Anyscale iconAnyscale

Managed Ray platform with the RayTurbo runtime for distributed training, batch inference and model serving, hosted or in your own cloud

44.2k+ GitHub Stars (Ray)
$100 Free Credits
Nscale Pending Acquirer

Overview

Anyscale is the company behind Ray, the open-source framework for scaling Python and AI workloads across many machines. Its product today is a managed Ray platform built on RayTurbo, Anyscale's optimized Ray runtime, for distributed training, batch inference, data processing and model serving with Ray Serve. You can run it hosted by Anyscale or bring your own cloud (your VPC, Kubernetes or on-premises), and KubeRay users get their own credit offer. Anyscale no longer sells a hosted per-token LLM API: Anyscale Endpoints was retired in 2024, and today you rent compute and serve your own models on it. Two ownership changes matter. Ray moved to the PyTorch Foundation under the Linux Foundation in October 2025, so the open-source project is community governed. And on July 30, 2026, London AI cloud Nscale agreed to acquire Anyscale; reports put the price at about $1.65B, which Nscale has not confirmed. As of November 2026 the deal is still described as pending, expected to close in the second half of 2026, with Anyscale keeping its brand, platform and customers such as Coinbase and Runway. For the open-source framework itself, see the Ray page. Alternatives include Modal and Baseten.

The Verdict

Who Should Use Anyscale?

Best For

  • Teams already using Ray who want managed clusters
  • Large batch inference and distributed training jobs
  • Companies that must keep compute in their own cloud account

Not Ideal For

  • Developers who just want a per-token LLM API
  • Beginners new to distributed computing

What's Great

  • Saves DevOps time managing Ray clusters
  • Scale workloads without major code changes
  • Bring-your-own-cloud option with 24x7 SLA support

Watch Out For

  • Docs are not beginner friendly; debugging takes effort
  • No hosted per-token model API
  • Ownership change to Nscale is pending

Pricing

View all features & details

Key Features

  • Managed Ray clusters with autoscaling
  • RayTurbo optimized Ray runtime
  • Ray Serve for model deployment
  • Workspaces and jobs
  • Hosted or bring your own cloud, Kubernetes or on-prem
  • Observability and debugging tools
  • KubeRay support with its own credit offer

Company

  • Created Ray, now a PyTorch Foundation project (Oct 2025)
  • Nscale agreed to acquire Anyscale (Jul 30, 2026)
  • About 200 staff to join Nscale on close

How It Compares

Feature Anyscale Modal Baseten
Core model Managed Ray clusters Serverless Python functions Model serving platform
Best for Distributed training and batch jobs Bursty Python and GPU jobs Production inference

Sources: Anyscale · Pricing (fetched Nov 2026) · Docs · Unite.AI (Jul 2026) · Futurum (2026)

Is this your tool? Add a badge