Google launches Gemini Deep Think — the most powerful AI model for logical reasoning

Google DeepMind has introduced Gemini 2.5 Deep Think — its most advanced artificial intelligence model for logical tasks. It can simultaneously consider and analyze multiple answer options, then select the best one.

Gemini 2.5 Deep Think will be available to Ultra plan subscribers at $250 per month through the Gemini app.

Deep Think may be particularly useful for solving tasks related to creativity, strategic planning, and step-by-step process improvement.

The model demonstrates outstanding results on the Humanity’s Last Exam test, which evaluates AI’s ability to answer thousands of complex questions in mathematics, humanities, and natural sciences. Gemini 2.5 Deep Think scored 34.8%, while Grok 4 from xAI received 25.4%, and o3 from OpenAI scored 20.3%.

On the LiveCodeBench 6 test, which includes complex programming tasks, Google’s model also outperformed competitors, achieving a score of 87.6%. Grok 4 scored 79%, and o3 scored 72%.

The model can automatically use tools such as Google Search and code execution, and generates longer and more detailed responses compared to traditional AI models.

In experiments, Google Deep Think showed better results in web development and visual content creation tasks. The company expects the model to help accelerate scientific discoveries.

In the coming weeks, Google plans to provide access to Gemini 2.5 Deep Think via API to a limited number of developers and corporate clients to explore application scenarios in business and science.

Source: TechCrunch