× News Alerts AI News CyberSec News Let's Talk Local AI Bank Tech News Cyber Advisories Contact

Google Gemini 2.5 Deep Think: Rolled-out for AI Ultra Subs

Google has rolled out Gemini 1.5 Pro to Gemini Advanced subscribers, offering a massive 1 million token context window for advanced data analysis. The model supports multiple modalities, including video, audio, and large documents, and is available in over 35 languages.

Google Gemini 2.5 Deep Think: Rolled-out for AI Ultra Subs

Google has rolled out Gemini 2.5 Deep Think, an advanced reasoning model, to all Google AI Ultra subscribers. The new feature, which was initially announced at Google I/O, is designed to handle complex problems that require a more strategic and analytical approach than previous models.

Deep Think utilizes what Google describes as "parallel thinking techniques," allowing it to generate and evaluate multiple potential solutions simultaneously. This method is intended to give the model more time to explore different approaches before arriving at a final response, potentially leading to more nuanced outcomes.

The announcement follows an advanced version of Deep Think reportedly achieving a gold-medal standard at the 2025 International Mathematical Olympiad (IMO). According to a post on X from Google CEO Sundar Pichai, the model available to subscribers is a more user-focused version, evaluated to perform at a bronze-medal level on the same benchmark.

Google states that Deep Think is suited for tasks requiring iterative development, such as web development and complex coding problems. The company also suggests the model can be a tool for researchers, capable of processing scientific literature and formulating mathematical conjectures. It was credited with assisting a mathematician in proving long-standing conjectures.

For Google AI Ultra subscribers, Deep Think is now accessible within the Gemini app. Users can activate the new mode with a dedicated button in the prompt bar. The release follows a testing period involving a small group of academics and mathematicians.

This latest version of Deep Think is presented as an improvement over the model first previewed in May, with Google citing enhanced benchmark scores in coding and reasoning. The model's "adaptive and budgeted thinking" is said to allow it to assess a task's complexity and allocate an appropriate amount of time and resources for problem-solving.

Google plans to make a more research-oriented version of the gold-medal-winning Deep Think model available to a select group of mathematicians and academics to gather further feedback. For now, the standard Deep Think model is live for Google AI Ultra subscribers.

Subscribe for AI & Cybersecurity news and insights