inference/research
verifiedhigh · 0.90updated 2026-10-04

Anyscale ACU pricing - Compute unit rates and instance mapping

database

Fields

id
sources/ppl-cloud-anyscale-acu-detail
Publisher
Anyscale
Kind
database
Accessed
2026-10-04
notes
Fetched 2026-10-04. Anyscale uses ACU (Anyscale Compute Unit) as its billing metric. Published rates: CPU Only $0.0135 ACU/hr; T4 $0.5682 ACU/hr; L4 $0.9542 ACU/hr; A10G $1.3635 ACU/hr; A100 $4.9591 ACU/hr. The ACU is a normalized compute unit - 1 ACU = 1 vCPU + 2GB RAM approximately. For GPU instances, the ACU rate includes the GPU plus its attached CPU/RAM. H/B/GB families are all 'Contact Us'. The $100 free credit is a trial, not a sustained free tier. BYOC model lets you use existing GPU reservations. Key comparison: Anyscale A100 at $4.9591/hr is close to CoreWeave A100 at $2.70/GPU-hr (on-demand $21.60/8) but Anyscale includes the Ray platform orchestration layer. The ACU model means you cannot directly compare to per-GPU-hour rates without knowing the ACU-to-instance mapping.

Notes

Fetched 2026-10-04. Anyscale uses ACU (Anyscale Compute Unit) as its billing metric. Published rates: CPU Only $0.0135 ACU/hr; T4 $0.5682 ACU/hr; L4 $0.9542 ACU/hr; A10G $1.3635 ACU/hr; A100 $4.9591 ACU/hr. The ACU is a normalized compute unit - 1 ACU = 1 vCPU + 2GB RAM approximately. For GPU instances, the ACU rate includes the GPU plus its attached CPU/RAM. H/B/GB families are all 'Contact Us'. The $100 free credit is a trial, not a sustained free tier. BYOC model lets you use existing GPU reservations. Key comparison: Anyscale A100 at $4.9591/hr is close to CoreWeave A100 at $2.70/GPU-hr (on-demand $21.60/8) but Anyscale includes the Ray platform orchestration layer. The ACU model means you cannot directly compare to per-GPU-hour rates without knowing the ACU-to-instance mapping.

Cross-references

This record references no other records.