OpenEvolve
About this tool
A community implementation inspired by Google DeepMind's AlphaEvolve. Starting with your program and evaluation function, language models repeatedly revise code while MAP-Elites quality-diversity search and parallel island evolution seek improved algorithms. Examples include circle packing, function optimization, GPU kernels and symbolic regression.
- Feature 1: Model-driven evolutionary code optimization with model combinations and fallbacks.
- Feature 2: MAP-Elites quality-diversity search and parallel island evolution.
- Feature 3: Cascaded evaluation feeds execution results into the next prompt.
- Feature 4: Support for Python, Rust, R, Metal and other languages.
- Feature 5: Evolution-tree and performance visualizations, with random seeds for reproducibility.
Resource information
- Maintainer
- algorithmicsuperintelligence community organization; not an official DeepMind project.
- Type
- Tool / library
- Category
- Research agents & deep research
- Runs on
- Tool or library
- Who it's for
- Researchers in algorithms, combinatorial optimization, mathematics and computational science improving heuristics, mathematical constructions, GPU kernels or scientific code.
- License at time of listing
- Apache-2.0
- Source
- GitHub
