Google Launches Deep Think Mode for Gemini 3: Setting New AI Reasoning Benchmarks

Google has officially released the Deep Think mode for its Gemini 3 model, introducing advanced reasoning capabilities that allow the AI to explore multiple hypotheses simultaneously to arrive at the most accurate answers.

The enhanced reasoning capabilities enable Gemini 3 to significantly outperform competitors in industry-standard benchmarks. In the Humanity's Last Exam benchmark, Gemini 3 with Deep Think confidently surpasses both GPT-5 Pro and Claude Sonnet 4.5. The model demonstrates particularly impressive results in the ARC-AGI-2 visual reasoning test, performing 2.5 times better than its competitors.

The Deep Think mode makes Gemini 3 well-suited for demanding tasks including:

• Advanced code generation and development
• Complex mathematical problem-solving
• Scientific research and analysis

The new capabilities are now available for testing at the official Gemini platform.

Source: https://gemini.google.com/