- Services
Live per-GPU prices from GPU Mart and Vast.ai, side by side. 10 GPU models are listed on both. Prices come from each provider’s own pricing data and refresh hourly where the provider has a pricing API.
TL;DR
Vast.ai is cheaper on 10 of the 10 GPU models both providers list. GPU Mart lists 17 GPU models and Vast.ai lists 28.
Advertising disclosure: we earn commissions when you shop through the links below.
Prices refreshed:
The cheapest current per-GPU rate on each provider. The lower price is highlighted. Spot rates are interruptible and shown underneath. Click a price to open the provider.
Methodology: Prices for GPU Mart and Vast.ai are refreshed hourly and normalized to per-GPU USD/hr. No affiliate weighting. GPU Mart has no API. Its prices come from the published rate card, are updated by hand, and bill monthly. Vast.ai figures come from its marketplace API and cover verified hosts only.
Price is one input. The hosting model, reliability and day-to-day experience decide whether a provider fits your team and workload.
The cheapest single GPU on each provider with enough memory to run the model at 4-bit quantization. Models that need several GPUs are left out.
| Model | VRAM at Q4 | GPU Mart | Vast.ai |
|---|---|---|---|
| Mistral 7B Instruct | 7 GB | $0.14/hrNVIDIA GeForce RTX 5060 | $0.06/hrNVIDIA GeForce RTX 3060 |
| Gemma 4 E4B IT | 7 GB | $0.14/hrNVIDIA GeForce RTX 5060 | $0.06/hrNVIDIA GeForce RTX 3060 |
| Qwen3.6 35B-A3B | 9 GB | $0.20/hrNVIDIA RTX A4000 | $0.06/hrNVIDIA GeForce RTX 3060 |
| Qwen3.5-35B-A3B | 9 GB | $0.20/hrNVIDIA RTX A4000 | $0.06/hrNVIDIA GeForce RTX 3060 |
| Gemma 4 26B-A4B IT | 12 GB | $0.20/hrNVIDIA RTX A4000 | $0.06/hrNVIDIA GeForce RTX 3060 |
| Llama 3.1 8B Instruct | 14 GB | $0.20/hrNVIDIA RTX A4000 | $0.09/hrNVIDIA Tesla V100 16GB |
| Gemma 4 12B | 33 GB | $0.62/hrNVIDIA RTX PRO 5000 Blackwell | $0.40/hrNVIDIA RTX A6000 |
| DeepSeek-R1 | 60 GB | $1.09/hrNVIDIA RTX PRO 6000 Blackwell | $0.44/hrNVIDIA A100 80GB SXM |
| DeepSeek-V3 | 60 GB | $1.09/hrNVIDIA RTX PRO 6000 Blackwell | $0.44/hrNVIDIA A100 80GB SXM |
| DeepSeek-V3.2 | 60 GB | $1.09/hrNVIDIA RTX PRO 6000 Blackwell | $0.44/hrNVIDIA A100 80GB SXM |
| Qwen3.6-27B | 73 GB | $1.09/hrNVIDIA RTX PRO 6000 Blackwell | $0.44/hrNVIDIA A100 80GB SXM |
| Gemma 4 31B IT | 82 GB | $1.09/hrNVIDIA RTX PRO 6000 Blackwell | $2.66/hrNVIDIA H100 NVL |
34 vs 28 GPU models listed
See the comparison17 vs 34 GPU models listed
See the comparison10 vs 28 GPU models listed
See the comparison28 vs 0 GPU models listed
See the comparison10 vs 17 GPU models listed
See the comparison17 vs 0 GPU models listed
See the comparisonStraight answers to the questions we hear most often.
Can't find what you're looking for? Book a Discovery Call
Right now Vast.ai has the lower per-GPU rate on 10 of the 10 GPU models both providers list, and GPU Mart on 0. Prices move every hour, so check the matrix above for the card you need.
Vast.ai lists the NVIDIA GeForce RTX 4090 at $0.41 per GPU-hour (on-demand), against $0.62 on GPU Mart. That is 34% less, or about $152 saved per month on a GPU that runs around the clock.
Vast.ai lists the NVIDIA GeForce RTX 5090 at $0.53 per GPU-hour (on-demand), against $0.74 on GPU Mart. That is 28% less, or about $149 saved per month on a GPU that runs around the clock.
Right now GPU Mart lists these GPU models that Vast.ai does not: NVIDIA RTX PRO 6000 Blackwell, NVIDIA H100 PCIe, NVIDIA A100 40GB PCIe, NVIDIA RTX PRO 2000 Blackwell, NVIDIA GeForce RTX 3060 Ti, NVIDIA GeForce RTX 4060, and NVIDIA GeForce RTX 5060.
Right now Vast.ai lists these GPU models that GPU Mart does not: NVIDIA B200, NVIDIA H200 NVL, NVIDIA H200 SXM, NVIDIA H100 NVL, NVIDIA A100 80GB SXM, NVIDIA H100 SXM, NVIDIA L40, NVIDIA L40S, NVIDIA GeForce RTX 3090, NVIDIA GeForce RTX 5060 Ti, NVIDIA GeForce RTX 5070 Ti, and NVIDIA GeForce RTX 5080.
GPU Mart has spot pricing on 0 GPU models and Vast.ai on 28. Spot instances are the cheapest way to rent but can be reclaimed at short notice, so use them for jobs that checkpoint and resume.
GPU Mart is best for Vast.ai is best for
GPU Mart: Vast.ai:
Every hour for providers with a pricing API. GPU Mart has none, so its rate card is updated by hand. We normalize the GPU names and convert multi-GPU instances to a per-GPU rate so the two columns compare like for like.
Rates vary by region, host and availability. We show the cheapest per-GPU rate currently rentable on each provider. Marketplace hosts change their prices often, so the provider page may show a different default.
