Google Announces Major Upgrade to Gemini 3 Deep Think AI Model
Google has unveiled a significant update to its Gemini 3 Deep Think AI model, designed to accelerate the handling of complex, real-world problems. The tech giant's CEO, Sundar Pichai, described this as a "significant upgrade," emphasizing that the model has been refined through close collaboration with scientists and researchers. This enhancement aims to tackle "messy" challenges in fields such as science, engineering, and advanced mathematics, where solutions often lack a single correct answer.
Record-Breaking Performance on Key Benchmarks
In a post on X, Pichai highlighted that Gemini 3 Deep Think is pushing the boundaries of AI capabilities. The model has achieved an unprecedented 84.6% score on the ARC-AGI-2 benchmark, setting a new standard for general intelligence reasoning. Additionally, it scored 48.4% on Humanity's Last Exam (HLE), a difficult test for advanced language models, without using any external tools. Google claims these results demonstrate "frontier" levels of intelligence, with the model "crushing" some of the hardest AI tests ever created.
Further achievements include reaching Gold Medal levels on the 2025 International Math Olympiad and attaining a massive Elo rating of 3455 in competitive programming. These milestones underscore the model's enhanced reasoning and problem-solving abilities, positioning it as a leader in the AI landscape.
Expanding Capabilities in Science and Engineering
While previous versions excelled in logic, the upgraded Gemini 3 Deep Think is now being promoted as a "scientist's assistant." Google reports that the model can navigate complex domains like advanced chemistry and theoretical physics with greater proficiency. For instance, engineers are utilizing it to model physical systems through code, and researchers are employing it to interpret dense data sets that were previously too challenging for AI to handle effectively.
This advancement marks a shift towards more practical applications, enabling professionals to leverage AI for intricate scientific and technical tasks.
Availability for Consumers and Professionals
Google has made the updated Gemini 3 Deep Think mode available to consumers who are Google AI Ultra subscribers, accessible directly through the Gemini app starting today. For professionals, the company is launching an Early Access Program for the Gemini API, allowing select researchers, engineers, and large businesses to build custom tools using Deep Think's reasoning power.
Pichai confirmed, "The updated Gemini 3 Deep Think mode is now available for Ultra subscribers in the Gemini app. We're also, for the first time, making Deep Think available via the Gemini API to select researchers & enterprises through an early access program." This move aims to foster innovation and collaboration in the AI community.
Implications for the Future of AI
This upgrade signifies Google's commitment to advancing AI technology for real-world problem-solving. By achieving record scores on rigorous benchmarks and expanding into scientific fields, Gemini 3 Deep Think sets a new precedent for AI performance. The availability through both consumer and professional channels ensures that a wide range of users can benefit from its enhanced capabilities, potentially accelerating breakthroughs in various industries.
As AI continues to evolve, models like Gemini 3 Deep Think are poised to play a crucial role in addressing complex global challenges, from scientific research to engineering innovations.
