{"slug": "google-deepmind-open-sources-alphaevolve", "title": "Google DeepMind open-sources AlphaEvolve", "summary": "Google DeepMind released an open-source repository accompanying its AlphaEvolve paper, including a minimal implementation of the evolutionary coding agent with local evaluation under /src and verification code for the new upper bound on the matrix multiplication exponent AlphaEvolve discovered. The /omega directory holds a JAX/Haiku gradient-based optimization program that runs gradient descent over 32,000 steps with 32 random seeds at recursion depth 3, plus a NumPy/SciPy verifier and a Colab notebook using rational arithmetic for rigorous proof. The repository also provides initial programs, discovered results and verification notebooks for 16 mathematical problems where AlphaEvolve beat the state of the art, including circle packing, kissing numbers, cap sets and Shannon capacity of cycle graphs, along with matrix multiplication tensor decomposition notebooks for standard tensor rank and support rank.", "body_md": "This repository accompanies the paper\n\nTODO(franrruiz): Add link to paper.\n\nThere are the following directories:\n\n- \n`/src` contains a minimal implementation of AlphaEvolve with local evaluation.\n- \n`/omega` contains code and data related to the new upper bound on the matrix\nmultiplication exponent discovered by AlphaEvolve. It includes a Colab notebook\nfor rigorous verification using rational arithmetic, and a`level_3/` subdirectory with the gradient-based optimisation code at recursion depth 3:\n  - \n`level_3/final_program.py` — the optimisation program evolved by\nAlphaEvolve. It builds a differentiable upper-bound model in JAX/Haiku,\nruns gradient descent over 32k steps with 32 random seeds, and converts the\nbest solution into a format suitable for verification. Run it directly with`python3 final_program.py` .\n  - \n`level_3/verification.py` — a NumPy/SciPy verifier that takes a flat\nparameter vector produced by the optimiser and evaluates the proven omega\nupper bound. It uses floating-point arithmetic for fast hill-climbing; the\ncompanion Colab notebook uses rational arithmetic for a rigorous proof.\n A `requirements.txt` listing the pinned Python dependencies and a`Dockerfile` for containerised execution are also provided in`level_3/` .\n- \n- \n`/maths` contains the initial programs, discovered results, and verification\ncode for the mathematical problems on which AlphaEvolve found a solution\noutperforming the state-of-the-art (see the \"Results\" section of the paper).\nThere is a subfolder for each of such problems, each containing the files needed\nto run an AlphaEvolve experiment and a Colab notebook to verify the found\nconstruction.\n- \n`/matmul_tensor_decompositions` contains Colab notebooks with the\ndecompositions of matrix multiplication tensors across the matrix sizes for\nwhich the gradient-based optimization method from AlphaEvolve found a solution\noutperforming the state-of-the-art, along with the code to verify their\ncorrectness. Specifically, there are two Colabs: one for the standard tensor\nrank decomposition, and another one for the support rank.\n\nFor the folders `/maths` and `/matmul_tensor_decompositions`, no\ninstallation is required. The provided notebooks can be opened and run in Google\nColab.\n\nFor the folder `/src`, these are the installation instructions:\n\n1. \nInstall docker (required for program evaluation): [https://www.docker.com/get-started/](https://www.docker.com/get-started/)\n2. \n(optional) Create venv:\n\n```\nsudo apt-get install virtualenv python3-venv\nvirtualenv -p python3.11 alpha_evolve\nsource alpha_evolve/bin/activate\n```\n\n1. Install pip dependencies:\n\n```\npip install -r src/requirements.txt\n```\n\n1. Set [Gemini API key](https://ai.google.dev/gemini-api/docs/api-key) :\n\n```\nexport GEMINI_API_KEY=<your_api_key>\n```\n\n- \n`/omega` : The Colab notebook to verify the new upper bound on the matrix\nmultiplication exponent can be opened via:*Note: To verify our best omega upper bound, you must download the\n`clean_level4.npz.xz` file from this repository to your local machine,\nand then upload it when prompted by the Colab notebook.*\n- \n`/maths` : The Colab notebook to verify the found solution for each problem can be opened via:\n  - \nCircle packing:   - \nErdös minimum overlap:   - \nFirst autocorrelation inequality:   - \nSecond autocorrelation inequality:   - \nThird autocorrelation inequality:   - \nHeilbronn problem for convex regions:   - \nHeilbronn problem for triangles:   - \nHexagon packing:   - \nKissing numbers:   - \nMax to min ratio:   - \nSums and differences:   - \nUncertainty principles (Hermite polynomials approach):   - \nUncertainty principles (Laguerre polynomials approach):   - \nExtremal girth 5:   - \nShannon capacity of cycle graphs:   - \nCap sets: \n- \n- \n`/matmul_tensor_decompositions` : There are two Colab notebooks:\n- \n`/src` : We provide an example launch command (for the circle packing problem) below:\n\n```\npython run.py --problem_config=./data/problem_config/circle_packing.yaml\n```\n\n**DISCLAIMER:** AlphaEvolve involves the execution of code generated by Large\nLanguage Models (LLMs). LLM-generated code is intrinsically unpredictable and\nshould be handled as untrusted code. In this version of AlphaEvolve, this code\nis run in a Docker sandbox as an isolation measure. However, despite sandboxing,\nthe potential for unintended behavior, resource misuse, or security\nvulnerabilities cannot be fully eliminated. All users should be aware of these\ninherent risks and understand that they use this software at their own\ndiscretion. Google DeepMind shall not be liable for any claims, damages, or\nother liability arising from the execution of LLM-generated code.\n\nIf you use the code or data in this package, please cite:\n\n```\nTODO(franrruiz): Add citation.\n```\n\nCopyright 2026 Google LLC\n\nAll software is licensed under the Apache License, Version 2.0 (Apache 2.0);\nyou may not use this file except in compliance with the Apache 2.0 license.\nYou may obtain a copy of the Apache 2.0 license at:\n[https://www.apache.org/licenses/LICENSE-2.0](https://www.apache.org/licenses/LICENSE-2.0)\n\nAll other materials are licensed under the Creative Commons Attribution 4.0\nInternational License (CC-BY). You may obtain a copy of the CC-BY license at:\n[https://creativecommons.org/licenses/by/4.0/legalcode](https://creativecommons.org/licenses/by/4.0/legalcode)\n\nUnless required by applicable law or agreed to in writing, all software and materials distributed here under the Apache 2.0 or CC-BY licenses are distributed on an \"AS IS\" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the licenses for the specific language governing permissions and limitations under those licenses.\n\nThis is not an official Google product.", "url": "https://wpnews.pro/news/google-deepmind-open-sources-alphaevolve", "canonical_source": "https://github.com/google-deepmind/alphaevolve", "published_at": "2026-11-04 12:10:20+00:00", "updated_at": "2026-11-04 12:21:01.808336+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-research", "machine-learning", "developer-tools"], "entities": ["Google DeepMind", "AlphaEvolve", "JAX", "Haiku", "NumPy", "SciPy", "Gemini API", "Google Colab"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/google-deepmind-open-sources-alphaevolve", "markdown": "https://wpnews.pro/news/google-deepmind-open-sources-alphaevolve.md", "text": "https://wpnews.pro/news/google-deepmind-open-sources-alphaevolve.txt", "jsonld": "https://wpnews.pro/news/google-deepmind-open-sources-alphaevolve.jsonld"}}