During Google I/O, the company announced upgrades to Gemini 2.5, highlighting the introduction of Deep Think, an advanced reasoning mode for Gemini 2.5 Pro that excels in complex math and coding tasks. Deep Think utilizes innovative research techniques to evaluate multiple hypotheses, scoring an 84% on the multimodal reasoning test MMMU and performing well on the USAMO. Google is prioritizing safety by involving trusted testers before broader release, and introducing 'thinking budgets' for efficient token management. These updates position Gemini 2.5 as a leader in AI-driven reasoning capabilities.
Google is adding an experimental new reasoning mode to 2.5 Pro called Deep Think, which allows the model to consider multiple hypotheses before responding.
Deep Think scored an 84% on the multimodal reasoning test MMMU and achieved an impressive score on the Mathematics Olympiad, signifying significant enhancement in reasoning capabilities.
We're going to make it available to trusted testers via the Gemini API to get their feedback before making it widely available, focusing on safety evaluations.
Pro updates will include thinking budgets, allowing developers to manage cost and quality in AI responses, which enhances efficiency in resource usage.
Collection
[
|
...
]