Gemini News 07/09/2026 AI Rating: Medium

AlphaEvolve, Google's Gemini-Powered Code-Optimization Agent, Is Now Generally Available on Google Cloud

#Gemini#AlphaEvolve#Google Cloud#Google DeepMind#Code Optimization

Google Cloud has made AlphaEvolve, a Gemini-powered agent that automatically discovers more efficient algorithms, generally available to all customers as of July 9, 2026 — expanding a tool that had previously been limited to private preview since December 2025.

Details

  • What it is: AlphaEvolve is a Gemini-powered code-optimization agent, co-developed with Google DeepMind, that acts as an “evolutionary collaborator” for algorithmic problems too large for developers to search by hand
  • How it works: Users supply a baseline algorithm and describe their optimization goals; AlphaEvolve then automatically searches for better-performing solutions and returns human-readable, optimized code rather than a black-box result
  • What’s new: The tool moves from private preview (which began December 2025) to general availability for all Google Cloud customers
  • Where to access it: Through the Gemini Enterprise Agent Platform on Google Cloud
  • Use cases highlighted: Microchip design, logistics network routing, and accelerating medical research, alongside earlier preview work in semiconductors, genomics, financial services, and computing research
  • Early adopters: Google cites companies including BASF, JetBrains, and Kinaxis as early adopters who used AlphaEvolve to solve previously intractable business and research problems, though specific results weren’t detailed in the announcement
  • Pricing: Not disclosed in the announcement

How to try it

  • Google Cloud customers can access AlphaEvolve through the Gemini Enterprise Agent Platform
  • Start with a baseline algorithm and a clear description of what you’re optimizing for (speed, resource usage, cost, etc.)
  • Google Cloud’s blog points to step-by-step implementation guides for getting started
  • Consider it first for algorithmically hard problems — like scheduling, routing, or chip layout — where an exhaustive manual search isn’t practical