Vast.ai vs Runpod: cost per hour, side by side
Per GPU, per hour, on-demand, in US dollars. Runpod sets a fixed price list in two tiers (Community Cloud and Secure Cloud). Vast.ai is a marketplace where each host sets its own price, so we show the lowest offer and the median of the offers we read.
| GPU | VRAM | Runpod Community | Runpod Secure | Vast.ai lowest | Vast.ai median | Vast.ai offers read |
|---|---|---|---|---|---|---|
| RTX 3090 | 24 GB | $0.22 | $0.50 | $0.16 | $0.24 | 64 |
| RTX 4090 | 24 GB | $0.34 | $0.89 | $0.33 | $0.48 | 64 |
| RTX 5090 | 32 GB | $0.69 | $1.19 | $0.47 | $0.74 | 64 |
| A100 PCIe | 80 GB | $1.19 | $1.79 | $0.87 | $0.93 | 4 |
| A100 SXM | 80 GB | $1.39 | $1.79 | $1.07 | $1.14 | 8 |
| H100 PCIe | 80 GB | $1.99 | $2.89 | $2.85 | $3.18 | 4 |
| H100 SXM | 80 GB | $2.69 | $3.99 | $2.14 | $2.92 | 2 |
Short answer: in our November 1, 2026 sample, the cheapest Vast.ai offer beat Runpod Community Cloud on 6 of these 7 GPUs (all but the H100 PCIe). But the typical (median) Vast.ai offer was cheaper only for the two A100 rows; for the RTX 3090, 4090, 5090 and both H100s, Runpod Community Cloud's fixed price was lower than the Vast.ai median. So Vast.ai wins if you shop for the lowest host; Runpod wins if you want a predictable price without comparing hosts.
What a 10-hour job costs
Same prices as the table, times 10 hours, GPU time only (storage and data transfer are extra):
| GPU | Runpod Community | Runpod Secure | Vast.ai lowest | Vast.ai median |
|---|---|---|---|---|
| RTX 3090 | $2.20 | $5.00 | $1.60 | $2.40 |
| RTX 4090 | $3.40 | $8.90 | $3.30 | $4.80 |
| RTX 5090 | $6.90 | $11.90 | $4.70 | $7.40 |
| A100 SXM 80 GB | $13.90 | $17.90 | $10.70 | $11.40 |
| H100 SXM 80 GB | $26.90 | $39.90 | $21.40 | $29.20 |
Sources and dates
- Runpod: on-demand Pods prices from runpod.io/pricing, read November 1, 2026 and re-checked the evening of November 1, 2026 (no change; the page is marked "Updated October 28, 2026").
- Vast.ai: live offers from Vast.ai's public marketplace search (console.vast.ai/api/v0/bundles), read November 1, 2026 at about 9:12 AM PT. Only rentable, verified, on-demand offers. We read up to the 64 cheapest offers per GPU, so for the RTX 3090, 4090 and 5090 the median is the median of those 64, not of the whole market. Per-GPU price is the listed total hourly rate divided by the number of GPUs. A100 and H100 rows are 80 GB cards only.
- Vast.ai prices change by the hour, and Runpod stock varies by region. Check the price on the provider's page before you rent.
- Download this table as JSON.
How to read it
- Cheapest is not always best. Vast.ai hosts vary in reliability, disk speed and bandwidth price. Runpod Secure Cloud runs in data centers; Community Cloud runs on third-party hosts.
- Storage and data transfer are extra on both. The table only shows the GPU's hourly rate.
- Need more than one GPU? Multiply by the GPU count, or use the cheapest setup finder to see which 1 to 8 GPU setup holds your model.
How Vast.ai and Runpod differ
| Vast.ai | Runpod | |
|---|---|---|
| Who sets the price | Each host sets its own; prices move with supply and demand | Runpod publishes one price list |
| Paying | Add credits before you start; billed by the second while running | Prepaid credit; the pricing page lists per-hour and per-second rates |
| Cheaper tier | Interruptible instances, "often 50%+ cheaper" but can be paused | Community Cloud (third-party hosts) vs Secure Cloud (data centers) |
| Long-term discount | Reserved instances, up to 50% off with pre-payment | Reserved clusters and contracts through Runpod sales |
| Storage | Each host sets the price; billed while the instance exists, even when stopped | Container disk $0.10/GB/month; volume disk $0.10 running, $0.20 idle; network storage $0.07/GB/month under 1 TB, $0.05 over 1 TB |
| Data transfer | Each host sets upload and download prices; check before moving large models | See Runpod's pricing page |
| Serverless | Vast Serverless: no extra fee, you pay the underlying instance | Separate serverless price list (for example $0.75/hr for 24 GB L4, A5000 or 3090 workers) |
Sources: Vast.ai pricing docs and Runpod pricing, both read November 1, 2026.
Watch the storage bill. On both, storage keeps costing money after you stop the GPU. A 100 GB Runpod volume costs about $20 a month while idle ($0.20 × 100 GB). On Vast.ai, delete the instance, not just stop it, to end storage charges.
Which one suits you
- Pick Vast.ai if you want the lowest hourly rate and don't mind checking each host's reliability, disk and bandwidth prices first. It is strongest on consumer cards and A100s.
- Pick Runpod if you want one stable price, data-center hosting (Secure Cloud), templates or serverless endpoints, and don't want to compare hosts.
- For a short test either works. Size the model first with the VRAM calculator or the GPU VRAM lookup so you don't rent more GPU than you need.
What about Novita?
Novita no longer rents GPU instances (Novita told us by email on October 30, 2026), so there is no Novita hourly GPU price. If you only need inference, Novita's model APIs Affiliate partner bill per token instead.
Rent a GPU
- Rent on Vast.ai Affiliate partner
- Rent on Runpod Referral
Need to know which card fits your model first? Use the VRAM calculator or what can I run.
Some links on this page are affiliate or referral links: we may get a commission or account credit if you sign up. The prices above are the same either way. See our disclosure.