vast ai pricing

Vast.ai operates as a peer-to-peer GPU rental marketplace rather than a traditional subscription-based product. Pricing is dynamic and set by individual hosts based on supply and demand, with no monthly or yearly subscription fees. Users prepay credits (starting with a $5 minimum) and are charged per second of usage across GPU compute, storage, and bandwidth.

Key pricing models:

• On-Demand: Guaranteed availability, standard rates.
• Interruptible: Significantly cheaper (often 50%+ lower), but instances may be preempted.
• Reserved: Discounts for 1, 3, or 6-month commitments with guaranteed capacity.

Example live rates (per GPU/hour, approximate as of recent data):

• RTX 4090: ~$0.28/hr (range $0.11–$1.33/hr)
• RTX 5090: ~$0.37/hr (range $0.13–$1.33/hr)
• H100 SXM: ~$1.53/hr (range $0.80–$6.67/hr)
• H200: ~$2.32/hr

Additional costs may apply for storage ($/GB/hr) and data transfer ($/TB). There is no fixed tiered subscription; costs scale directly with usage. Visit https://vast.ai/pricing for real-time marketplace rates.