“Rent a GPU” means different things on different platforms. On some you get a full container you log into. On others you get an endpoint that runs your container for you. On GPUFlow you get an API key for an AI model. The right choice depends less on price and more on what you’re trying to do.
We compared four platforms that rent consumer GPUs like the RTX 3090 and 4090. Everything here was checked in September 2026 on each platform’s own docs and pricing pages; sources are at the end.
What you actually get
| GPUFlow | Vast.ai | RunPod | SaladCloud | |
|---|---|---|---|---|
| What you rent | An API key for AI models on one GPU | A container on a host’s machine | A pod (container), or serverless workers | Container groups on home PCs |
| How you use it | OpenAI-compatible API: /v1/models, /v1/chat/completions | SSH, Jupyter | SSH, JupyterLab, VS Code, web proxy | Your container’s API; SSH into running instances |
| Run your own code | No | Yes | Yes | Yes |
| Setup before first use | None | Pick an image, download your model | Pick a template, download your model | Build and deploy a container |
| Where the GPUs are | Providers’ own computers | Individuals up to data centers | Secure Cloud (data centers) and Community Cloud | Consumer PCs (“Chefs”) |
Money
| GPUFlow | Vast.ai | RunPod | SaladCloud | |
|---|---|---|---|---|
| Billing | Per second, 1-minute minimum | Per second, no minimum | Per second | Per second |
| Storage charges | None | Set by host, also while stopped | $0.10/GB/month; $0.20 for a stopped pod’s volume | Not stated (the GPU price includes vCPU and RAM) |
| Data transfer | None | Set by host, every byte | Free | Not stated |
| To get started | $10 top-up minimum, no fee | $5 minimum deposit | At least 1 hour of credit; $100 for prepaid cards | Top-ups from $5 |
| Payment | Card (Stripe) | Card, BitPay, Crypto.com | Card, crypto, invoicing over $5,000 | Card, crypto on Solana |
| Credit expiry | Never; unused rental time is refunded | — | — | 12 months after purchase |
A dash means we didn’t find a rule in that platform’s docs.
Prices for common cards, September 2026
Per GPU per hour, on demand:
| GPU | Vast.ai | RunPod Community / Secure | SaladCloud |
|---|---|---|---|
| RTX 3090 | from about $0.11 – $0.12 | $0.22 / $0.50 | $0.17 |
| RTX 4090 | from about $0.37 | $0.34 (out of stock) / $0.74 | $0.33 |
| RTX 5090 | about $0.43 | $0.69 (out of stock) / $0.99 | $0.50 |
Vast.ai and SaladCloud prices are from getdeploying.com, because Vast’s own price table didn’t load when we checked. SaladCloud also sells lower-priority capacity for less, which can be interrupted. RunPod raised its Secure Cloud prices on September 20, 2026; Community Cloud prices didn’t change.
On GPUFlow each provider sets their own price. The typical range across rental sites is $0.11 – $0.31 for an RTX 3090 and $0.30 – $0.46 for an RTX 4090, and the GPUFlow listing form shows providers where their price sits in that range.
Remember that these aren’t the same product. A $0.30 container that takes 20 minutes to set up and a $0.35 API key that works immediately cost different amounts for a one-hour job. The real cost of renting a GPU goes through the extras.
Which one fits your job
Training or fine-tuning a model
Vast.ai or RunPod. You need the whole environment: your code, your data, your libraries. Vast.ai is usually cheaper; RunPod has more ready-made templates and a data-center tier. Our Stable Diffusion LoRA training walkthrough prices a typical run on both. GPUFlow can’t do this: it doesn’t give you a machine.
Running your own container at scale
SaladCloud or RunPod serverless. Both run your container across many GPUs and handle scaling. Salad runs on consumer PCs, so instances can be interrupted and local storage doesn’t persist; design your job for that. RunPod serverless bills start time and an idle timeout on top of processing time.
Calling an open model from an app, a script or a chat tool
GPUFlow, if a provider runs the model you want. You get an OpenAI-compatible key, so the OpenAI libraries, LangChain, Open WebUI and most chat apps work by changing the base URL. There’s no server to maintain, and you pay by the second for the hours you book. If you end early, the rest goes back to your credits. How to use the key in your tools.
What GPUFlow doesn’t do: embeddings, image generation, the Responses API, or running your own code. And the model runs on the provider’s own computer, so your prompts pass through it. Don’t send anything you wouldn’t share with a stranger.
Trying a model before you buy a GPU
Any of them. On GPUFlow it takes a few minutes and no setup. On Vast.ai or RunPod you can also test your own inference server settings. Either way, an hour costs less than a coffee.
Only need cheap tokens from a popular model
None of these, maybe. If a hosted API offers the model you want, per-token pricing can be far cheaper than renting a GPU. We did the math in hourly GPU or per-token API.
If you have a GPU to rent out
| GPUFlow | Vast.ai | Salad | |
|---|---|---|---|
| Operating system | 64-bit Linux with systemd | Ubuntu | Windows 10/11 |
| What renters can reach | Your AI models through GPUFlow’s relay. No shell, no open ports | A container on your machine | Salad’s workloads |
| Your share | 88% | Vast says listed prices are typically about 25% above what hosts earn | Not published |
| Payouts | Bank through Stripe, $25 minimum, $2.50 per cash-out | Wise, PayPal or Stripe, $20 minimum | PayPal, gift cards and more |
The full math per card is in what your gaming GPU can earn.
Summary
- Need a machine? Vast.ai for price, RunPod for convenience and a data-center option.
- Need a scalable container service? SaladCloud or RunPod serverless.
- Need an AI model behind an OpenAI-style API, with nothing to set up? GPUFlow.
- Need the cheapest tokens from a popular model? Check hosted per-token APIs first.
Sources
All checked in September 2026.
- GPUFlow: renting, billing, API, providers, getting paid, price ranges
- Vast.ai: quickstart, pricing, billing, hosting, host payouts, host earnings
- RunPod: pricing, pod pricing, serverless pricing, billing, pods
- RunPod Secure Cloud price change: usagepricing.com
- SaladCloud: billing, container billing, priority pricing, SSH, Salad for hosts
- Prices: getdeploying.com for RTX 3090, RTX 4090, RTX 5090, Vast.ai, Salad