Google releases smaller, more efficient Gemini 2.5 Flash

Google has released the Gemini 2.5 Flash, their new AI model that focuses on delivering good performance with efficiency. The model will be launched soon on Google’s AI development platform, Vertex AI. 

“Enterprises face challenges that require intricate information landscapes, multi-step analyses, and making nuanced decisions – tasks demanding that AI doesn’t just process, but also reasons. For these situations, we offer Gemini 2.5 Pro on Vertex AI, engineered for maximum quality and tackling the most complex tasks demanding deep reasoning and coding expertise,” Google shared in a blog post. 

Just like Gemini 2.5 Pro, which was released last month, 2.5 Flash is also a reasoning model. The Gemini 2.5 range of models employ a technique called ‘dynamic thinking’ which allows models to regulate how much time they spend over an answer. 

Developers can also control this “thinking budget” to control costs and fasten the pace when it feels apt. 

“Coupled with a one million token context window, Gemini 2.5 Pro performs deep data analysis, extracts key insights from dense documents like legal contracts or medical records, and handles complex coding tasks by comprehending entire codebases,” the blog post noted. 

Google also said that the newer Gemini models will be brought to on-premises environments like Google Distributed Cloud. The company has also partnered with Nvidia so Gemini models on GDC-compliant Nvidia Blackwells can be bought by customers via Google or other channels.

Additionally, Gemini 2.5 Pro will now power the Deep Research tool which was earlier run on Gemini 2.0 Pro.