Google has officially rolled out an enhanced version of its Gemini 2.5 Deep Think model, now accessible to subscribers of the Google AI Ultra plan. This upgraded reasoning model, first introduced at Google I/O in May, is designed to tackle complex challenges, including advanced mathematics and coding tasks.
The latest iteration of Deep Think is engineered for speed and efficiency, allowing users to engage with intricate problems that demand strategic thinking and creativity. Unlike its predecessor, which could take hours to solve difficult issues, this new version is optimized for everyday applications while still achieving a commendable Bronze-level score in internal International Math Olympiad (IMO) assessments.
Strong benchmark results show Deep Think’s potential
While Google has been somewhat reticent about the specific architecture of Deep Think, it is believed to operate similarly to OpenAI’s advanced models, utilizing a sophisticated engine to synthesize optimal solutions. This model has distinguished itself in various benchmarks, excelling in LiveCodeBench V6, a competitive coding evaluation, and performing admirably on Humanity’s Last Exam, a rigorous assessment of advanced scientific and mathematical concepts.
Gemini 2.5 Deep Think significantly enhances the value of the Google AI Ultra subscription, which also offers early access to experimental video models, increased limits on tools like NotebookLM, a prototype AI agent manager, and a generous 30TB of storage. The cost for the Google AI Ultra plan is set at 0 per month. For those seeking a more economical option, the newly renamed Google AI Pro plan remains available at annually, catering to a wider audience interested in leveraging AI capabilities.
