Cheapest on-demand GPU for inference
Depot Pick: Vast.ai. Runner-up: RunPod (then Oracle Cloud GPU). Job: Cheapest on-demand GPUs. Updated 2026-11-03 (newest stored date on this page).
Minimize $/GPU-hr for fault-tolerant training/batch.
Ready call for agents
MCP https://getagentdepot.com/api/mcp → decide {"task":"cheapest on-demand GPU for inference"}
HTTP GET /api/decide?jtbd=cheap-gpuNo key. The reply carries the pick, runner-up, the pick's sourced price with its date, and a recheck token.
Depot Pick: Vast.ai
Editorial score 6.94 / 10 (curated estimate, as of 2026-10-28). Prices below are copied from the provider's own page; units are never converted.
pricePerf primary; reliability heavily discounted for marketplaces.
No sourced price stored (null). Depot never estimates a price.
Runner-up 1: RunPod
Editorial score 8.32 / 10 (curated estimate, as of 2026-10-28). Prices below are copied from the provider's own page; units are never converted.
| Item | Price (USD) | Source | Price as of | Last re-checked |
|---|---|---|---|---|
| H100 SXM Pod (Secure Cloud) | $3.99 / gpu hour | official page | 2026-11-03 | not reconfirmed |
| H100 SXM Pod (Community Cloud) | $2.69 / gpu hour | official page | 2026-11-03 | not reconfirmed |
| H200 Pod (Secure Cloud) | $5.29 / gpu hour | official page | 2026-11-03 | not reconfirmed |
| H200 Pod (Community Cloud) | $3.59 / gpu hour | official page | 2026-11-03 | not reconfirmed |
Runner-up 2: Oracle Cloud GPU
Editorial score 7.65 / 10 (curated estimate, as of 2026-10-28). Prices below are copied from the provider's own page; units are never converted.
| Item | Price (USD) | Source | Price as of | Last re-checked |
|---|---|---|---|---|
| BM.GPU.H100.8 | $10 / gpu hour | official page | 2026-11-03 | not reconfirmed |
| BM.GPU.H200.8 | $10 / gpu hour | official page | 2026-11-03 | not reconfirmed |
| BM.GPU.A100-v2.8 (A100 80GB) | $4 / gpu hour | official page | 2026-11-03 | not reconfirmed |
| BM.GPU.L40S-NC.4 | $3.5 / gpu hour | official page | 2026-11-03 | not reconfirmed |
Lowest sourced prices per gpu hour
Same unit only, price only (not quality-adjusted). The pick is editorial and is not chosen by price.
| Provider | Item | USD / gpu hour | Source | Price as of | Last re-checked |
|---|---|---|---|---|---|
| SaladCloud | RTX 3090 24GB (Lowest priority, batch) | $0.09 | official page | 2026-11-03 | not reconfirmed |
| RunPod | RTX 4090 Pod (Community Cloud) | $0.34 | official page | 2026-11-03 | not reconfirmed |
| Massed Compute | L40S (48GB) x1 | $0.97 | official page | 2026-11-03 | not reconfirmed |
| Verda | 1x L40S 48GB | $1.882 | official page | 2026-10-29 | not reconfirmed |
| Oracle Cloud GPU | VM.GPU.A10.1 | $2 | official page | 2026-11-03 | not reconfirmed |
Full list: /cheapest/cheap-gpu
Questions
- What does Agent Depot pick for Cheapest on-demand GPUs?
- Vast.ai is the stored Depot Pick for the job cheap-gpu. pricePerf primary; reliability heavily discounted for marketplaces.
- What is the runner-up?
- RunPod, then Oracle Cloud GPU. The ranking is editorial (scores as of 2026-10-28) and is not chosen by price.
- When should I not use Vast.ai?
- Need SLA/security — RunPod Secure Cloud, Lambda, or CoreWeave. Never Vast for sensitive PII without review.
- How much does Vast.ai cost?
- Depot has no sourced price stored for Vast.ai on this job, so none is shown. It never estimates one.
Picks are editorial and dated; community reports never change a pick or a score. Method: /methodology · Corrections: /corrections · Compare: Vast.ai vs RunPod