Google publicly publishes the Gemini winners of the Olympiac medal.
Do you want smarter information in your reception box? Sign up for our weekly newsletters to obtain only what matters for business managers, data and security managers. Subscribe now
Google officially launched Gemini 2.5 Deep Think, a new variation of its AI model designed for deeper reasoning and a resolution of complex problems, which made the headlines last month for having won a gold medal at the International Mathematical Olympiad (IMO) – The first time that an AI model has achieved the feat.
However, unfortunately it is not The identical model of gold medal. It is in fact a less powerful “bronze” version according to the blog post of Google and Logan Kilpatrick, Lead-Lead for Google AI Studio.
As Kilpatrick has published on the social network X: “This is a variation in our OMO Gold model which is faster and more optimized for daily use. We also give the full model of Gold IMO to a set of mathematicians to test the value of complete capacities. ”
Now available via the Gemini mobile applicationThis bronze model is accessible to subscribers of the Google’s most expensive IA individual plan, AI Ultra, which costs $ 249.99 per month with a 3 -month departure promotion at a reduced rate of $ 124.99 / month for new subscribers.
The IA Impact series returns to San Francisco – August 5
The next AI phase is here – are you ready? Join the Block, GSK and SAP leaders for an exclusive overview of how autonomous agents reshape business workflows – from real -time decision -making to end -to -end automation.
Secure your place now – space is limited:
Google also declared in its publication blog article that it would provide a deep reflection with and without integration of use of tools to “trust testers” via the Gemini application programming interface (API) “in the coming weeks”.
Why “deep thought” is so powerful
Gemini 2.5 Deep Think is based on the family of gemini of large language models (LLM), adding new capacities aimed at reasoning through sophisticated problems.
He Using techniques of “parallel reflection” to explore several ideas simultaneously and includes learning to strengthen to strengthen its resolution capacity step by step over time.
The model is Designed for use cases which benefit from prolonged deliberation, such as mathematical conjecture tests, scientific research, algorithm design, and creative iteration tasks such as code refinement and design.
The first testers, including mathematicians such as Michel Van Garrel, used it to probe unresolved problems and generate potential evidence.
The AI user and expert Ethan Mollick, professor of the Wharton School of Business at the University of Pennsylvania, also published on X that he was able to take an invitation that he often uses to test the capacities of the new models – “create something that I can stick in a P5J that will surprise me with his intelligence” and his adventure “and at the origin of the control group in the distant future” and transformed it into a 3D graphic, which is the first time that any model does it.
Performance benchmarks and use cases
Google highlights several areas of key application for a deep reflection:
- Mathematics and Sciences: The model can simulate reasoning for complex evidence, explore conjectures and interpret dense scientific literature
- Algorithm coding and design: It works well on tasks involving performance compromises, temporal complexity and logic in several stages
- Creative development: In design scenarios such as Voxel art versions or user interface, Deep Think demonstrates an iterative improvement and an improvement in details
The model too leads to performance in reference assessments such as Livecodebench V6 (for coding capacity) And the last examination of humanity (covering mathematics, sciences and reasoning).
He OwSettectured Gemini 2.5 PRO and competing models like the OPENAI GPT-4 and the XAI GROK 4 By two -digit margins on certain categories (reasoning and knowledge, generation of code and mathematics of the OMI 2025).
Gemini 2.5 Deep Think vs Gemini 2.5 Pro
While Deep Think and Gemini 2.5 Pro are part of the family of Gemini 2.5 models, Google positions Deep Think as a More capable and analytically qualified variantIn particular with regard to complex reasoning and problem solving in several stages.
This improvement stems from the use of parallel thought And Reinforcement learning techniqueswhich allow the model to simulate a deeper cognitive deliberation.
In its official communication, Google describes Deep Think as the best to Management of nuanced prompts, exploring several hypotheses and produces more refined outings. This is supported by comparisons side by side in the Voxel art generation, where Deep Think adds more texture, structural fidelity and diversity of composition than 2.5 pro.
Improvements are not only visual or anecdotal. Google reports that Deep Think GEMINI 2.5 PRO SURPACE ON SUPPORTS TECHNICAL WARDES linked to reasoning, code generation and expertise between the cross -country. However, these gains come with compromise in responsiveness and rapid acceptance.
Here is a ventilation:
| Capacity / attribute | Gemini 2.5 Pro | Gemini 2.5 Deep Think |
|---|---|---|
| Speed of inference | Faster and low latency | “Slow and prolonged thinking time” |
| Reasoning complexity | Moderate | High – use a parallel thought |
| Prompt depth and creativity | GOOD | More detailed and nuanced |
| Reference performance | Strong | State of art |
| Content security and objectivity of the tone | Improved on older models | Still improved |
| Refusal rate (benign prompts) | Lower | Higher |
| Output length | Standard | Supports longer responses |
| Voxel Art / Design Fidelity | Basic scene structure | Improved details and wealth |
Google notes that The higher refusal rate of Deep Think is an active field of survey. This can limit its flexibility in the management of ambiguous or informal requests compared to 2.5 pro. On the other hand, 2.5 Pro remains better suited to users who prioritize Speed and responsivenessEspecially for lighter and general use tasks.
This differentiation allows users to choose according to their priorities: 2.5 Pro for speed and fluidityOr Deep thought for rigor and reflection.
Not the winning model of the gold medal, just a bronze
In July, Google Deepmind made the headlines when a more advanced version of the Gemini Deep Think model obtained the official status of the gold medal at the OMI 2025 – the most prestigious mathematics competition in the world for high school students.
The system Solved five of the six difficult problems and became the first AI to receive a score at the OMI level.
Demis Hassabis, CEO of Google Deepmind, announced the realization on X, indicating that the model had resolved end -to -end problems in natural language – without the need for translation in a syntax of formal programming.
The IMO card confirmed that the model had marked 35 out of 42 possible points, well above the gold threshold. Gemini 2.5 Deep Think’s Solutions were Described by the president of the Gregor Dolinar competition as clear, precise and in many cases, Easier to follow than those of human competitors.
However, the Gemini 2.5 Deep Think published to users is not the same competition model, rather, a more efficient but apparently faster version.
How to access Deep Think now
Gemini 2.5 Deep Think is Available exclusively on the Google Gemini mobile application for iOS and Android for the moment users in terms of Google AI UltraPart of the Google One subscription range, with prices as follows.
- Promotional offer: $ 124.99 for 3 months, then it starts at…
- Standard rate: $ 249.99 / month
- Characteristics included: 30 TB of storage, access to the Gemini application with Deep Think and Veo 3, as well as tools such as flow, whip and 12,500 monthly AI credits
Subscribers can deeply activate reflection in the Gemini application by selecting the 2.5 Pro model and tilting the “Think Deep” option.
It supports a fixed number of prompts per day and is integrated into capacities such as the execution of the code and Google research. The model also generates longer and more detailed outputs compared to standard versions.
The lower level Google AI plan, at the price of $ 19.99 / month (with a free trial), does not include access to Deep Think, any more than the free gemini ai service.
Why this counts for corporate technical decision -makers
Gemini 2.5 Deep Think represents the practical application of an important step in research.
He Allows companies and organizations to draw on a model of medal medal in mathematics and to make it join their staff, Although only via an individual user account.
For researchers who receive the complete model of Grade IO, it offers an overview of the future of collaborative AI in mathematics. For Ultra subscribers, Deep Think provides a powerful step towards more competent and aware assistance of AI, which now takes place in the palm of their hand.
About The Author
