates integration in real world applications.

One of the most notable advantages of O3-MINI is its cost efficiency: it is 63% cheaper than OPENAI O1-mini and 93% cheaper than the complete O1 model, with a price of $ 1.10/$ 4.40 per million Tokens in/out (with a 50% discount in cache).

However, it still pales compared to the official’s affordability Speeek APIThe offer from R1 to $ 0.14/$ 0.55 per million tokens entering/out. But given Depseek it is based on China and comes with geopolitical awareness and security concerns about user/company data that flow inside and outside the model, it is likely that Operai is still the preferred API for some clients and companies centered on safety in safety in The United States. and Europe.

Developers can also adjust the level of reasoning effort (low, medium, high) depending on their application needs, allowing greater control over latency and precision compensation.

On security, Operai says he used something called “deliberative alignment” with O3-mini. This means that the model was asked to reason about the human author’s safety guidelines that were given to him, they include more of his intention and the damages that are designed to prevent and prepare their own ways to ensure that these damages are avoided. Operai says that it allows the model to be less censored when delicate issues are discussed while preserving security.

Operai says that the model exceeds GPT-4O in the management of security and jailbreak challenges, and that it performed a wide external security test before its launch today.

TO Recent covered report in Cabling (Where my wife works) He showed that Depseek succumbed to each Jailbreak notice and attempts 50 proven by security researchers, which can give OpenAi O3-mini the advantage over Deepseek R1 in cases where security and safety are primordial

That next?

The launch of O3-MINI represents Openai’s broader effort to make advanced AI reasoning more accessible and profitable against a more intense competition than ever of Deepseek’s R1 and others. This includes Google, which recently launched a free version of its own rival Reasoning model Gemini 2 Flash Thinking with an expanded entry context of up to 1 million tokens.

With its approach to the reasoning and stake of Stem, OpenAi aims to expand the scope of problem solving promoted by AI in consumer and developer applications.

But as the company becomes more ambitious than ever, for example, recently announcing a data center infrastructure project of $ 500 billion called Stargate with the support of SoftBank, the question remains if its strategy will be worth penalty or Not well enough to justify the sunken billions. It is for pocket investors deeply like Microsoft and other VC.

As the open source models close more and more the gap with openai in performance and exceed it in cost, according to reports, its superior safety measures, powerful capabilities, easy -to -use API and easy to use interfaces to be enough To keep customers, especially in the company. – Who can prioritize the cost and efficiency over these attributes? As always, we will inform about developments as they develop.

Daily insights on commercial use cases with VB daily

If you want to impress your boss, VB Daily has you covered you. We give the interior account of what companies are doing with generative AI, from regulatory changes to practical implementations, so you can share ideas for the maximum ROI.

Read our Privacy Policy

Thanks for subscribing. Look more VB bulletins here.

A mistake happened.