Directory › GPU & inference hosting › Vast.ai
Vast.ai
Cheapest GPU hours if you can tolerate interruption.
GPU & inference hosting Price verified 2026-10-16
Visit provider ↗ Direct link, no affiliate relationship. Disclosure
| Starting price | Live 'from' rates, e.g. RTX 5090 from $0.39/hr; H100 SXM from $1.73/hr |
|---|---|
| Pricing model | Usage-based |
| Key specs | Marketplace across 40+ data centers with live prices set by supply and demand. Snapshot on our check date: RTX 5090 from $0.39/hr (median $0.63), RTX 3090 Ti from $0.23/hr, H100 SXM from $1.73/hr (median $2.40); older cards (e.g. RTX A2000 6 GB, Tesla P100) from $0.03/hr. Tiers: On-Demand (guaranteed, no interruptions), Interruptible (50%+ cheaper, may be reclaimed), Reserved. Per-second billing. |
| Date checked | 2026-10-16 |
| Sources | vast.ai/pricing |
Agent-friendly features
- AI Agents and text-generation use-case pages
- Serverless and clusters
Security notes
- Marketplace model: host quality varies by listing.
Security notes are what the provider states on the pages we read, or gaps we noticed. Treat vendor claims as claims, not audits.
Not verified or not checked
- Prices move constantly; the figures are a snapshot from 2026-10-16.
- Host security varies by listing; not assessed.
Tags
Similar providers
Spot something out of date? Send a correction with a link to the provider page. Providers can fix factual errors for free.