How do I access AlphaEvolve now that it's GA?

AlphaEvolve is available to every Google Cloud customer through the Gemini Enterprise Agent Platform. You submit a baseline seed program and a client-side evaluator script; AlphaEvolve then runs its evolutionary search on Google's side and returns optimized human-readable code you can drop back into production.

What problems has AlphaEvolve actually solved so far?

Since December 2025, private-preview adopters BASF, JetBrains and Kinaxis have applied AlphaEvolve to logistics, supply-chain, semiconductor, genomics, HPC and financial-services problems. Earlier DeepMind runs used the same agent to improve matrix-multiplication kernels and Google's own Spanner write amplification and data-center scheduling.

How does AlphaEvolve differ from a normal coding agent like Codex or Claude Code?

AlphaEvolve is an evolutionary search agent, not an IDE assistant. It generates candidate code, runs each variant through a user-supplied evaluator that scores correctness and performance, keeps the winners, and iterates for hours or days — so the win case is a single hot algorithm you want to squeeze, not day-to-day coding.

Do I need to write anything besides my baseline algorithm?

AlphaEvolve requires two inputs from you. First, a seed program with the code segments you want optimized marked out. Second, a deterministic client-side evaluator that compiles, runs and scores each candidate and returns a scalar metric. The evaluator is what pins the search to real correctness.

Is AlphaEvolve open source?

No. AlphaEvolve is a managed Google Cloud service on the Gemini Enterprise Agent Platform. The research paper and DeepMind's write-ups are public, but the agent, the Gemini backbone, and the search harness stay behind Google Cloud. Google published a `Google-Cloud-AI/alphaevolve-on-googlecloud` sample repo for integration patterns.

Google Cloud · 2026-07-09 · major

AlphaEvolve GA — Google Cloud opens Gemini's code-optimization agent to everyone

AlphaEvolve, the Gemini-powered agent that DeepMind used to break decades-old algorithms, is now generally available on the Gemini Enterprise Agent Platform for every Google Cloud customer after seven months in private preview.

AlphaEvolve hero graphic from Google Cloud's GA announcement

The Gemini agent that beat decades-old algorithms now runs for every Google Cloud customer.

Quick facts

Maker	Google Cloud (Google DeepMind)
Availability	GA on Gemini Enterprise Agent Platform
Prior status	Private preview since December 2025
Backbone	Gemini models
Early adopters	BASF, JetBrains, Kinaxis
Inputs	Seed program + deterministic evaluator

What is it?

AlphaEvolve is Google DeepMind's evolutionary code-search agent, wrapped as a managed Google Cloud service. You give it a baseline algorithm and an evaluator that scores candidate programs; AlphaEvolve then uses Gemini to propose variants, runs each one through your evaluator, keeps the winners, and iterates for hours or days.

How does it work?

The workflow has four steps: define the problem and a seed program, define an evaluator that returns a scalar score, let AlphaEvolve's agentic harness run the search, then take the winning optimized code back to production. Gemini does the creative candidate-generation on Google's side; the evaluator runs client-side, so correctness and performance are pinned to code you control.

Why does it matter?

Until today, AlphaEvolve was locked to a handful of private-preview partners like BASF, JetBrains and Kinaxis. Opening it up means any Google Cloud team with a hot algorithm — routing, chip placement, genomics pipelines, HPC kernels — can throw a Gemini-scale search at it instead of hand-optimizing.

Who is it for?

Google Cloud teams with a specific algorithm they need to squeeze — logistics, semiconductor design, HPC, computational biology, quant finance.

Frequently asked questions

How do I access AlphaEvolve now that it's GA?: AlphaEvolve is available to every Google Cloud customer through the Gemini Enterprise Agent Platform. You submit a baseline seed program and a client-side evaluator script; AlphaEvolve then runs its evolutionary search on Google's side and returns optimized human-readable code you can drop back into production.
What problems has AlphaEvolve actually solved so far?: Since December 2025, private-preview adopters BASF, JetBrains and Kinaxis have applied AlphaEvolve to logistics, supply-chain, semiconductor, genomics, HPC and financial-services problems. Earlier DeepMind runs used the same agent to improve matrix-multiplication kernels and Google's own Spanner write amplification and data-center scheduling.
How does AlphaEvolve differ from a normal coding agent like Codex or Claude Code?: AlphaEvolve is an evolutionary search agent, not an IDE assistant. It generates candidate code, runs each variant through a user-supplied evaluator that scores correctness and performance, keeps the winners, and iterates for hours or days — so the win case is a single hot algorithm you want to squeeze, not day-to-day coding.
Do I need to write anything besides my baseline algorithm?: AlphaEvolve requires two inputs from you. First, a seed program with the code segments you want optimized marked out. Second, a deterministic client-side evaluator that compiles, runs and scores each candidate and returns a scalar metric. The evaluator is what pins the search to real correctness.
Is AlphaEvolve open source?: No. AlphaEvolve is a managed Google Cloud service on the Gemini Enterprise Agent Platform. The research paper and DeepMind's write-ups are public, but the agent, the Gemini backbone, and the search harness stay behind Google Cloud. Google published a `Google-Cloud-AI/alphaevolve-on-googlecloud` sample repo for integration patterns.

Try it

Console: console.cloud.google.com → Gemini Enterprise Agent Platform → AlphaEvolve