Want more intelligent visions of your inbox? Subscribe to our weekly newsletters to get what is concerned only for institutions AI, data and security leaders. Subscribe now
Google Officially launched Gemini 2.5 Deep Think A new difference in the artificial intelligence model designed for deeper thinking and complex problem solving, which is Newspaper headlines last month to win a gold medal at the International Mathematics Olympics (IMO) The first time that the artificial intelligence model has achieved this feat.
but, Unfortunately no The matching model for the gold medal. In fact, the “Bron’s” release is less powerful according to the publication of the Google and Logan Kilpatric Blog, on which the Google Ai Studio product is based.
like Kilpatrick has been published on the x social network: “This is a variation in our IMO gold model, which is faster and more improved for daily use. We also offer the IMO full model for a group of mathematicians to test the full capabilities value.”
Available now through the Gemini Mobile appThis bronze model is available to subscribers Amnesty International Plan is more expensive than the AI in Google, AI UltraWhich costs $ 249.99 per month with a 3 -month start promotion at a reduced rate of $ 124.99 per month for new subscribers.
AI Impact series returns to San Francisco – August 5
The next stage of artificial intelligence here – are you ready? Join the leaders from Block, GSK and SAP to take an exclusive look on how to restart independent agents from the Foundation’s workflow tasks-from decisions in an actual time to comprehensive automation.
Securing your place now – the space is limited: https://bit.ly/3GUPLF
Google also said in the publication blog publication that it would bring deep thinking with and without the integration of tools to “trusted laboratories” through the API applications interface in the coming weeks.
Why is “deep thinking” very strong
Gemini 2.5 Deep Think depends on the Gemini family from the large LLMS models, adding new capabilities aimed at thinking through advanced problems.
He – she It employs “parallel thinking” techniques to explore multiple ideas simultaneously and include learning to enhance to enhance their ability to solve problems step by step over time.
The model is Designed for use cases that benefit from expanded deliberations, such as a guessing testing, scientific research, and algorithm design, Creative repetition tasks such as software instructions and refinement of design.
The first laboratories, including mathematicians such as Michelle van Jarrel, used to investigate problems that have not been solved and generated possible evidence.
User and power expert from artificial intelligence, Ethan Malik, Professor of Warton Business at the University of Pennsylvania, It was also published on X He was able to take a claim that he often used to test the possibilities of new models – “Create something I can paste in P5Js, which will amaze me with his intelligence in creating something that calls for the stars control panel in the distant future” – and I turned it into a three -dimensional drawing, the first time that any model has done so.
Performance standards and cases of use
Google highlights many major application areas for deep thinking:
- Mathematics and science: The model can simulate thinking about complex proofs, explore guesses, and the interpretation of heavy scientific literature
- Coding and algorithm design: It works well on the tasks that involve performance bodies, the complexity of time, and a multi -step logic
- Creative development: In design scenarios such as Foxel Art or User interface, deep thinking shows a stronger recurrence and enhancement of details
Model also Performing standard assessments such as LiveCodebeench v6 (For coding ability) And the last exam for humanity (Mathematics coverage, science and thinking).
He – she GIMINI 2.5 Pro and competing models such as Openai’s GPT-4 and Xai’s GROK 4 surpassed Through the dual numbers margins in some categories (thinking and knowledge, generating the code, and IMO 2025 mathematics).
Gemini 2.5 Deep Think versus Gemini 2.5 Pro
While the Deep Think and Gemini 2.5 Pro are part of the Gemini 2.5 Form, Google put Google Think A more capable variable and analyticalEspecially when it comes to complex thinking and solving multiple problems.
This improvement stems from use Parallel thinking and Reinforce learning techniquesWhich enables the model to simulate the deepest cognitive deliberations.
In its official contact, Google describes deep thinking as the best in Dealing with accurate claims, exploring multiple hypotheses, and producing more accurate outputs. This is supported by comparisons side by side in the Foxel Art, where deep thinking adds more texture, structural loyalty and synthetic diversity of 2.5 Pro.
Improvements are not only visual or anecdotal. Google mentioned that Deep Think GIMINI 2.5 Pro outperforms multiple technical standards Concerning, generating code, and domain experience across the field. However, these gains come with symptoms of response and immediate acceptance.
This is a collapse:
| The ability / feature | Gemini 2.5 Pro | Gemini 2.5 deep thought |
|---|---|---|
| Conclusion speed | Faster, low transition time | Slow down, the extended “thinking time” |
| The complexity of thinking | moderate | High – uses parallel thinking |
| Immediate depth and creativity | good | More detailed and accurate |
| Standard performance | strong | state of the art |
| Safety of content and objectivity of tone | It improved the old models | Improved more improvement |
| Rejection rate (benign claims) | minimum | higher |
| The length of the output | standard | Supports longer responses |
| Faksel Art / Design is sincerity | The basic scene structure | Enhanced details and wealth |
Google notice that High rejection rate in deep thinking It is the field of active investigation. This may limit its flexibility in dealing with mysterious or informal queries compared to 2.5 Pro. In contrast, 2.5 Pro is still more suitable for users who give priority Speed and responseEspecially for lighter tasks for general purposes.
This distinction allows users to choose based on their priorities: 2.5 Pro for speed and liquidityOr Deep thought of rigor and thinking.
It is not the model of winning the gold medal, just a bronze
In July, Google DeepMind topped the headlines when it was a more advanced version than The Gemini Deep Think model has made a golden position on the IMO 2025 – The most famous mathematics competition in the world for high school students.
order Five of six difficult problems and became the first Amnesty International to receive a gold level from IMO.
Demis Hassabis, CEO of Google DeepMind, has announced the achievement of X, saying that the model had solved problems from one side-without the need for translation to build a total programming sentence.
The IMO board confirmed that the model scored 35 points from 42 potential points, much higher than the gold threshold. Gemini 2.5 Deep Think solutions Description of the head of competition, Gregor Dolinear Obvious and accurate and in many cases, The easiest to follow up on human competitors.
However, Gueini 2.5 Deep Think, which was released to users is not the same as the competition model, instead, it appears to be less performance but apparently faster.
How to get to Deep Think now
Gemini 2.5 deep thinking Available exclusively on the Google Gemini Mobile app for iOS and Android at this time for users on the Google Ai Ultra planPart of the Google One subscription collection, with prices as follows.
- Promotional offer: 124.99 dollars per month for 3 months, then starts to …
- Standard rate: 249.99 dollars/month
- Features included: 30 TB of storage, access to Gemini with Deep Think and Veo 3, as well as tools such as Flow, Westisk and 12,500 monthly credits of artificial intelligence
Subscribers can activate deep thinking in the Gemini app by selecting the 2.5 Pro model and switching the “deep thinking” option.
It supports a fixed number of claims daily and is combined with capabilities such as implementing software and research in Google. The model also generates longer and more detailed outputs compared to standard versions.
The low -level Google Ai Pro plan, at $ 19.99 per month (with a free experience), does not include deep Think, nor free GEMINI AI.
Why do the technical decision makers of the institutions concern?
Gemini 2.5 Deep Think represents the practical app for a major search teacher.
He – she Institutions and institutions are allowed to take advantage of the Media Olympiad model and make them join their employees, And if it is only through the individual user account now.
For researchers who receive a full IMO model, it offers a glimpse of the future of cooperative artificial intelligence in mathematics. For superior subscribers, Deep Think provides a strong step towards artificial intelligence help, which you are now running at their hands.
Source link