Openai responds to the Deepseek contest with detailed effects of O3-MINI

Openai now displays more details about the thinking of O3-MINI, its latest model. The change has been announced X Openai account It comes as it is the artificial intelligence laboratory Under the increasing pressure By DeepSeek-R1, an open competing model completely displays the distinctive symbols of thinking.

Models such as O3 and R1 are subject to a long “chain” process in which additional symbols are born to break the problem, the reason for various answers, test them and reach a final solution. Previously, the thinking models in Openai hidden her series of ideas and only issued a high -level overview of the thinking steps. This made it difficult for users and developers to understand the logic of thinking about the model and changing their instructions and demanding that it be directed in the right direction.

In a series of thought, Openai looked at a competitive advantage and hidden it to prevent its competitors from copying to train their models. But with R1 and other open models Show fully logic trackingLack of transparency becomes a defect in Openai.

The new version of O3-MINI displays a more detailed copy of COT. Although we still do not see raw symbols, they provide more clarity about the thinking process.

Why are the requests?

In our Previous experiences On O1 and R1, we found that O1 was a little better in solving data analysis and thinking problems. However, one of the main restrictions was that there was no way to find out why the model committed errors-he committed mistakes when facing the data of the chaotic real world obtained from the web. On the other hand, the R1 R1 thought series enabled us to explore problems and change our claims to improve thinking.

For example, in one of our experiences, both models failed to provide the correct answer. But thanks to the detailed thinking series of R1, we were able to discover that the problem was not with the model itself but with the retrieval stage that collected information from the web. In other experiences, the R1 ideas series managed to provide us with hints when it failed to analyze the information we provided, while O1 only gave us a general overview of how to formulate its response.

We have experienced the new O3-MINI model on a variable of a previous experience that we have played with O1. We have provided the model with a text file that contains different prices from January 2024 to January 2025. The file was noisy and unprocessed, a mixture of normal text and HTML elements. Then we asked the form to calculate the value of the portfolio that invested 140 dollars in the wonderful stocks 7 on the first day of each month of January 2024 to January 2025, and distributed it evenly in all shares (we used the term “MAG 7” in a router to make it more challenging).

The O3-Mini bed was really useful this time. First, the model was nominated about what MAG 7 was for data to maintain relevant shares only (to make the problem difficult, we added a little shares, and made the final accounts to provide the correct answer (the value of the portfolio is about 2200 dollars at the last time it was recorded in the data we provided For model).

It will take more tests to know the boundaries of the new thinking chain, because Openai still hides many details. But in the check checks, the new coordination appears to be more useful.

What does it mean to Openai

When Deepseek-R1 was released, it had three clear advantages on the thinking forms in Openai: it was open, cheap and transparent.

Since then, Openai has managed to shorten the gap. While O1 costs $ 60 per million output symbols, O3-MINI costs only $ 4.40, while O beat over many thinking criteria. R1 costs about $ 7 and $ 8 per million symbols for American service providers. (Deepseek R1 offers $ 2.19 per million symbols on their own servers, but many organizations will not be able to use them because they host them in China.)

With the new change in Cot output, Openai enables somewhat action about the problem of transparency.

It remains to see what Openai will do about open sources. Since its release, the R1 has already been adapted, similar and hosted by many different laboratories and companies that are likely to make the preferred thinking model for institutions. Openai Sam Altman CEO recently admitted that he was “was” recentlyOn the wrong side of history“In an open source discussion, we will have to see how this perception will appear in future Openai versions.

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.