Home/Large language models/Anyscale

Anyscale pricing

Managed platform that runs Ray workloads (training, batch inference, data processing, serving) on hosted or customer-owned GPUs.

Large language modelsPriced per minuteSelf-serve + sales
Visit AnyscalePrices checked Oct 29, 2026

How much does Anyscale cost?

Usage price from $0.01/minute
Free plan
Not stated
Free trial
$100 in Anyscale credits on signup (one-time credit, not time-limited as stated)
Usage price
$0.01/minute
Enterprise
Contact sales
Read from www.anyscale.com/pricing on Oct 29, 2026. Price changed? Tell usSee the line each price came from

In your favor

  • No long-term contract. Pay-as-you-go is month-to-month; prepaid committed services are invoiced in full on order
  • Free trial. $100 in Anyscale credits on signup (one-time credit, not time-limited as stated)

Watch out for

  • Extra usage is billed. Excess usage is invoiced monthly in arrears

How you buy and get started

Self-serve + salesButtons on their site: Get started with $100 credit, Start for Free, Get a Demo, Talk to Sales
Sign up
Self-serve signup with a free account and $100 credit
"Start building with a free Anyscale account and access to dozens of code templates on Anyscale Platform."
Setup
Hosted option is fully managed with no setup required
"Fully managed infrastructure with no setup required."
When sales steps in
Committed contracts and NVIDIA H/B/GB GPU pricing require contacting sales
"Committed contracts"
Support
Hosted: business hours only, 5 case submissions
"Business hours only"
Support
BYOC: enterprise SLAs with 24x7 coverage, unlimited cases
"Enterprise SLAs with 24x7 coverage"

What Anyscale does

Anyscale is a hosted and bring-your-own-cloud platform built on the open source Ray engine for distributed training, data curation, batch embedding generation, and post-training of AI models. Billing is pay-as-you-go by instance hour in Anyscale credits, with committed contracts available through sales.

Good for
AI and ML teams, especially foundation model builders, that need to scale data-intensive training and inference pipelines across GPU clusters and clouds.
Channels
Web console, Python SDK (Ray), Code templates
Main product
Anyscale Platform (managed Ray)
Also includes
Distributed model training with Ray Train, Multimodal data curation with Ray Data, Batch embedding generation, Post-training with SkyRL and veRL, Hosted and bring-your-own-cloud deployment, Code templates
Connects to
MongoDB, Redis, Databricks, Snowflake, Prefect, Airflow, Dagster, Hugging Face, Lightning, PyTorch, Keras, Hudi
Public API
Not stated
MCP server
Not stated
  • Public API accessNot stated
  • Long context windowNot stated
  • Fine-tuning supportFine-tune an LLM to perform batch inference and online serving for entity recognition.
  • Tool and function callingNot stated
  • Image and audio inputNot stated
  • SOC 2 complianceNot stated
  • Zero data retention optionNot stated
  • VPC or on-prem deploymentDeploy inside your own cloud, or on-prem.

The same 8 checks appear on every tool in Large language models. Each tick quotes the vendor's own site.

Security and your data

QuestionTheir answerWhere it says so
Trains AI on your data?no"Anyscale shall not, nor shall it permit any AI Services to, use Customer Property to train, fine-tune, or improve any artificial intelligence or machine learning model"terms
HIPAA (signs a BAA)Not stated
SOC 2Not stated
GDPRNot stated
CCPAPrivacy notice includes a supplemental CCPA annex"Annex A – Supplemental CCPA Privacy Notice"privacy
PCINot stated
ISO 27001Not stated
Single sign-onSSO, SAML, SCIM supported (plan not stated)"Access controls and authentication including SSO, SAML, SCIM, and audit logs for secure multi-team security and governance."home
Trust centerNot stated

Security answers are the vendor's own claims, quoted. We don't audit them.

Every plan

PlanMonthlyBilled yearlyIncludedOver the allowanceWhere it says so
Pay as you go (Hosted and BYOC), CPU Only/hr $0.01per minute – – – "CPU Only AC 0.0135 /hr"
Pay as you go, NVIDIA T4/hr $0.57per minute – – – "NVIDIA T4 AC 0.5682 /hr"
Pay as you go, NVIDIA L4/hr $0.95per minute – – – "NVIDIA L4 AC 0.9542 /hr"
Pay as you go, NVIDIA A10G/hr $1.36per minute – – – "NVIDIA A10G AC 1.3635 /hr"
Pay as you go, NVIDIA A100/hr $4.96per minute – – – "NVIDIA A100 AC 4.9591 /hr"
NVIDIA H/B/GB GPU families Quote – – – "NVIDIA H/B/GB GPU families [Contact Us](/contact-sales)"
Committed contracts Quote – – – "Unlock volume discounts and other benefits that scales with your usage."

Fine print

Billing increment
Not stated
Billed beyond usage
Excess usage is invoiced monthly in arrears
"all excess usage will be invoiced monthly in arrears"
Contract
Pay-as-you-go is month-to-month; prepaid committed services are invoiced in full on order
"Fees for all prepaid committed Anyscale Services will be invoiced in full upon execution of the applicable Order Form"
Refunds
Fees are non-refundable except as expressly provided
"All payment of Fees are non-refundable and non-transferable except as expressly provided in this Agreement."
Auto renew
Not stated
Price change notice
Not stated

"Not stated" means we read their pricing page and terms and they don't say.

About the company

Legal name
Anyscale, Inc.
Address
600 Harrison Street, 4th Floor, San Francisco, CA 94107
Online since
Jan 31, 2000
Domain registration date
Funding
Anyscale to join Nscale, which acquires Anyscale
"Anyscale to Join Nscale"
Customers
"500M+ all-time Ray downloads, 41K+ GitHub stars, 1.2k+ contributors"
their claim

Anyscale alternatives

Reviews elsewhere

We link to reviews instead of copying ratings: those sites don't allow reuse, and we don't test products ourselves.

Questions

Is Anyscale free?

There's no free plan. Free trial: $100 in Anyscale credits on signup (one-time credit, not time-limited as stated).

Does Anyscale train AI on your data?

No, according to Anyscale: "Anyscale shall not, nor shall it permit any AI Services to, use Customer Property to train, fine-tune, or improve any artificial intelligenc"