GPU Cost Calculator
Compare RunPod, Modal, Replicate, and Vast.ai across hardware × hours. Same workload, four price tags.
Estimate monthly spend
INTERACTIVE · LIVE · VERIFIED RATESDrag to your expected GPU hours. 720 is 24/7. Most indie workloads land between 40 and 200 hours a month. All four provider numbers update live.
Active GPU rate × hours. Storage, bandwidth, and idle time not included.
Rates verified from each provider's pricing page in 2026-04. RunPod uses Serverless active rates. Modal lists per-second pricing — we've normalized to per-hour. Modal doesn't sell RTX 4090s; we show A10G as a rough proxy on the A40 row. Replicate doesn't publish a dedicated 4090 SKU either, so 4090 reads "—" on that provider. Vast.ai is a peer-to-peer marketplace, so the actual listings on any given day vary a lot — the figure shown is a typical median. Storage, bandwidth, and idle workers are excluded across the board.
Read the underlying reviews
FOUR PROVIDERS · HONEST TAKESRunPod
Indie GPU rental sweet spot. Real per-second billing, rough community-tier edges.
Modal
Polished Python-native serverless. Higher floor, smoother cold starts.
Replicate
One-click APIs over open-source models. Fastest path; pricey at volume.
Vast.ai
Peer-to-peer marketplace. Cheapest rates if you can tolerate the rough edges.