menu
close

Google Unveils Gemini 2.5 and Ironwood TPU at I/O 2025

At Google I/O 2025, the company showcased significant advancements in AI with Gemini 2.5 Pro now leading the LMArena leaderboard across all categories. The company also introduced Deep Think, an experimental enhanced reasoning mode for complex math and coding tasks. Additionally, Google unveiled its seventh-generation TPU called Ironwood, delivering 42.5 exaflops of compute power per pod, designed specifically for inferential AI workloads.
Google Unveils Gemini 2.5 and Ironwood TPU at I/O 2025

Google has marked a significant milestone in artificial intelligence development at its annual I/O 2025 conference, demonstrating how its latest technologies are transforming the AI landscape.

CEO Sundar Pichai highlighted the remarkable progress in Google's AI models, noting that Elo scores have increased by more than 300 points since their first-generation Gemini Pro model. The star of the show was Gemini 2.5 Pro, which now dominates the LMArena leaderboard in all categories, showcasing Google's leadership in developing models that align with human preferences.

A major announcement was the introduction of Deep Think, an experimental enhanced reasoning mode for Gemini 2.5 Pro. This feature enables the model to consider multiple hypotheses before responding, resulting in impressive performance on challenging benchmarks. According to Google, Deep Think achieves outstanding scores on the 2025 USAMO mathematics benchmark and leads on LiveCodeBench for competition-level coding.

On the hardware front, Google unveiled Ironwood, its seventh-generation Tensor Processing Unit. This custom AI accelerator is the first designed specifically for inference workloads, delivering 10 times the performance over the previous generation. Each Ironwood pod contains 9,216 chips and provides a massive 42.5 exaflops of computing power, making it ideal for powering thinking and inferential AI models at scale.

Google also showcased Project Mariner, an AI agent that can browse and interact with websites on users' behalf. Initially released as a research prototype in December 2024, Project Mariner has been significantly upgraded to run on cloud-based virtual machines and can now handle up to ten tasks simultaneously. This capability is being integrated into the Gemini API, allowing developers to build more autonomous digital experiences.

These announcements signal what Google calls "a new phase of the AI platform shift" where decades of research are becoming practical reality for people, businesses, and communities worldwide. With these advancements, Google is positioning itself at the forefront of the AI revolution, providing both the models and infrastructure needed to power the next generation of intelligent applications.

Source:

Latest News