Ray

What it does

Ray is an open-source Python framework for scaling AI and Python applications from a laptop to a cluster. You give it Python code and data, including structured data and unstructured data such as images, video, and audio, and it runs the work in parallel across CPUs and GPUs. Its libraries cover data processing, distributed model training, hyperparameter tuning, reinforcement learning, and model serving.

Teams typically use Ray to parallelize existing Python code with few or no changes, or to run distributed training and LLM inference across multiple nodes. Its distinguishing feature is a unified compute layer: the same code can move from a laptop to a cluster on AWS, GCP, Azure, or Kubernetes, with tasks, actors, and fault tolerance handled by the framework.

Ray suits ML engineers, data scientists, and developers who write Python and need to scale workloads without deep distributed-systems expertise. Installation is via pip (pip install ray). Anyscale offers a fully managed platform for Ray; pricing for that service is not covered here and should be confirmed directly with Anyscale. Users need Python and enough compute to run their workloads, and Ray is Python-native, so it is a poor fit for teams whose workloads are not written in Python.

What it can do

• Distributed training of ML and deep learning models using frameworks such as PyTorch and TensorFlow
• Batch inference using CPUs and GPUs in the same pipeline
• Model serving with independent scaling and fractional resource allocation
• Parallel hyperparameter tuning
• Reinforcement learning via the RLlib library
• Scaling general Python code with tasks, actors, and objects

Works with

• Kubernetes
• YARN
• Slurm
• AWS
• GCP
• Azure
• PyTorch
• TensorFlow
• Keras
• XGBoost

Example workflows

Batch inference over a dataset

Input: a CSV or Parquet dataset in cloud storage. Steps: read it with Ray Data, transform batches in parallel with map_batches, then write results to Parquet. Output: transformed files written to local or cloud storage.

Distributed model training

Input: a single-worker PyTorch training function. Steps: wrap the model and data loaders with Ray Train utilities and run with TorchTrainer across the requested number of workers. Output: trained model results returned from trainer.fit().

Before you choose

• Ray is Python-native; it is not a fit for teams whose workloads are not written in Python.
• Pricing for Anyscale's managed platform should be confirmed directly with Anyscale.
• Scaling to GPU clusters requires GPU infrastructure; the docs point to Anyscale workspaces for GPU environments without a local GPU.

Listed under

• Coding
• Sandboxes and infrastructure
• AI infrastructure
• Developers
• Distributed computing
• LLM inference
• ML engineers
• ML training

Get started

Install with pip install ray; the Getting Started guide covers the Ray Libraries, Core, and Clusters paths.

1. Install Ray with pip install ray; add extras such as ray[data], ray[train], ray[serve], ray[tune], or ray[rllib] for specific libraries.
2. Follow the Getting Started guide to choose a path: Ray Libraries, Ray Core, Ray Clusters, or debugging and monitoring.
3. To run on a cluster, deploy Ray on AWS, GCP, Azure, or Kubernetes using the Ray Clusters documentation.

The software can be installed under its repository license. Infrastructure and model-provider charges remain separate. Optional commercial hosting and support have their own terms.

Ray is free to use. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Scale Machine Learning & AI Computing | Ray by Anyscale ↗www.ray.io
• Overview — Ray 2.59.0 ↗docs.ray.io
• Getting Started — Ray 2.59.0 ↗docs.ray.io
• GitHub - ray-project/ray: Ray is an AI compute engine. Ray consists of a core distributed runtime and a set of AI Libraries for accelerating ML workloads. · GitHub ↗github.com
• Welcome to Ray! — Ray 2.59.0 ↗docs.ray.io

Company website

ray.io: product, pricing and policy pages

YC directory mirror

Batch and company name

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.