Skip to main content
Back

Google's Massive Open Research Monorepo Just Hit 38,000 Stars

Google's sprawling research monorepo has quietly grown to 38k+ stars and thousands of project subfolders spanning nearly every corner of ML.

Repo, Full Access
  • google-research/google-research is Google's public monorepo, now past 38k stars and 8.4k forks.
  • Hundreds of paper-accompanying projects including ALBERT, AutoML-Zero, ZipNeRF, Charformer, CoDi, and BBF.
  • Language mix is roughly 48% Jupyter Notebook and 45% Python with 5,000+ commits on master.
  • Datasets released under CC BY 4.0, code under Apache 2.0, with over 1,000 open issues.
  • Maintainers recommend downloading single subdirectories via github.dev instead of cloning the full repo.
  • Larger active projects like TimesFM and Kauldron live in separate repos under the same org.

If you have ever chased down a Google paper and wondered where the reference code went, there is a decent chance it landed in one enormous shared repository. google-research/google-research is the catch-all monorepo where research teams across Google drop the code that accompanies their publications, and it has now crossed 38,000 stars with more than 8,400 forks.

The repo is not a single project. It is closer to a public shelf where hundreds of independent research folders sit side by side, each with its own README, dependencies, and reproduction scripts. It contains code released by Google Research, with datasets released under CC BY 4.0 and source files released under Apache 2.0.

What actually lives inside

Browsing the top-level directory listing feels like scrolling a conference proceedings. A quick sample of the folders currently checked in:

  • albert - the lite BERT variant that once held SOTA on GLUE and SQuAD
  • automl_zero - evolving ML algorithms from scratch using only basic math ops
  • camp_zipnerf - anti-aliased neural radiance fields for 3D scene reconstruction
  • autoregressive_diffusion