Google ⁢AI has launched Gemini 3 Deep Think mode within the Gemini app for its⁢ Ultra subscribers. This‍ new mode represents a significant leap in advanced reasoning ⁣capabilities, specifically designed ⁣to address complex challenges in mathematics, science, and logic – areas that traditionally push the boundaries of even the most sophisticated artificial intelligence models.

Unprecedented Performance on Key‍ Benchmarks

Gemini 3 ⁤deep ‍Think demonstrates industry-leading performance on‍ demanding benchmarks. It⁣ achieved a score⁢ of 41.0% on Humanity’s Last Exam without utilizing external tools, and an exceptional 45.1% on the ARC-AGI-2 benchmark when code execution was ⁢enabled. these results highlight the model’s ability to not only understand complex problems but also to formulate and⁣ test multiple⁣ hypotheses in parallel.

Building on a Foundation of Competitive Success

The advancements in Gemini 3 Deep Think are rooted in the success of earlier Gemini 2.5 deep Think variants. These predecessors earned gold-medal recognition at both the International Mathematical Olympiad and the International Collegiate Programming Contest World Finals, demonstrating a proven ability to excel in highly ⁢competitive⁤ problem-solving environments.

Accessing Deep Think Mode

Google AI Ultra subscribers can instantly access Gemini 3 Deep ⁢Think mode. To activate ⁣it, simply select “Deep Think” from the prompt bar⁢ and choose⁢ “Gemini 3 Pro” from the model dropdown menu within the Gemini app.