Anyscale · Pricing Plans

Anyscale Plans Pricing

Anyscale uses a usage-based pay-as-you-go model billed in compute instance hours, with rates that vary by GPU type. Two delivery modes are offered - Hosted (monthly credit-card invoicing, business-hours support) and Bring Your Own Cloud (BYOC) (invoiced via cloud marketplaces, 24/7 support, can consume customer GPU reservations). New users receive $100 in free credits; committed contracts unlock volume discounts.

3 Plans API Commons Plans
View Source
AIDistributed ComputingRayML PlatformInferenceGPUPlans

Plans

Hosted (Pay-as-you-go) usage

Anyscale-managed compute on AWS / GCP, billed per instance-hour by node type with monthly credit-card invoicing and business-hours support.

CPU Only (per hour) (hours · month) $0.0135 per hour USD
NVIDIA T4 (per hour) (hours · month) $0.5682 per hour USD
NVIDIA L4 (per hour) (hours · month) $0.9542 per hour USD
NVIDIA A10G (per hour) (hours · month) $1.3635 per hour USD
NVIDIA A100 (per hour) (hours · month) $4.9591 per hour USD
NVIDIA H100 (per hour) (hours · month) $9.2880 per hour USD
NVIDIA H200 (per hour) (hours · month) $10.6812 per hour USD
Bring Your Own Cloud (BYOC) enterprise

Anyscale runs the control plane while compute lives in the customer's AWS / GCP account, leveraging existing GPU reservations. Invoiced via cloud marketplaces with 24/7 support.

Anyscale Platform Fee (hours · month) per instance-hour, contract-dependent (contact sales) USD
Underlying Cloud Compute (hours · month) billed by AWS / GCP at customer's rates USD
Enterprise / Committed enterprise

Volume commitments unlock discounts on hosted hourly rates and BYOC platform fees. Contact Anyscale sales.

Committed Spend (contract · year) contact sales (volume discount) USD

Sources