Google unveils Gemini 2.5 AI models for enterprises

Google has unveiled the Gemini 2.5 series of AI models, designed to enhance performance in enterprise applications through improved reasoning capabilities.

The Gemini 2.5 Pro model, now available for public preview on Vertex AI, represents a significant development in AI technology. According to Jason Gelman, Director of Product Management for Vertex AI, it ranks highly in global benchmarks for tasks requiring advanced reasoning and coding. Known for its deep reasoning abilities, the model allows for comprehensive data analysis and handling of complex coding tasks.

Yashodha Bhavnani, Vice President of AI Product Management at Box, highlights how Gemini is transforming content management. "At Box, we're redefining how enterprises apply intelligence to their content. With Box AI extract agents, powered by Gemini, users can instantly streamline tasks by making unstructured data actionable, as seen in millions of extractions supporting a variety of use cases, including procurement and reporting. Gemini 2.5 represents a leap forward in advanced reasoning, enabling us to envision building more powerful agent systems where extracted insights automatically trigger downstream actions and coordinate across multiple steps. This evolution pushes the boundaries of automation, allowing businesses to unlock and act upon their most valuable information with even greater impact and efficiency."

Moody's is also exploring the capabilities of Gemini 2.5, as noted by Wade Moss, Senior Director of AI Data Solutions. "Moody's leverages Gemini's advanced reasoning capabilities on Vertex AI within a model-agnostic framework. Our current production system uses Gemini 2.0 Flash for intelligent filtering and Gemini 1.5 Pro for high-precision extraction, achieving over 95% accuracy and an 80% reduction in processing time for complex PDFs. Building on this success, we are now in the early stages of testing Gemini 2.5 Pro. Its potential for deeper, structured reasoning across extensive document sets, thanks to features like its large context window, looks very promising for tackling even more complex data challenges and enhancing our data coverage further. While it's not in production, the initial results are very encouraging."

To address different enterprise needs, Google is launching Gemini 2.5 Flash, which is designed for applications requiring low latency and cost efficiency. This model is targeted at scenarios such as customer service and real-time information processing, offering dynamic reasoning capabilities.

Rajesh Bhagwat, Vice President of Engineering at Palo Alto Networks, discussed the utility of Gemini 2.5 Flash for the company's AI-driven initiatives. "Gemini 2.5 Flash's enhanced reasoning ability, including its insightful responses, holds immense potential for Palo Alto Networks, including detection of future AI-powered threats and more effective customer support across our AI portfolio. We are focused on evaluating the latest model's impact on AI-assistant performance, including its summaries and responses, with the intention of migrating to this model to unlock its advanced capabilities."

Google is also introducing new features in Vertex AI, such as supervised tuning and context caching for the Gemini 2.5 models. These advancements aim to tailor AI capabilities to specific enterprise requirements, enhancing performance while managing costs. Furthermore, a new feature called Vertex AI Model Optimizer is being trialled to automatically balance quality and cost for each task.

Additionally, the company is expanding its services with innovations supporting multi-agent ecosystems, including a Live API. This is intended to facilitate real-time, interactive AI applications across various domains, providing capabilities such as streaming audio processing and integration of multilingual outputs.