Google DeepMind’s AlphaEvolve is now generally available on Google Cloud, a little over a year after it was first announced. AlphaEvolve is an evolutionary coding agent built on Gemini: you give it a working program and a scoring function, it generates mutated versions of the code, tests each against your evaluator, keeps the best, and repeats until it finds something better than where you started. DeepMind says the same engine has already helped redesign Google’s next-generation TPU chips, cut the data written to its Spanner database by 20%, and reduced software storage footprints by nearly 9%.
The general release is what makes this matter beyond Google. Early customers report large gains: BASF improved its supply-chain planning and forecasting models by over 80%, Klarna doubled the training speed of a large AI model without hurting quality, and FM Logistic cut warehouse routing waste by 10.4%, saving more than 15,000 kilometers of staff travel a year. Schrödinger roughly quadrupled the speed of its molecular simulations for drug discovery. You supply only a seed program and an evaluator, and AlphaEvolve does the searching.
The larger shift is that AI is beginning to improve the machinery underneath itself. A circuit AlphaEvolve proposed was integrated directly into Google’s next-generation TPUs, which Jeff Dean described as “TPU brains helping design next-generation TPU bodies.” Algorithm discovery is becoming a loop that runs largely on its own.
Sources:
- AlphaEvolve is available for everyone (Google Cloud, July 10, 2026)
- AlphaEvolve: How our Gemini-powered coding agent is scaling impact across fields (Google DeepMind)
- AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms (Google DeepMind)
- How FM Logistic tackled the traveling salesman problem at warehouse scale with AlphaEvolve (Google Cloud)
- Overview of AlphaEvolve (Google Cloud docs)
Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service
Reuse
Citation
@misc{kabui2026,
author = {{Kabui, Charles}},
title = {AlphaEvolve: {The} {AI} {Agent} {That} {Evolves} {Better}
{Algorithms,} {Now} on {Google} {Cloud}},
date = {2026-08-26},
url = {https://toknow.ai/posts/alphaevolve-gemini-evolutionary-algorithm-agent-google-cloud/},
langid = {en-GB}
}
