Ray

license

Apache-2.0(OSI)

source

public(all Ray libraries OSS)

commercial-tier

Anyscale managed SaaS on top

core-gated

ungated(a full 11,885-entry tree carries one Apache-2.0 LICENSE and no enterprise/ or ee/ path, no license-key machinery, and Anyscale sells managed infrastructure rather than a Ray feature the open project lacks)

Apache-2.0 throughout with a managed platform sold beside it, and nothing in the core held back. That reading rests on the repo and on Anyscale’s own page. The 11,885-entry untruncated tree of ray-project/ray@master carries a single governing Apache-2.0 LICENSE plus three vendored third-party ones, and no enterprise/, ee/, commercial/ or proprietary/ path; the only Anyscale-named files are CI image definitions and release-test harnesses, not a closed package the open tree depends on. Anyscale’s Ray page sells infrastructure management - "Ray runs anywhere. Or choose Anyscale, and leave the infrastructure to us" - and claims no Ray capability the open project withholds. So the published source is the whole product, and it scores at the top of the ladder. Ray is also governed by the PyTorch Foundation rather than by Anyscale, which is why the managed platform can be moved out of the openness question entirely.

• https://github.com/ray-project/ray recorded 2026-08-15

  Repository page for ray-project/ray: 'Apache-2.0 license' in the sidebar, 43.5k stars, 7.9k forks, public and unarchived. Establishes the licence and that the source is public.

• https://www.anyscale.com/product/open-source/ray recorded 2026-08-12

  Anyscale positions its offering as managed infrastructure, not a fuller Ray: "Scale on any cloud or private data centers, bare metal, Kubernetes cluster - Ray runs anywhere. Or choose Anyscale, and leave the infrastructure to us." The page makes no claim that open-source Ray lacks a feature the paid platform adds.

• https://www.anyscale.com/blog/ray-by-anyscale-joins-pytorch-foundation recorded 2026-08-15

  Anyscale's announcement that Ray moved under the PyTorch Foundation. Establishes the governance reading — the managed platform is sold by a company that does not control the project — rather than any licence or source dimension.

• https://api.github.com/repos/ray-project/ray/git/trees/master?recursive=1 recorded 2026-08-12

  Full untruncated recursive tree of the default branch master, 11,885 entries. The only license paths are the root LICENSE, python/LICENSE.txt and three vendored third-party licenses (dacite, pathspec, speedscope). No enterprise/, ee/, commercial/ or proprietary/ path. The only anyscale-named paths are CI and release-test tooling (ci/docker/ray-anyscale-*.wanda.yaml, release/ray_release/anyscale_job_runner.py); no closed package the open tree depends on.

• https://api.github.com/repos/ray-project/ray recorded 2026-08-12

  Repo metadata: language Python, license spdx_id Apache-2.0, 43,497 stars, default branch master, size 779,592 KB.

The PyPI download API reports 63,273,910 downloads of ray in the trailing 30 days and 15,105,708 in the trailing week, comfortably above the >10M a month floor for level 5 - and above even the stricter 50M floor applied to the highest-volume software records, where PyTorch sits at ~88.6M and Triton at ~64.7M. The trailing-week figure alone clears the floor, so the band does not turn on which window is read. The figure is read from the download API rather than from the rendered pepy page, and PyPI is the product’s declared distribution channel, so the level can be recomputed from that feed and challenged. Ray is the standard open distributed-compute and model-serving substrate (Ray Serve, Tune, Train, Data) under the PyTorch Foundation.

• https://pepy.tech/projects/ray recorded 2026-06-25

  61,306,813 downloads in last 30 days

• https://pypistats.org/api/packages/ray/recent recorded 2026-08-13

  {"data":{"last_day":2156063,"last_month":63273910,"last_week":15105708},"package":"ray"}. The declared PyPI artifact, read directly - 63,273,910 downloads in the trailing month, above the >10M level-5 floor for software usage_volume.

• https://www.anyscale.com/blog/ray-by-anyscale-joins-pytorch-foundation recorded 2026-06-04

  Ray under the PyTorch Foundation (governance / ecosystem standing)

General distributed-compute + model-serving framework (Ray Core, Serve, Tune, Train, Data). No standardized inference/training benchmark (MLPerf) applies to Ray as a whole; scored on breadth of distributed-serving feature coverage. Not directly comparable to single-engine inference benchmarks. It is a scheduling and serving substrate rather than an isolation boundary, which is why it sits at 4 and not at the microVM frontier this category's 5s occupy: Ray Core plus the AI libraries (Data, Train, Tune, Serve, RLlib), clusters on VMs or Kubernetes, mixed CPU/GPU scheduling, and workloads spanning batch inference, model serving, distributed training, hyperparameter tuning, feature engineering and reinforcement learning.

• https://www.anyscale.com/product/open-source/ray recorded 2026-08-15

  Anyscale's open-source Ray page, describing Ray Core plus the Data, Train, Tune and Serve libraries and deployment on VMs or Kubernetes. Establishes the feature breadth the capability band rests on.

• https://docs.anyscale.com/get-started/what-is-ray recorded 2026-08-13

  "Ray is an open source framework for building, deploying, and serving AI and ML applications using distributed compute resources. Ray excels when applications need to scale to many machines and optimizes resource utilization for workloads that require both CPUs and GPUs." Lists the libraries built on Ray Core (Data, Train, Tune, Serve) and the common workloads - batch inference, LLMs and gen AI, model serving, feature engineering, hyperparameter tuning, distributed training, reinforcement learning. "All Ray apps deploy as Ray clusters", on virtual machines or Kubernetes.