Baseten vs Vast.ai
- GPUs
- 6 80
- Cheapest GPU
- $0.63/hr $0.0296/hr
- Price wins
- 0/4 4/4
Vast.ai is cheaper on all 4 GPUs both rent. Vast.ai carries 74 more GPUs.
Runs dedicated model inference billed per minute of GPU time, plus pre-optimised model APIs billed per token.
Best for: Production inference for custom models
Marketplace where independent hosts list GPUs and renters pick machines by price, with on-demand and interruptible bids.
Best for: Low-cost GPUs from independent hosts
Price by GPU
The 4 GPUs both rent. Cheapest live rate per GPU-hour, most powerful first. The cheaper side is in purple.
| GPU | Baseten | Vast.ai | Gap |
|---|---|---|---|
| NVIDIA B200 SXM 180GB | $9.98/hr | $6.25/hr community | 60% |
| NVIDIA H100 SXM5 80GB | $6.50/hr | $1.88/hr community | 245% |
| NVIDIA A100 SXM4 80GB | $4.00/hr | $0.95/hr community | 322% |
| NVIDIA L4 24GB | $0.85/hr | $0.27/hr community | 216% |
Prices last checked: Baseten 1 Oct 2026, Vast.ai 1 Oct 2026.
Only on one side
Only on Baseten 2
Only on Vast.ai 76
- NVIDIA B300 SXM 262GB $9.94
- NVIDIA H200 SXM 141GB $3.56
- NVIDIA H100 NVL 94GB $3.02
- NVIDIA H200 NVL 141GB $3.91
- NVIDIA H100 PCIe 80GB $2.33
- NVIDIA RTX PRO 6000 Blackwell Workstation Edition $1.10
- NVIDIA RTX PRO 6000 Blackwell Server Edition $1.01
- NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition $1.10
- NVIDIA GeForce RTX 5090 32GB $0.43
- NVIDIA RTX 6000 Ada 48GB $0.49
- NVIDIA L40S 48GB $0.73
- NVIDIA A100 PCIe 40GB $0.40
- NVIDIA A100 PCIe 80GB $1.00
- NVIDIA RTX 5880 Ada 48GB $0.67
- NVIDIA RTX 5000 Ada 32GB $0.34
- NVIDIA RTX PRO 5000 Blackwell 48GB $0.61
- NVIDIA RTX PRO 5000 Blackwell 72GB $0.94
- NVIDIA GeForce RTX 5080 16GB $0.20
- +58 more on the Vast.ai page
Company facts
| Fact | Baseten | Vast.ai |
|---|---|---|
| Headquarters |  San Francisco, CA | United States |
| Type | GPU cloud | Marketplace |
| Founded | n/a | 2018 |
| GPUs priced | 6 | 80 |
| Fastest GPU | NVIDIA B200 SXM 180GB | NVIDIA B200 SXM 180GB |
| Services | Dedicated model inference, pre-optimised model APIs, training (Loops SDK), self-hosted deployment | n/a |
| Pricing models | Per-minute GPU compute, per-token for model APIs | n/a |
Compliance
Frameworks each provider states it holds, linked to its own statement. Not published means we found no statement, not that the provider fails it. Check the scope with the provider.
Pick your provider
Questions
- What is the difference between Baseten and Vast.ai?
- Baseten: Runs dedicated model inference billed per minute of GPU time, plus pre-optimised model APIs billed per token. Vast.ai: Marketplace where independent hosts list GPUs and renters pick machines by price, with on-demand and interruptible bids.
- Is Baseten cheaper than Vast.ai?
- Vast.ai is cheaper on all 4 GPUs both rent. Vast.ai carries 74 more GPUs.
- How much does a NVIDIA B200 SXM 180GB cost on Baseten and Vast.ai?
- Baseten charges $9.98 per GPU-hour and Vast.ai charges $6.25 per GPU-hour, the cheapest live rate on each.
- Which GPUs can I rent on both Baseten and Vast.ai?
- 4 GPUs, including NVIDIA B200 SXM 180GB, NVIDIA H100 SXM5 80GB, NVIDIA A100 SXM4 80GB, NVIDIA L4 24GB.
- How are these prices compared?
- Each figure is the cheapest live rate per GPU-hour for that GPU. We use on-demand pricing where a provider sells it and fall back to community or reserved terms, labelled, where that is all they sell. Spot capacity is left out because it can be reclaimed. Rates are refreshed by our daily ingest.
How we define a live price: methodology.
