Google has announced that AlphaEvolve, a project that was initiated by DeepMind, is now generally available on the Gemini Enterprise Agent Platform.
AlphaEvolve is an AI coding agent developed by Google DeepMind that uses large language models to automatically generate and improve algorithms. Google DeepMind is the main AI research lab and division of Alphabet, Google’s parent company.
Announcing the release of AlphaEvolve, Tom Beyer, Group AI Product Manager, Google Cloud, said:
Rather than rewriting code from scratch, AlphaEvolve acts as an evolutionary collaborator: you provide a baseline algorithm and your goals, and it automatically searches for better solutions, returning human-readable, optimized code.
AlphaEvolve pairs the problem-solving capabilities of Google’s Gemini models with automated evaluators that verify answers, along with an evolutionary framework to improve upon the most promising ideas.
Developers work with AlphaEvolve by first generating a “seed” algorithm alongside background information and details of what you want to optimize.
The developer also sets out a scoring function that can objectively score potentially better versions against one or more metrics that are important for the problem such as correctness, performance, and operational constraints.
Gemini models are then used to come up with variations of your code. explicitly optimized against the metrics in the scoring function.
The system then selects the best candidates, and further modifies them, combining the best ideas to work towards a better solution over successive generations.
Google says that AlphaEvolve has been integrated into its own operations and enterprise applications, and has delivered measurable improvements. Specifically, it improved the performance of matrix multiplication kernels in the Gemini architecture by 23% and reduced Gemini’s overall training time.
It was also used in the design of arithmetic circuits and cache replacement policies for Google’s TPUs (Tensor Processing Units). These are Google’s custom-designed application-specific integrated circuits (ASICs) that have been designed to accelerate machine learning and AI workloads. AlphaEvolve also refined Google Spanner’s Log-Structured Merge-tree compaction heuristics to reduce write amplification by 20%.
AlphaEvolve is available now on the Gemini Enterprise Agent Platform
More Information
Gemini Enterprise Agent Platform
Related Articles
Google Introduces Developer Knowledge API
Google Releases FunctionGemma Model
Google Releases Gemini 3 Flash
To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Facebook or Linkedin.
Comments
Make a Comment or View Existing Comments Using Disqus
or email your comment to: comments@i-programmer.info