Google has upgraded Gemini 2.5 with a feature called Deep Think that “could be a powerful tool in creative problem solving,” according to the company. Deep Think uses parallel thinking techniques, letting Gemini “generate many ideas at once and consider them simultaneously, even revising or combining different ideas over time, before arriving at the best answer.” Its ability to reason through highly complex problems makes Deep Think mode a powerful tool for researchers, Google says, adding that it also excels at coding. Deep Think is available in the Gemini app to subscribers of Google AI Ultra, priced at $250 per month.

Introduced at Google I/O 2025 in May, Gemini 2.5 Deep Think mode utilizes what TechCrunch calls a “multi-agent” approach, unleashing “multiple AI agents to tackle a question in parallel, a process that uses significantly more computational resources than a single agent, but tends to result in better answers.”

“By extending the inference time or ‘thinking time,’ we give Gemini more time to explore different hypotheses, and arrive at creative solutions to complex problems,” Google explains in a blog post, adding that “novel reinforcement learning techniques” have been added to leverage the extended reasoning paths, “thus enabling Deep Think to become a better, more intuitive problem-solver over time.”

Google used a version of Gemini 2.5 Deep Think to achieve the equivalent of a gold medal at the 2025 International Math Olympiad (IMO). TechCrunch reports that the math-focused variant “‘takes hours to reason,’ instead of seconds or minutes like most consumer-facing AI models.”

Simultaneous with the release of Deep Think to AI Ultra subscribers, Google is making the official IMO model available to “select mathematicians.” “The company hopes the IMO model will enhance research efforts, and aims to get feedback on how to improve the multi-agent system for academic use cases,” TechCrunch writes.

Ars Technica provides benchmark charts showing how Gemini 2.5 Deep Think “surpasses the standard Gemini 2.5 Pro and competing models like OpenAI o3 and [xAI’s] Grok 4,” exhibiting “a particularly large gain in Humanity’s Last Exam, a collection of 2,500 complex, multi-modal questions that cover more than 100 subjects”

Google also plans to release Deep Think via the Gemini API, “with and without tools to a set of trusted testers” in the coming weeks “to better understand its usability for developer and enterprise use cases.”