Vast.ai review

Facts from Vast.ai’s own pricing and docs pages, September 202650 read from a quoted sentence on the page we archived.

Summary

Vast.ai rents GPUs from $0.10 per GPU-hour on demand, bills per second. The minimum to start is $5 minimum to start. Vast.ai does not publish an uptime SLA.

SOC 2 Type IISpend limitsRefundable balanceVolume discounts
From
$0.10/GPU-hr
Billing
Per second
Minimum commitment
$5 minimum to start
Egress
Priced per byte by each host
Uptime SLA
not stated
Registered in
🇺🇸
Visit Vast.ai
On-demand pricing
×2, on demand · RTX PRO 6000 S
$1.34/hr
Deploy
×4, on demand · H200
$4.61/hr
Deploy
×1, on demand · B200
$6.25/hr
Deploy
×1, on demand · A100 SXM4
$0.75/hr
Deploy
×1, on demand · RTX 4090
$0.43/hr
Deploy

See all 53 Vast.ai offers →

What’s good
Lowest published on-demand rate is $0.10/GPU-hr (NVIDIA RTX 5060 8GB)
Billing is per second
Self-serve signup — no sales call to get started
Commitment discounts: reserved: up to 50% off for 1/3/6-month terms
Spend controls: low-balance email alerts
SOC 2 Type II (Secure Cloud)
Plans
On-DemandInterruptibleReserved
FromMarket-set; H100 from $0.90/hrOften 50%+ below on-demandUp to 50% off on-demand
Minimum termNone; $5 to get startedNone1, 3 or 6 months
BillingPer secondPer secondPrepaid for the term
Uptime SLAGuaranteed uptime, no interruptionsNone — preemptibleGuaranteed capacity
TenancyIsolated instance on a host's machineSame machines; may be paused when outbidGuaranteed capacity
Provisioning~60 seconds~60 secondsConvert a running on-demand instance
PaymentPrepaid creditsPrepaid creditsPrepaid credits
Details
Billing & access
How you pay
Prepaid credits; optional auto top-up
Currency
USD
Commitment discounts
Reserved: up to 50% off for 1/3/6-month terms
Approval or KYC
Card and email verification
When funds run out
Instances stopped; deletion after a spend-based grace period
more
Stopped instances keep accruing storage charges; with no card on file, instances and data are destroyed.
Late payment
Instances stopped at $0; card charged for negative balance if saved
Provisioning guarantee
Reserved only
What you get
GPUs listed
68+ GPU types, RTX 3060 to B200
more
Architectures listed: Blackwell (RTX 5090), Hopper (H100 NVL/SXM/PCIe, H200), Ada (RTX 4090, L-series, RTX Ada), Ampere (RTX 3090, RTX A-series, A100), Volta/Tesla, Turing, Pascal, Maxwell.
Regions
40+ data centres worldwide
Node configuration
Set by each host; shown per offer
Local storage
Per-second storage charge, even while stopped
Persistent storage
Storage volumes
Ingress
Priced per byte by each host
Public endpoints
Direct SSH; ports per instance
Private networking
Private VPN access (enterprise)
Images
Official and community templates; bring your own
Access methods
Console, SSH, CLI, API, Python SDK
Model deployment
Serverless endpoints; model library
Managed inference
Serverless, at the underlying instance cost
Serverless
Yes, no separate pricing tier
FAQ
How much does Vast.ai cost?

As of September 2026, the lowest published on-demand rate is $0.10 per GPU-hour for the NVIDIA RTX 5060 8GB. Vast.ai publishes an on-demand rate for 17 GPU models. Billing is per second.

Does Vast.ai have an uptime SLA?

Vast.ai does not publish an uptime commitment on its own site.

Is there a minimum commitment on Vast.ai?

Vast.ai states its minimum as $5 minimum to start.

Does Vast.ai charge for egress?

Vast.ai states its egress terms as priced per byte by each host.

Can I sign up to Vast.ai without approval?

Vast.ai states that signup is self-serve. On approval and identity checks it states: card and email verification.

Are Vast.ai payments refundable?

Vast.ai states: spent credits final; unspent balance refundable at discretion.

Where are Vast.ai's data centres?

Vast.ai lists 40+ data centres worldwide.

Who do you contract with at Vast.ai?

The contracting entity is Vast.ai Inc. (Delaware).

Reviewed September 2026, from Vast.ai’s own site. 50 facts, each quoted from the page it was read on.