Google announced Thursday that the new Deep Think mode is now available to Google AI Ultra subscribers. In this mode, Gemini 3 gains access to extensive computing power and advanced parallel reasoning, enabling it to solve complex problems in disciplines such as mathematics, science, and logic.
According to Google, Gemini 3 Deep Think achieves a score of 41.0 percent on Humanity’s Last Exam without the use of tools and 45.1 percent on ARC-AGI-2 when code execution is available. In addition, Google points to strong performances by previous Deep Think variants in international competitions, including the International Mathematical Olympiad and the International Collegiate Programming Contest World Finals, where models had to solve problems and formulate complete mathematical proofs on their own.
More compute than Gemini 3 Pro
According to Neowin, Gemini 3 Deep Think also scores 93.8 percent on the GPQA Diamond benchmark, a scientific knowledge test that Google itself does not mention in the announcement. Neowin also reports that Deep Think uses more compute and new techniques than the regular Gemini 3 Pro mode, which is in line with Google’s stated emphasis on intensive parallel reasoning methods.
Google says users can activate the new mode by opening the prompt bar in the Gemini app, selecting Deep Think, and choosing Gemini 3 Pro as the model. The feature is currently only available to Google AI Ultra subscribers. According to Neowin, the introduction of Deep Think positions Google prominently in the competitive market for advanced reasoning models, in which OpenAI has also recently shared experimental results without releasing these models.
With the introduction of Gemini 3 Deep Think, Google is emphasizing its ambition to further refine generative AI toward multi-layered and accurate reasoning. The new mode should not only produce answers, but actually reason about complex issues, a step that, according to the official announcement, is intended to shape the next generation of intelligent AI systems.