Google Rolls Out Deep Think, Its Most Advanced AI Model for Gemini Ultra Subscribers
Google has launched Deep Think, its most powerful reasoning model yet, for Gemini 2.5 Ultra subscribers. The high-tier AI plan, priced at $250 per month, now includes access to this cutting-edge technology, which was first previewed at Google I/O. After refining the model based on feedback from early testers, Google is making it available to a broader audience.
Deep Think is a modified version of the AI that recently earned a gold-medal standard at the International Mathematical Olympiad. While the public release is not the full gold-medal version, Google states it performs at a bronze-medal level in internal tests. The gold-medal model remains under evaluation, with select mathematicians and researchers gaining exclusive access for further study.
The model uses parallel thinking techniques, enabling Gemini to generate and evaluate multiple ideas simultaneously. It is also multi-modal, capable of processing text, images, and audio. This makes it highly versatile for complex problem-solving and everyday applications.
In benchmark tests, Deep Think outperformed competing models like OpenAI’s o3, xAI’s Grok 4, and even Google’s own Gemini 2.5 Pro. It excelled in assessments such as Humanity’s Last Exam and LiveCodeBench, showcasing superior reasoning and analytical abilities.
Ultra subscribers can now use Deep Think within the Gemini app, though daily usage will be restricted to a fixed number of prompts. Google has not disclosed the exact limit, leaving users to explore the model’s capabilities within these constraints.
This rollout marks another step in Google’s push to lead the AI race, offering premium users access to its most sophisticated reasoning technology. With Deep Think, the company aims to set a new standard for AI performance, though its high subscription cost may limit widespread adoption.
