AlphaEvolve, Google’s Gemini-powered evolutionary coding agent, is now generally available to all Google Cloud customers — rolled out widely in partnership with Google DeepMind.
The agent targets algorithmic optimization problems with solution spaces too large for manual exploration: microchip design, logistics routing, GPU kernel generation, quantum computing.
Concrete results so far:
- GPU kernels — generated exascale kernels running on the Frontier supercomputer
- Quantum algorithms — discovered new approaches to quantum computation
- Hardware design — applied to microchip layout optimization
- Logistics — routing network efficiency gains
- Medical research — computational pipeline acceleration
AlphaEvolve orchestrates a pipeline of LLMs that propose and refine code through iterative edits. An evaluator function scores each variant, and the best survive to the next generation. It’s evolutionary search applied to code — not a single-pass generator.
NOTE
AlphaEvolve needs a defined evaluator function to work. Problems where solution quality isn’t measurable as a score won’t benefit. This is an optimization agent for hard search problems, not a general-purpose coding assistant.
Available now to all Google Cloud customers. Details and access are on the Google Cloud AlphaEvolve page.