Google officially launched Gemini 2.5 deep pondering, A brand new variety of the AI model designed for deeper reasoning and complex problem solving, which last month was on the first pages of newspapers for winning the gold medal at the International Mathematical Olympics (IMO)-after the first time the AI model achieved this feat.
However, Unfortunately, this is an equivalent model of the golden medal. In fact, this is a less powerful “brown” version in accordance with the post on the Google blog and Logan Kilpatrick, a product wire for Google AI Studio.
How Kilpatrick has published in the social network x: “This is a variety of our IMO Gold model, which is faster and more optimized for everyday use. We also provide the full IMO Gold model of the mathematician set to test the value of full possibilities.”
Now available via the Gemini mobile applicationThis brown model is available for subscribers The costliest individual plan AI Google, AI Ultrawhich costs USD 249.99 per 30 days with a 3-month initial promotion at a reduced rate of USD 124.99/month for latest subscribers.
The AI Impact series returns to San Francisco – August 5
The next AI phase is here – are you ready? Join the leaders from Block, GSK and SAP to see the exclusive look at how autonomous agents transform the flows of the work of the company-decision-making in real time for comprehensive automation.
Secure your home now – the space is limited: https://bit.ly/3guplf
Google also said in his blog post that he would bring deep pondering with integration and without integration of the wear of tools for “trusted testers” via the programming interface of the Gemini application (API) “In the coming weeks”.
Why “deep thinking” is so powerful
Gemini 2.5 Deep Think is based on the family of large large language models (LLM), adding latest possibilities aimed at reasoning through sophisticated problems.
This He uses “parallel thinking” techniques to concurrently discover many ideas and includes learning to strengthen to strengthen the ability to resolve problems step by step in time.
The model is Designed for use cases that use an prolonged meeting, comparable to testing mathematical mortgments, scientific research, algorithm design, and the tasks of creative iteration, comparable to code and improvement of design.
Early testers, including mathematicians comparable to Michel Van Garrel, used it to look at unsolved problems and generate potential evidence.
AI Power User and expert Ethan Mollick, Professor Wharton School of Business at the University of Pennsylvania, also published on X That he was in a position to take a quick one, which he often uses to check the possibilities of latest models – “create something that I can paste in P5J, which will surprise me with his clever in creating something that causes a ship control panel in the distant future” – and He turned it into 3D graphics, which for the first time did every model.
Performance indicators and cases of use
Google emphasizes several key areas of the application for deep pondering:
- Mathematics and science: The model can simulate the reasoning of the provided evidence, examine the hypothesis and interpret dense scientific literature
- Coding and algorithm design: Works well on tasks related to performance compromises, complexity of time and multi -stage logic
- Creative development: In design scenarios comparable to Voxel art compilations or user interface, Deep Think shows a stronger improvement and iterative improvement and improvement of details
Model also It conducts performance in comparative assessments comparable to LivecodeBench V6 (for coding capability) and the last exam of humanity (Covering mathematics, science and reasoning).
This Outsced Gemini 2.5 Pro and competitive models comparable to OpenAI’s GPT-4 and XAI’s GROK 4 by double digital margins in some categories (reasoning and knowledge, code generation and mathematics IMO 2025).
Gemini 2.5 Deep Think vs. Gemini 2.5 Pro
While each deep pondering and Gemini 2.5 PRO are part of the family of the Gemini 2.5 model, Google is deep pondering as more talented and qualified analytically variantEspecially when it involves complex reasoning and multi -stage problem solving.
This improvement results from use Parallel pondering AND Strengthening learning techniqueswhich enable the model simulation of deeper cognitive sessions.
In its official communication, Google describes deep pondering as higher Service. This is supported by comparisons next to each other in the generation of Wokseels, in which Deep Think adds more texture, structural loyalty and compositional diversity than 2.5 PRO.
Improvements are not only visual or anecdotal. Google reports that deep pondering exceeds Gemini 2.5 Pro on many technical comparative tests related to the justification, generation of code and specialist knowledge between domains. However, these profits have compromises in response and quick acceptance.
Here is a failure:
| Ability / attribute | Gemini 2.5 Pro | Gemini 2.5 Deep Think |
|---|---|---|
| Application speed | Faster, low delay | Slower, longer “thinking time” |
| Complexity of reasoning | Moderate | High – uses parallel pondering |
| Fast depth and creativity | Good | More detailed and refined |
| Comparative performance | Strong | Most modern |
| CONTENT SAFETY AND OBJECTIVE TON | Improved in relation to older models | Further improved |
| Refusal indicator (mild hints) | Lower | Higher |
| Output length | Standard | Supports longer answers |
| Voxel Art / Design Fidelity | The basic structure of the stage | Increased details and wealth |
Google notes that Higher indicator of deep pondering refusal Is an area of energetic investigation. This may reduce its flexibility in service of ambiguous or informal queries in comparison with 2.5 PRO. However, 2.5 Pro stays higher for users who set priorities speed and responseEspecially in the case of increasing, general tasks.
This diversity allows users to decide on on the basis of their priorities: 2.5 Pro for speed and liquidityOr Deep think about rigors and reflection.
Not the golden victorious model, but bronze
In July, Google Deepmind was on the headlines, when the more advanced version of the Gemini Deep Think model achieved the official gold status in IMO in 2025-the most prestigious mathematics competition in the world for highschool students.
System He solved five of the six difficult problems and became the first artificial intelligence that received a gold level of gold from IMO.
Demis Hassabis, general director of Google Deepmind, announced the achievement on X, stating that the model has solved problems in natural language-including the have to translate into formal programming syntax.
The IMO board confirmed that the model obtained 35 of the possible 42 points, much above gold. Think twins 2.5 Deep Think solutions described by the president of the Gregor Dolinar competition as clear, precise and in many cases, easier to follow than people from human competitors.
However, Gemini 2.5 Deep pondering released to users is not the same competition model, but relatively a lower, but seemingly faster version.
How to access a deep think now
Gemini 2.5 Deep Think IS Available only in the Google Gemini mobile application on iOS and Android at this time to users in the Ultra AI Google planPart of the Google One subscription composition, with a quote as follows.
- Promotional offer: USD 124.99/month for 3 months, and then ends …
- Standard rate: USD 249.99/month
- Functions included: 30 TB of memory, access to the Gemini application with deep pondering and VEO 3, in addition to tools comparable to Flow, Whiskey and 12,500 monthly AI loans
Subscribers can activate deep pondering in the Gemini application by choosing the 2.5 Pro model and switching the “deep thinking” option.
It supports a fixed number of prompts a day and is integrated with capabilities comparable to performing code and Google search. The model also generates longer and more detailed results compared to plain versions.
Google AI Pro plan of a lower level, at a price of $ 19.99/month (with a free process), does not include access to deep pondering, or the free AI Gemini service.
Why is this vital for technical decision makers of enterprises
Bligini 2.5 Deep Thin represents the practical application of the predominant milestone of research.
This It allows enterprises and organizations to make use of the Mathematics Olympics model Although now only through an individual user account.
For scientists receiving the full IMO class model, it offers insight into the future of joint artificial intelligence in mathematics. For the ultra subscribers, Deep Think is a powerful step towards a more talented and conscious context of artificial intelligence help, which was now working in the hand.
