Vast.ai
Vast.ai is a GPU infrastructure marketplace offering programmatic provisioning via CLI, SDK, or REST API across many GPU types and locations with per-second billing, on-demand/interruptible/reserved instance options, and real-time market pricing for cost savings.
The linked official pages describe live, market-driven rates and three pricing tiers (On-Demand, Interruptible, Reserved) but do not list fixed per-instance price tables on the linked pages.Is Vast.ai right for you?
Good for
- Teams that need programmatic, market-priced GPU access to deploy and scale training or inference via CLI, SDK, or API across a wide hardware inventory.
- Programmatic provisioning via CLI, Python SDK, and REST API for automated deployments.
- Large, diverse GPU supply across many locations enabling model-to-hardware matching.
Keep in mind
- Pricing is market-driven and dynamic rather than fixed list prices on the linked official pages.
- The linked official pages do not publish fixed per-instance price tables or a documented SLA percentage.
- Buyers seeking fixed, published per-instance sticker prices or a documented uptime SLA on the linked official pages; pricing is market-driven and dynamic—confirm expected spot pricing for your GPU types.
Pricing and plan limits
Sign-up and API key require adding credit (product page mentions a $5 starting amount). The pricing pages explain per-second billing and reserved terms but show live dynamic rates rather than static plan prices.
More about capabilities and limits
Documented strengths
Programmatic provisioning via CLI, Python SDK, and REST API for automated deployments.
Large, diverse GPU supply across many locations enabling model-to-hardware matching.
Per-second billing and real-time market pricing to minimize wasted billed time.
Limitations to consider
Pricing is market-driven and dynamic rather than fixed list prices on the linked official pages.
The linked official pages do not publish fixed per-instance price tables or a documented SLA percentage.
Choosing a plan
Run representative jobs on your target GPU types and instance classes to capture real per-second market prices, preemption behavior, and end-to-end throughput for budgeting.
Based on the linked product documentation.
Available on
Web.
Official sources
Alternatives to Vast.ai
Choose around the work you need to do.
Lambda
Use Lambda to provision on-demand GPU instances or reserved clusters with high-density NVIDIA hardware and single-tenant options for model training and inference workloads.
