Need smarter insights in your inbox? Sign up for our weekly newsletter to get only the things that matter to enterprise AI, data and security leaders. Subscribe now
Google has officially launched the Gemini 2.5 Deep Think, a new variation of an AI model designed for deeper reasoning and complex problem solving. This led to the headline being that he won a gold medal at the International Mathematics Olympiad (IMO) last month.
but, Unfortunately, this is a shame do not have A model who won the same gold medal. In fact, it’s a less powerful “bronze” version, according to Google’s blog post and Google AI Studio’s product lead Logan Kilpatrick.
As Kilpatrick posted on Social Network X, “This is a variation of the IMO Gold model that is faster and optimized for daily use. We also provide an IMO Gold Full model to a set of mathematicians to test the value of the full functionality.”
Now available from the Gemini Mobile appThis bronze model is accessible to subscribers of Google’s most expensive individual AI plan, AI Ultra. This comes with a three-month startup promotion at a rate of $124.99 per month for new subscribers, and costs $249.99 per month.
The AI Impact Series returns to San Francisco – August 5th
The next phase of AI is here – Are you ready? Join Block, GSK and SAP leaders to see exclusively how autonomous agents are reshaping their enterprise workflows, from real-time decision-making to end-to-end automation.
Secure your spot now – Space is limited: https://bit.ly/3guplf
Google also said in a blog post of the release that the Gemini Application Programming Interface (API) will bring “trust testers” to think deeply without the integration of tool use through “in the coming weeks.”
Why “deep thinking” is so powerful
Gemini 2.5 Deep Think is built on the Gemini family of large-scale language models (LLMS) and adds new features aimed at inference through sophisticated problems.
that It employs “parallel thinking” techniques to explore multiple ideas simultaneously, and includes reinforcement learning to enhance step-by-step problem-solving capabilities over time.
The model is It is designed for use cases that benefit from extended deliberations, such as mathematical inference testing, scientific research, and algorithm design. Creative iteration tasks such as code and design improvements.
Early testers, including mathematicians such as Michelle Van Garrell, used them to investigate unresolved problems and generate potential proofs.
Ethan Ethan Mollick, a power user and expert in AI, is also a professor at Wharton School of Business at the University of Pennsylvania, and posted on X. I posted that this allows you to take prompts that you often use to test the capabilities of new models. I changed it to 3D graphics..
Performance Benchmarks and Use Cases
Google highlights some important application areas for deep thought:
- Mathematics and ScienceModels can simulate complex evidence inference, explore speculation, and interpret dense scientific literature
- Coding and Algorithm Design: Works well with tasks that include performance trade-offs, time complexity, and multi-step logic
- Creative Development: In design scenarios such as Voxel Art and user interface builds, Deep Think shows more powerful improvements and enhancements to detail
The model too Leads performance in benchmark ratings such as LiveCodebench V6 (Because of coding ability) And the final exam of mankind (covers mathematics, science, and reasoning).
that Models that compete with Gemini 2.5 Pro such as OutscoredGpt-4 and Xai’s Grok 4 By two digit margins in several categories (inference and knowledge, code generation, and IMO 2025 mathematics).
Gemini 2.5 Deep Think vs. Gemini 2.5 Pro
Both Deep Think and Gemini 2.5 Pro are part of the Gemini 2.5 model family, but Google has Deeps as Deeps Thinks More competent and analytically skilled variantespecially when it comes to complex inference and multi-step problem solving.
This improvement is due to the use of Parallel thinking and Reinforcement learning technologyallowing the model to simulate deeper cognitive deliberations.
In official communications, Google explains that Deep Thinking is better Handling subtle prompts, exploring multiple hypotheses, creating more sophisticated outputs. This is compared side-by-side in Voxel Art Generation. This adds more texture, structural fidelity and compositional diversity than Deep Think’s 2.5 Pro.
The improvements aren’t just visual or anecdote. Google reports Deep Think Overcome Gemini 2.5 Pro with multiple technical benchmarks It relates to inference, code generation, and cross-domain expertise. However, these benefits come with responsive trade-offs and quick acceptance.
This is the fault:
| Functions/Attributes | Gemini 2.5 Pro | Gemini 2.5 Deep Think |
|---|---|---|
| Inference speed | Faster, lower latency | Later, extended “thinking time” |
| The complexity of reasoning | Moderate | High – Use parallel thinking |
| Quick depth and creativity | good | More detailed and subtle |
| Benchmark Performance | strong | The cutting edge |
| Content safety and tone objectivity | Improved on older models | It has been further improved |
| Rejection rate (benign prompt) | Lower | Higher |
| Output length | standard | Supports longer responses |
| Voxel Art/Design faithfulness | Basic scene structure | Enhance details and richness |
Google points it out Higher rejection rates for Deep Think This is an area of active research. This may limit the flexibility in handling ambiguous or informal queries compared to 2.5 Pro. In contrast, 2.5 Pro remains more suited to prioritizing users Speed and Responsivenessespecially for lighter, general purpose tasks.
This differentiation allows users to choose based on their priorities. 2.5 Pro for velocity and fluidityor Think deeply for rigor and reflection.
It’s not a gold medal model, it’s just a bronze model.
In July, Google Deepmind made the headline when a more advanced version of the Gemini Deep Think model achieved official gold medal status at the 2025 IMO.
system He solved five of the six challenging issues and became the first AI to receive gold-level scoring from IMO.
Demis Hassabis, CEO of Google Deepmind, announced his achievements in X, saying the model solved end-to-end problems in natural language.
The IMO board confirmed that the model scored 35 out of 42 points, which are well above the gold threshold. That’s how Gemini 2.5 Deep Think’s solution was Competitor President Gregor Drinner explained Clear, accurate, and often It’s easier to follow than your competitors.
However, the user-released Gemini 2.5 Deep Think is not a very similar competitive model and has a lower performance, but is clearly a faster version.
How to access deep thinking now
Gemini 2.5 is a deep way of thinking Currently, Google AI Ultra plans are available only to users on the Google Gemini mobile app for iOS and Androidpart of the Google One subscription lineup, prices are as follows:
- Promotional offers: $124.99/month for 3 months, then kick up…
- Standard Rate: $249.99 per month
- Features Included: 30 TB of Storage, Access to Gemini App with Deep Think and Veo 3, and Tools such as Flow, Whisk, 12,500 Monthly AI Credits
Subscribers can activate Deep Think in the Gemini app by selecting a 2.5 Pro model and toggling the “Deep Think” option.
It supports a fixed number of prompts per day and is integrated with features such as code execution and Google search. This model also produces longer and more detailed output compared to the standard version.
The low-level Google AI Pro plan, which costs $19.99 a month (free trial), does not include access to Deep Think and does not include the free Gemini AI service.
Why it’s important for enterprise technical decision makers
Gemini 2.5 Deep Think represents a practical application of a major research milestone.
that Companies and organizations can use models that have won Olympia medals in mathematics to join their staff. However, only through individual user accounts.
Researchers who have received a complete IMO grade model will have a glimpse into the future of collaborative AI in mathematics. For Ultra subscribers, Deep Think offers a powerful step towards more capable, contextual AI support, and is now running in the palm of your hand.
Source link
