Wednesday, July 23, 2025
HomeGadgetsGoogle launches Gemini 2.5 Pro IO edition, optimized for coding and software...

Google launches Gemini 2.5 Pro IO edition, optimized for coding and software development

Published on

spot_img



Google is gearing up for its annual showcase — Google I/O 2025, scheduled for May 20. While the event is expected to unveil major software updates like Android 16, Android XR, and new Gemini AI features, Google decided to drop a little surprise ahead of time. DeepMind CEO Demis Hassabis couldn’t hold back and revealed upgrades to what he calls their “most intelligent model” yet. In his words, these are “pre-I/O goodies” — with the highlight being that Gemini 2.5 Pro (Preview) now comes with massively improved coding capabilities.

Google Gemini 2.5 Pro (I/O Edition) Brings Major Upgrades for Developers

Just ahead of its big stage debut at Google I/O 2025, Google has quietly released a preview of its enhanced Gemini 2.5 Pro, now branded the I/O Edition. Internally labelled gemini-2.5-pro-preview-05-06, this version takes a big step forward, especially when it comes to coding. Google says the model now delivers significant improvements in code transformation, code editing, and even in developing complex agentic workflows — making it far more capable for software developers and engineers.

This isn’t a minor bump — it builds on the original Gemini 2.5 Pro, which launched in March, and pushes the architecture to handle more demanding and intricate programming tasks. Google calls every model in the 2.5 line a “thinking model,” designed to reason through its responses, and that design is clearly paying off here.

In terms of performance, Gemini 2.5 Pro (I/O Edition) now leads the LMArena leaderboard (which is based on human preference), and tops academic benchmarks like AIME 2025 for math and GPQA diamond for science — and it does all this without relying on test-time tricks like majority voting.

Where it really shines is programming. Google claims it’s a “big leap over 2.0” with better support for building rich web apps, agent-driven tools, and performing advanced code edits. In the SWE-Bench Verified test for autonomous software engineering, it scored 63.8% using a custom agent-based setup — showing it’s more than ready for real-world development use cases.



Source link

Latest articles

Release Notes podcast on AI features in Search

What does it take to enable billions of people to truly ask anything...

Tech Billionaires Accused of Quietly Working to Implement “Corporate Dictatorship”

In a world increasingly ruled by tech companies, some of the industry’s most...

Early Anthropic hire raises $15M to insure AI agents and help startups deploy safely

Want smarter insights in your inbox? Sign up for our weekly newsletters to...

More like this

Release Notes podcast on AI features in Search

What does it take to enable billions of people to truly ask anything...

Tech Billionaires Accused of Quietly Working to Implement “Corporate Dictatorship”

In a world increasingly ruled by tech companies, some of the industry’s most...