# OpenEvolve review

> OpenEvolve is an open-source implementation of AlphaEvolve that uses LLMs in an evolutionary loop to iteratively improve programs against an evaluator.

Devtoolwatch Board, the week of Oct 25–31, 2026. Canonical page: https://index.devtoolwatch.com/tools/algorithmicsuperintelligence-openevolve/

## TL;DR

OpenEvolve is an open-source Python implementation of AlphaEvolve. It evolves code using LLMs combined with MAP-Elites and island-based populations, scored by your evaluator. It suits developers and researchers working on performance optimization, algorithm discovery or scientific computing, and it supports seeded, reproducible runs and many LLM backends.

## OpenEvolve at a glance

- **Category:** [Coding agents](https://index.devtoolwatch.com/categories/coding-agents/)
- **Maintained by:** Algorithmic SuperIntelligence Labs (organization, [@algorithmicsuperintelligence](https://github.com/algorithmicsuperintelligence) on GitHub)
- **License:** Apache 2.0
- **Language:** Python, JavaScript, CSS
- **Repository:** [github.com/algorithmicsuperintelligence/openevolve](https://github.com/algorithmicsuperintelligence/openevolve)
- **PyPI package:** [openevolve](https://pypi.org/project/openevolve/) 0.4.0
- **Latest release:** [v0.4.0](https://github.com/algorithmicsuperintelligence/openevolve/releases/tag/v0.4.0) on GitHub, Sep 28, 2026
- **First published:** May 15, 2025

## What OpenEvolve does

- OpenEvolve evolves an initial program over many iterations, using LLMs to propose changes and a user-supplied evaluator to score them.
- OpenEvolve maintains diverse populations with MAP-Elites quality-diversity evolution across multiple islands, with ring-topology migration between them.
- OpenEvolve can combine multiple LLMs in a weighted ensemble with fallback strategies.
- OpenEvolve seeds every component, with a default seed of 42, so runs can be reproduced across machines.
- OpenEvolve can be used as a Python library through run_evolution and evolve_function, with inline code and no external files.
- OpenEvolve supports pluggable island selectors and population strategy hooks for admission, cell replacement, archiving, eviction and migration.
- OpenEvolve tracks LLM token usage per iteration and as a run total for OpenAI-compatible responses.

### How it works

You give OpenEvolve an initial program, an evaluator and a YAML config, then run openevolve-run.py or call the library API. It connects to any OpenAI-compatible API, including OpenAI, Gemini, Ollama, vLLM and the OptiLLM proxy. It can also use the Claude Code CLI, GitHub Copilot CLI or Codex app-server as backends without API keys. Docker images are available.

## How to install OpenEvolve

**With pip**

```
pip install openevolve
```

**Development install**

```
pip install -e ".[dev]"
```

**With Docker**

```
docker pull ghcr.io/algorithmicsuperintelligence/openevolve:latest
```

### Requirements

- Python 3.10 or later.
- Access to an OpenAI-compatible API, or the Claude Code CLI, GitHub Copilot CLI or Codex app-server as a backend.
- Docker is optional, for containerized runs.

### First steps

1. Install OpenEvolve with pip install openevolve.
2. Get a Gemini API key and set it in the OPENAI_API_KEY environment variable, as the example config uses Gemini.
3. Run the function_minimization example with openevolve-run.py, passing the initial program, the evaluator, config.yaml and --iterations 50.
4. Edit config.yaml to switch to another OpenAI-compatible provider or a local model.

## Who OpenEvolve is for

- Developers optimizing performance-critical code, such as GPU kernels, against a measurable benchmark.
- Researchers doing algorithm discovery who need reproducible, seeded evolutionary runs.
- Scientific computing users automating manual tuning work, such as filter design.
- Teams with existing AlphaEvolve-style local evaluators they want to reuse unchanged.

### Limitations

- OpenEvolve requires you to write an evaluator that scores candidate programs. Results depend on that scoring.
- OpenEvolve's API-based runs incur per-iteration LLM costs that depend on the model and the code size.
- OpenEvolve's Codex app-server backend is unsupported on Windows.
- On macOS and Windows, OpenEvolve library calls must be wrapped in if __name__ == '__main__': to avoid multiprocessing errors.

## Usage data (the week of Oct 25–31, 2026)

- Board rank: #356 overall, #105 in Coding agents, momentum score 20
- Downloads that week: 7.1k (−30% vs 4-week average)
- New GitHub stars per week (last 12 weeks): Aug 9: 302; Aug 16: 80; Aug 23: 34; Aug 30: 50; Sep 6: 31; Sep 13: 46; Sep 20: 41; Sep 27: 46; Oct 4: 39; Oct 11: 67; Oct 18: 64; Oct 25: 43

## OpenEvolve alternatives

| Rank | Tool | Stars | Downloads/wk |
|---|---|---|---|
| #1 | [Cindy](https://index.devtoolwatch.com/tools/makecindy-cindy/) | 8.8k | – |
| #2 | [OpenMausBot](https://index.devtoolwatch.com/tools/milind-soni-openmausbot/) | 9.6k | – |
| #3 | [Orca](https://index.devtoolwatch.com/tools/stablyai-orca/) | 111.7k | – |
| #4 | [OpenBot](https://index.devtoolwatch.com/tools/copilotkit-openbot/) | 9.1k | – |
| #5 | [Agent Reach](https://index.devtoolwatch.com/tools/panniantong-agent-reach/) | 107.2k | – |
| #6 | [Skill Recorder](https://index.devtoolwatch.com/tools/microsoft-skill-recorder/) | 5.6k | – |

## OpenEvolve FAQ

### Is OpenEvolve free and open source?

Yes. OpenEvolve's source code is on GitHub (algorithmicsuperintelligence/openevolve) under the Apache 2.0 license, so you can use it for free.

### How do I install OpenEvolve?

Run `pip install openevolve`. The openevolve package needs Python 3.10 or newer. The install section above has the other options from the project's README.

### Is OpenEvolve still maintained?

Yes. The algorithmicsuperintelligence/openevolve repository was last updated on Oct 21, 2026, and the latest release, v0.4.0 on GitHub, came out on Sep 28, 2026.

### How popular is OpenEvolve?

OpenEvolve has 7.7k GitHub stars as of Nov 1, 2026. It gained 43 new stars in the week of Oct 25–31, 2026, against a 12-week average of 70 a week. Its main package had 7.1k downloads that week, −30% against its own 4-week average. That made it #105 among AI coding agents on the Devtoolwatch Board.

### What are the best alternatives to OpenEvolve?

Other AI coding agents on the Devtoolwatch Board in the week of Oct 25–31, 2026, by momentum: Cindy (#1), OpenMausBot (#2), Orca (#3), OpenBot (#4), Agent Reach (#5). The alternatives table above has their numbers.

### How does OpenEvolve work?

OpenEvolve runs an evolutionary loop. LLMs propose changes to programs, and your evaluator scores each candidate. MAP-Elites keeps the population diverse across feature dimensions, and several islands evolve in parallel with migration between them. Error feedback from evaluations is passed back as artifacts to inform later generations.

### Is OpenEvolve related to AlphaEvolve?

OpenEvolve describes itself as an open-source implementation of AlphaEvolve. It also includes a walkthrough for using unchanged local evaluators from Google's AlphaEvolve examples.

### Can OpenEvolve run without an API key?

OpenEvolve can run without an API key. The Claude Code CLI, GitHub Copilot CLI and Codex app-server backends authenticate through their own logins or subscriptions. Local models through Ollama or vLLM are another option.

### Are OpenEvolve runs reproducible?

OpenEvolve runs are designed to be reproducible. Every component, including the LLM, database and evaluation, is seeded, with a default seed of 42. The README describes this as exact reproduction of runs across machines.

### What languages can OpenEvolve evolve?

OpenEvolve's examples cover Python, Rust, R and Metal shaders.

## Sources

1. [algorithmicsuperintelligence/openevolve on GitHub](https://github.com/algorithmicsuperintelligence/openevolve)
2. [OpenEvolve README](https://github.com/algorithmicsuperintelligence/openevolve/blob/main/README.md)
3. [openevolve on PyPI](https://pypi.org/project/openevolve/)
4. [Release v0.4.0 (Sep 28, 2026)](https://github.com/algorithmicsuperintelligence/openevolve/releases/tag/v0.4.0)
5. [Release v0.3.2 (Jul 18, 2026)](https://github.com/algorithmicsuperintelligence/openevolve/releases/tag/v0.3.2)
6. [Release v0.3.1 (Jul 14, 2026)](https://github.com/algorithmicsuperintelligence/openevolve/releases/tag/v0.3.1)
7. [Devtoolwatch Board, the week of Oct 25–31, 2026](https://index.devtoolwatch.com/)