Google has launched Gemini 2.5 Deep Think, an advanced AI model tailored for complex reasoning tasks. Unlike traditional AIs that generate responses in seconds, Deep Think takes several minutes for thorough analysis, examining multiple problem-solving approaches simultaneously. Initially previewed at the I/O conference and now available to AI Ultra subscribers for $250/month, this model showcases impressive capabilities, achieving bronze-level performance in International Mathematical Olympiad problems and securing a gold medal in research variants.
Deep Think scored 87.6% on LiveCodeBench, outperforming earlier benchmarks, and registered a 34.8% accuracy on multi-subject assessments. Users get limited queries through the Gemini interface, with Google planning to roll out a developer API for enterprise applications soon. This release emphasizes the competitive landscape in AI, as Google continues to innovate amidst rivals like OpenAI. With premium access and ongoing safety evaluations, Google seeks trusted feedback before broader availability of this groundbreaking technology.
Source link
