Google Cloud Opens AlphaEvolve to All Enterprise Customers

Google Cloud has announced that AlphaEvolve, an AI agent built on Gemini models and developed jointly with Google DeepMind, has reached full general availability on the Gemini Enterprise Agent Platform. The tool, which autonomously searches solution spaces and discovers optimized algorithms, has left private preview after a year of testing with more than fifteen companies across various industries.

How AlphaEvolve works

The mechanism relies on four steps: the user provides a baseline algorithm and defines the problem, then sets an evaluation function to score candidate solutions, after which the system agentically generates and mutates successive versions of the code, testing them against the predefined evaluation function. Solutions that perform better become the starting point for the next generation, creating a feedback loop resembling evolution.

As a result, AlphaEvolve can find algorithms significantly more efficient than the ones the optimization process started from, returning human-readable code to the user that is ready for production deployment.

Results from early customers

Among the companies that tested the tool during private preview, Google Cloud cites BASF, which improved its supply chain planning and forecasting models by 80 percent, Kinaxis, with a 22 percent improvement in forecast accuracy and a 90 percent reduction in model runtime, FM Logistic, which saved 15,000 kilometers of warehouse routing, and Klarna, which doubled the throughput of its machine learning model training pipeline.

In science and engineering, the company points to JetBrains, which achieved 15-20 percent better IDE performance, PacBio, with a 30 percent reduction in errors in DNA sequencing variant detection, and Schrödinger, which used AlphaEvolve to speed up its molecule discovery process fourfold.

Applications inside Google itself

Google also uses AlphaEvolve within its own infrastructure. The tool helped design non-obvious, counterintuitive circuit layouts for its next-generation TPU chips, improved the Log-Structured Merge tree compaction mechanism in the Spanner database, cutting write overhead by 20 percent, and discovered quantum circuits with a tenfold lower error rate for the Willow quantum processor.

AI is no longer just a productivity assistant that speeds up our work, it is becoming a discovery engine that expands what we can achieve. By autonomously searching complex computational spaces, tools like AlphaEvolve help researchers and engineers uncover breakthrough algorithms that reinforce traditional human intuition - Pushmeet Kohli, Chief Scientist at Google Cloud and Vice President of Science at Google DeepMind

What this means for businesses and developers

For Polish tech companies and R&D teams, AlphaEvolve's general availability means that a tool once reserved for internal DeepMind projects is now part of Google Cloud's standard offering, accessible through the Gemini Enterprise platform. Google has not yet disclosed detailed pricing, but stresses that the tool applies to chip design, logistics, finance, genomics, and the optimization of machine learning training architectures.

AlphaEvolve's integration with development environments, including Antigravity and Claude Code, also means that this automated algorithm-discovery tool is starting to interact directly with developers' everyday workflow, not just with research teams working on narrowly defined scientific problems.

Sources: Google Cloud Blog (cloud.google.com), TechBuzz (techbuzz.ai)