Introducing Google’s Deep Think: A New Era in AI Problem Solving

Just weeks after Google and OpenAI celebrated their models’ stunning achievements at the International Math Olympiad (IMO), Google is set to unleash its cutting-edge Deep Think model to the world—but there’s a catch: it comes at a premium.

On Friday, Google rolled out Deep Think within the Gemini app, exclusively for subscribers of its Google Ultra plan, priced at $250 annually or just $125 for the first three months. This model isn’t just any adaptation; it’s designed for faster, everyday use while still packing a punch with bronze-level capabilities on the 2025 IMO benchmarks.

How Deep Think Works its Magic

What sets Deep Think apart? This AI marvel employs a unique parallel thinking technique, allowing it to juggle multiple ideas simultaneously—often merging concepts to pinpoint the best solution. It’s not just quick; it’s also smart.

Enhanced inference time—essentially its ‘thinking’ period—plays a pivotal role, as it explores various options before delivering answers. Coupled with innovative reinforcement learning strategies, Deep Think continually hones its problem-solving skills.

Excelling in Various Domains

According to Google, Deep Think shines in iterative development, scientific discovery, and coding. In fact, its performance metrics from benchmarks like Humanity's Last Exam—featuring over 100 subject areas—demonstrate its prowess across disciplines including math and the sciences.

Improved Safety and Objectivity