A supercharged version of Google’s Gemini 2.5 large language model recently reached gold medal status at the International Mathematical Olympiad. Now you can ask a version of it (only a bronze medalist) to answer your toughest math questions. Like, how am I going to pay $250 a month for this AI subscription?

Naturally, this new version of Google’s Gemini 2.5 Deep Think is designed for complicated questions that require much more work than you’d expect from a free or cheap AI chatbot. Google said it’s designed to handle prompts that require strategic planning, creativity and deep reasoning.

Google’s making it available now to subscribers of the company’s $250-per-month AI Ultra plan. That subscription, first announced earlier this year at Google’s I/O developer conference, is a steep price to pay for generative AI tools. The tech giant teased then that this model would be among the offerings to make that price worthwhile. (You can still access other Gemini models either for free or at a $20-per-month tier.)

Generative AI developers are intent on building ever more powerful models, and one big way they’re doing so is by expanding their ability for research and increasing the amount of time these models can work. OpenAI has a Deep Research tool for ChatGPT users, and Google’s Gemini has a tool by the same name for all users.

While those research modes allow models more time, Deep Think uses “parallel thinking” to consider different ideas at once, combining or revising them as it goes. Deep Think also just has more time to think and process, which helps it find better answers but also costs more money and uses more computing power.