Google Cloud makes AlphaEvolve code agent generally available

The Gemini-based system requires users to supply a seed program and evaluator, with research deployments spanning aging models, quantum computing and supercomputing

Google Cloud has made AlphaEvolve generally available through the Gemini Enterprise Agent Platform

Google Cloud has made AlphaEvolve generally available on the Gemini Enterprise Agent Platform, moving the algorithm discovery and code optimization agent beyond the private preview introduced in 2025.

Built on Gemini, AlphaEvolve is designed to search possible algorithms and code implementations using performance measures defined by the organization deploying it. Google positions the product for scientists, engineers and businesses working on problems across logistics, semiconductors, genomics, high-performance computing and financial services.

Pushmeet Kohli, Chief Scientist, Google Cloud and Vice President, Science at Google DeepMind, shared the general availability release on LinkedIn.

In the accompanying Google Cloud article, Kohli describes a shift from using AI for task support toward computational discovery: "AI is moving beyond acting as a productivity assistant that accelerates how we work to a discovery engine that expands what we can achieve."

General availability expands access to AlphaEvolve, but the product is not a one-step code generation tool. Users must already have a baseline algorithm and define how each proposed alternative will be tested and scored.

Users must define what successful optimization means

Google describes a four-stage process for deploying AlphaEvolve: define, measure, optimize and apply.

Organizations begin by supplying a seed algorithm, a problem definition and relevant background information. They must then create a scoring function covering measures such as correctness, performance and operational constraints.

AlphaEvolve uses that scoring function to generate and test candidate code. The user decides which parts of the original program can be changed and provides a deterministic client-side evaluator that compiles, runs and scores each candidate.

The client-side runner can operate in the user's own environment. It requests mutated programs through the AlphaEvolve API, evaluates them and returns scalar scores that the agent uses to guide further searches.

That process leaves benchmarking, validation and deployment decisions with engineers. Dmitrii Batkovich, Director of Engineering at JetBrains, says: "AlphaEvolve can change how we approach complex performance work. It turns optimizations that were once too time-consuming to explore into candidates we can test routinely. Engineers still own the benchmark, review, and release decision. The search space is what gets smaller."

JetBrains reports that AlphaEvolve improved its integrated development environment performance by more than 15% to 20%.

University and laboratory teams test research applications

Higher education and government research teams have used AlphaEvolve for work in biological aging, quantum computing and scientific software.

The Qin Lab at Old Dominion University used the agent with a high-performance computing cluster in Google Cloud as part of the university's MonarchSphere initiative. Across approximately 500 evaluations, the team says AlphaEvolve independently rediscovered the Kannisto logistic mortality model without prior knowledge of the 1990s research.

The lab also reports a 19% improvement in the Emergent Aging Model composite fitness score and a Strehler-Mildvan correlation of 0.949. Dr. Hong Qin of Old Dominion University's Department of Computer Science states: "The central finding is that structurally diverse models all converge on the same empirical aging laws, providing evidence that Gompertz, Strehler-Mildvan, and Kannisto regularities are robust attractors of biological systems."

The team plans to expand the work to datasets covering multiple species and connect the generated program structures to biological mechanisms that can be tested.

Oak Ridge National Laboratory has deployed AlphaEvolve on Frontier, its exascale supercomputer, through Google DeepMind's Genesis Mission partnership with the US Department of Energy.

The research team created a closed-loop system connecting cloud-based code generation with execution on Frontier's AMD graphics processing units. Candidate mixed-precision GPU kernels were generated, compiled, run and checked against numerical accuracy requirements on the supercomputer.

Google also cites internal uses of AlphaEvolve. The company says the agent reduced write amplification in Google Spanner by 20%, cut software storage footprints by almost 9% and found quantum circuits with 10 times lower error rates for molecular simulations on Google's Willow quantum processor.

Google directs new users to an onboarding guide, API documentation, repository examples and Colab notebooks explaining how the AlphaEvolve process works.

Developers can also access an AlphaEvolve Skill through supported development environments, including Antigravity and Claude Code. More complex projects can use Google's advanced examples and best-practice guidance.