Renting a GPU through an API

A GPU API turns a rental into a few HTTP calls: pick a GPU type and an image, create the pod, run commands or read its logs, and terminate it when the work is done. For an agent, that is the difference between doing the job and asking a person to click through a console. rent.dev puts GPU pods and sandboxes behind one API and an MCP server. GPU pods run on the agent's own key at the GPU provider rent.dev connects to, and that provider bills the key.

rent.dev: one API for the sandboxes and GPU pods an agent needs.

What the quotes API lists, seller by seller

Rates as each vendor publishes them, normalised per GPU, from rent.dev's price board, generated 2026-11-14. See the whole board. Browse the full catalog.

VendorTierGPU typesLowest per GPU hourForRows with stockRate read
Akash22$0.02gtx1050 2Gi PCIe02026-11-11
RunPodcommunity36$0.12NVIDIA RTX A2000362026-11-11
Verdaspot15$0.122Tesla V100 16GB02026-11-11
RunPodsecure32$0.24NVIDIA RTX 2000 Ada Generation322026-11-11
Verdaon demand15$0.244Tesla V100 16GB02026-11-11
AWS23$0.37853g4ad02026-11-11
Beam9$0.44RTX 4090 machine02026-11-11
Daytonapreemptible9$0.57Nvidia RTX 409002026-11-11
RunPodserverless29$0.58NVIDIA RTX 2000 Ada Generation292026-11-11
Modal11$0.5904Nvidia T402026-11-11
CoreWeavespot8$0.78375NVIDIA L4002026-11-11
Northflank7$0.80NVIDIA L402026-11-11

Questions

How do I rent a GPU through an API?
On rent.dev: create an account with POST https://api.rent.dev/v1/accounts, store your GPU provider key with POST /v1/byok/credentials, then create a pod with POST /v1/gpu-pods, naming the GPU type and a container image. Run commands through the pod's exec route, stream its logs, and terminate it when finished. The pod has no timeout and is billed until it is terminated.
What are the best GPU APIs for agents and developers?
Judge them by what an agent can do without a person: create and terminate by API, list GPU types and rates before creating, report stock, stream logs, and offer the whole flow as MCP tools. Look at the errors too, since a clear out-of-stock or quota answer lets an agent pick another GPU type instead of retrying blindly. rent.dev's REST API covers that whole flow. Its CLI and its MCP server create, list and terminate GPU pods, while running a command in a pod and reading its logs are REST calls. rent.dev's quotes answer, generated 2026-11-14, lists 448 priced GPU rows across 166 GPU types from 12 vendors, and 97 of those rows carry a stock level.
Can an AI agent get GPU compute without human signup?
Partly. An agent can create a rent.dev account itself with one unauthenticated call, or with the rent_signup tool over MCP, and it can read GPU rates with no account at all. GPU pods, though, run on a key at the GPU provider rent.dev connects to, and that account is normally opened by a person who adds a payment method. Once the key is stored, the agent creates and terminates pods on its own. Without a key, the quotes answer generated 2026-11-14 starts its GPU rates at $0.02 per GPU hour, for the gtx1050 2Gi PCIe (Akash).
What is the difference between a serverless GPU API and a GPU pod API?
A serverless GPU API runs a function or container per request and scales it down when idle, which suits inference with uneven traffic. A pod API gives you a whole machine with a GPU that runs until you end it, which suits training, long jobs and anything that keeps state. rent.dev's GPU API is the pod kind.