Want a smart insight into your inbox? Sign up for our weekly newsletters to get the only thing that is important to enterprise AI, data, and security leaders. Subscribe now
Google The Gemini 2.5 deep think is formally launched, A new variation of its AI model engineered to solve a deep reasoning and complex problem, which made headlines last month to win the gold medal at the International Mathematics Olympiad (IMO)-the first time the AI model achieved the feat.
However, This is unfortunately Not The same gold medalist model. In fact, Google’s Blog Post and Google AI Studio Product Lead, Logan Kill Patrick, is a less powerful “bronze” version.
As if Call Patrick posted on social network X: “This is a variation of our IMO Gold Model that is faster and better for daily use. We are also giving a full model to IMO Gold to a set of mathematicians to test the value of full potential.”
Now available via the Gemini Mobile appThis brass model is accessible to users Google’s most expensive individual AI plan, AI UltraWhich costs $ 249.99 per month, with a 3 -month start of 3 months for new consumers at a lower rate of 4 124.99/month.
AI Impact Series returning to San Francisco – August 5
The next step of the AI is here – are you ready? Block, GSK, and SAP leaders include for a special look on how autonomous agents are changing enterprise workflows-from real time decision-making to end to automation.
Now secure your place – space is limited:
Google has also said in its release blog post that it will bring deep thinking deeply with the integration of tools for “reliable testers” “in the coming weeks” through the Gemini application programming interface (API).
Why ‘deep thinking’ is so powerful
The Gemini 2.5 deep think builds on the Gemini family of large language models (LLM), which includes new abilities that aims to argue through sophisticated problems.
These At the same time, “parallel thinking” hires techniques to find numerous ideas, and it includes learning to strengthen its ability to solve his step -by -step problem over time.
Is the model Designed for use matters that benefit from expansional consideration, such as mathematical speculation testing, scientific research, algorithm design, And creative repetition works such as codes and designs.
Early testers, including mathematics such as Michelle Van Garl, have used it to investigate and create potential evidence of non -soluble problems.
AI Power User and Expert Ethan Mulk, Professor of the Wartton School of Business at the University of Pennsylvania, Also posted on x That he was able to take a hint that he often uses to test the capabilities of new models – “I create something that I can paste into P5Js that will surprise me to create something that requests a control panel of Starship in the future.” Converted it to 3D graphic, which models have done for the first time.
Performance standards and matters of use
Google deeply application highlights several key application areas:
- Maths and science: Model can imitate reasoning for complex evidence, detect speculation, and translate dense scientific literature
- Coding and algorithm design: This performance performs well on trade office, time complexity, and multi -faceted logic works
- Creative Development: In design scenarios, such as Vocal Art or User Interface, Deep Think demonstrates strong regeneration improvement and enhancing detail
Model Benchmark diagnosis guides performance such as Liu Kodch v6 (For coding qualification) And last test of humanity (Covering mathematics, science and reasoning).
These Openai’s GPT -4 and ZK Grick 4 like Outsord Gemini 2.5 Pro and Competitive Model Through a double -digit margin on some categories (reasoning and knowledge, code generation, and IMO 2025 mathematics).
Gemini 2.5 Deep Think vs. Gemini 2.5 Pro
Although the deep think and the Gemini 2.5 Pro are part of the Gemini 2.5 Model Family, Google’s positions think deeply More worthy and analytically skilled varietyEspecially when it comes to solving complex arguments and multi -faceted problems.
This is from the use of improvement Parallel And The reinforcement techniquesWhich enables the model to take deep knowledge.
In its official communication, Google has deepened the deep thinking better Dealing with newborn indicators, detecting multiple assumptions, and developing more better output. It is supported by comparison in vocal art generation, where Deep Think adds more than 2.5 Pro structure, structural loyalty, and structural diversity.
Improvement is not just visual or story. Google reports that deep thinking Many technical bench marks improve Gemini 2.5 Pro Related to reasoning, code generation, and cross domain skills. However, these benefits are with trade in response and immediate acceptance.
Here is an error:
| Capacity / attribute | Gemini 2.5 Pro | Gemini 2.5 deep thinking |
|---|---|---|
| Diagnosis speed | Faster, less delayed | Slowly, extended “Time to think” |
| The complexity of reasoning | Moderate | High – uses parallel thinking |
| Quick depth and creativity | Nice | Is more detailed and necessary |
| Benchmark performance | Concrete | Sophisticated |
| The safety and tone object of the contents | Improved old models | Got better |
| Refuse rate (benign indicators) | Lower | Instrument |
| Output length | Standard | Supports longer answers |
| Voxel Art / Design Loyalty | The basic scene structure | Better detail and Frawani |
Google notes High denial rate of deep think There is an area of active investigation. It can restrict its flexibility in handling vague or informal questions compared to 2.5 Pro. On the contrary, 2.5 Pro is better suitable for those users who prefer Speed and reactionEspecially for mild, general purpose tasks.
This distinction allows users to choose based on their preferences: 2.5 Pro for speed and fluencyOr Think deep for hardness and reflection.
Gold Medal Winning Not a Winning model, just a bronze
In July, Google Deep Mind made headlines when a sophisticated version of the Gemini Deep Think Model officially gained the status of gold medal in the 2025 IMO-which is the world’s most famous mathematics competition for high school students.
System Five of the six solved difficult problems and became the first AI to score gold from the IMO.
Google Deep Mind’s CEO, Dimmes Coshabie, announced the success on X, saying that the model has solved the problems from the end to the end-without any formal programming syntax.
The IMO Board confirmed that the model scored 35 out of possible 42 points, above the gold doorstep. Gemini had 2.5 deep think solutions The contest was described by President Gregor Dulner As clear, precise, and in many cases, It is easy to follow than human rivals.
However, Gemini 2.5 deep thinking is not the same competitive model released to users, but is a low -performing but seemingly faster version.
How to access deep thinking now
Gemini is 2.5 deep thinking Google A. Ultra Plan is currently available to users on the Google Gemini mobile app for iOS and Android.Part of the Google One Subscript Lineup, the following, with prices.
- Promotional Offer: 4 124.99/month for 3 months, then it …
- Standard rate: 9 249.99/Month
- Added Features: Tools such as 30 TB storage, deep think and VEO 3 with VEO 3, as well as flu, whispers, and 12,500 monthly AI credit
Users can activate a deep think in the Gemini app by selecting the 2.5 Pro model and togeling the “Deep Think” option.
It supports a fixed indicator daily and is integrated with the code implementation and capabilities like Google Search. The model also produces longer and more detailed output than the standard version.
The lower level Google AI Pro Plan, which costs. 19.99/month (with a free trial), it does not include deep thinking, nor is the free Gemini AI service.
Why this Enterprise is important to technical decision makers
The Gemini 2.5 deep think represents the practical application of a major research milestone.
These Allows businesses and organizations to tap in the math Olympiad medal winning model and allow it to be included in their staff, Although now only through an individual user account.
The Full IMO grade model researchers, it, reflects the future of AI with mutual cooperation in mathematics. For ultra -consumers, Deep Think provides a powerful step towards AI aid, more capable and familiar with context, which is now running in the palm of his hand.
