Beyond these sectors, its reach spans semiconductors, software performance, forecasting, and large ML training pipelines, with participants including Infineon, JetBrains, Kinaxis, Klarna, and Kuro Games.

AlphaEvolve, Google's Gemini-based code optimization and algorithm discovery agent, has been publicly released and is now generally available to all Google Cloud customers on the Gemini Enterprise Agent Platform.

Google provides a Cloud blog post with step-by-step guides detailing AlphaEvolve’s availability and how to use it.

The model reportedly rediscovered cutting-edge solutions for a majority of unsolved problems and aided AI model improvements for disaster-risk prediction and quantum-circuit design for molecular simulations.

In academia and research, AlphaEvolve has been applied to Frontier at Oak Ridge National Laboratory for mixed-precision GPU kernel optimization and to Python models of biological aging, with PacBio noting higher sequencing accuracy.

The tool is highlighted for tackling complex optimization challenges across chip design, logistics routing, and accelerating medical research.

In life sciences and industry, AlphaEvolve has improved DNA analysis models (reducing mutation detection errors by about 30%) and helped BASF boost planning and forecasting performance by more than 80%, with FM Logistics achieving near-optimal routing improvements.

Internally, Google has used AlphaEvolve to optimize silicon design for next-gen TPUs, reduce write amplification in Spanner by about 20%, and cut software storage footprints by roughly 9% through compiler changes.

The tool acts as an evolutionary collaborator, taking a baseline algorithm and goals and automatically searching for better solutions, returning optimized, human-readable code rather than starting from scratch.

Since private preview in December, early adopters like BASF, JetBrains, and Kinaxis have reported solving previously intractable problems with AlphaEvolve.

Early customers report performance gains across various domains, including BASF’s digital twin for supply networks, Coolblue’s 28-day forecast pipeline (over 5% accuracy gain), and FM Logistics’ warehouse routing (about 10% improvement).

Public deployment guidance outlines a four-step workflow: define context, establish an evaluation score, generate optimized code, and apply a production-ready algorithm to workloads.