google-deepmind/funsearch resource

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1110 stars · 183 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

FunSearch is the official repository accompanying DeepMind's Nature 2023 paper on discovering new mathematical results via program search with large language models. It provides discovered results (cap sets, admissible sets, bin packing heuristics, cyclic graph independent sets, corner-free sets) plus a single-threaded implementation of the evolutionary search pipeline.

Use cases

• reproduce FunSearch results from the Nature paper
• understand how LLM-driven evolutionary program search works
• adapt FunSearch pipeline to custom language models and sandboxes
• explore discovered cap set and admissible set constructions
• evaluate bin packing heuristics discovered by FunSearch
• learn about mathematical discovery with large language models

When to choose

• you want to study or reproduce the FunSearch method or paper results
• you want to adapt LLM-based program search to your own problems
• you need the discovered combinatorial constructions in numerical format

When to avoid

• you need a production-ready distributed FunSearch system with LLMs and sandbox included
• you want a turnkey library with pip installation and APIs
• you need support or frequent updates

Facets

learning-resource · maturity maintenance

llm-inference agent-framework data-generation benchmarking artificial-intelligence large-language-models mathematics python program-search evolutionary-algorithm combinatorics research-code llm-driven-discovery jupyter-notebooks algorithms research web-server

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google-deepmind/funsearch")