GPU Marketplace · Managed Inference

Rent GPUs. Run Models. Get Results.

Run pool-verified models on marketplace GPUs, or list your own hardware to earn.

No membership required. Add Compute Balance when you need it and pay only for what you use.

Text, Image & Video
Pool-Verified Models
Streaming & Cancellation
Managed Owner Payouts

What You Can Do

Run pool-verified text, image, video, and speech models on marketplace GPUs through one consistent API.

Text

Text Inference

Run one text request or submit independent prompts as an ordered batch.

Ollama models
Independent batches
Structured results
Image

Image Generation

Generate images with pool-managed workflows and verified model assets.

SDXL Base
Managed workflow
Verified weights
Video

Video Generation

Turn prompts into video with supported local-GPU workflows.

Wan 2.1 1.3B
LTX Video 0.9
Managed assets
Audio

Text to Speech

Turn prompts into speech with Kokoro-FastAPI on a rented GPU.

Kokoro 82M
US and UK voices
WAV playback
API

API Control

Submit, stream, inspect, and cancel jobs through the pool API.

Polling
SSE streaming
Cancellation

How It Works

01

Browse GPUs

Filter by VRAM, price, model, and region.

02

Choose Model

Select a verified text, image, video, or speech model.

03

Run on a Match

Your workload runs on a compatible marketplace GPU.

04

Get Output

Retrieve generated text, images, video, or speech.

Supported Models

The pool lists Ollama models and only those image, video, and speech presets whose assets are verified.

No models match this search.

Smarter Than a Cloud Provider

Under the hood, RunGPU automatically optimizes cost, performance, and reliability.

Smart Routing

Jobs matched to best GPU by VRAM, price, and cached models.

Workload Recovery

Interrupted queued work can return to the pool for another compatible host.

Managed Runtime

Verified model execution with workflow, capability, and memory checks.

Cost Optimization

Real-time pricing. Finds cheapest option meeting your requirements.

Isolated Workloads

Managed runtime boundaries help protect hosts while jobs are running.

Real-Time Alerts

Email and in-app notifications for every event. Full audit trail.

How a job flows through the platform

You

pick a model

→

Router

finds best GPU

→

Agent

pulls & runs

→

Output

delivered to you

Get Started in 3 Minutes

Whether you're renting or hosting — here's how to go from zero to running.

Run a Model

For users who want results without integration work

1

Browse available GPUs

Compare VRAM, price, and supported workloads.

2

Choose a verified model

Select text, image, or video generation.

3

Enter your prompt

Adjust supported generation controls and run.

4

Collect your result

View progress and retrieve output in the dashboard.

Browse GPUs

Host Your GPU

For GPU owners who want to earn

1

List your GPU

Set price, choose region. 30 seconds.

2

Download the agent

Pre-built for macOS, Linux, Windows.

3

Connect Stripe

Payouts directly to your bank account.

4

Earn automatically

Keep 85% of marketplace rentals. Fully managed.

List Your GPU
API

Automate inference from your application

API reference 
const response = await fetch("https://pool.rungpu.io/api/v1/infer", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.RUNGPU_PROJECT_KEY}`,
    "Content-Type": "application/json"
  },
  body: JSON.stringify({
    model_name: "llama3.2",
    input: { prompt: "Summarize this document" }
  })
});
const job = await response.json();
// → job.status_url, job.stream_url, job.cancel_url
For GPU Owners

Have a GPU? Start Earning.

List your GPU. Install the agent. We handle billing, security, and fault tolerance.

0%

Rental share

0s

Setup time

0/7

Auto-managed

How to start hosting

1

List your GPU

Select model, set price, choose region.

2

Download the agent

One binary — macOS, Linux, or Windows.

3

Connect Stripe

Receive payouts directly to your bank.

4

Earn automatically

Jobs arrive, agent runs them, you get paid.

List Your GPU

Download the AgentLatest release

Two options after download: Command or GUI.

Norton or other security software may flag a newly released, low-prevalence agent binary as suspicious. Download only from the official RunGPU GitHub release, verify the published checksum, and review the source before allowing it. Never disable security software just to install the agent.

All releases on GitHub →

Built for Trust

Isolated Workloads

Managed workload boundaries help protect hosts while customer jobs are running.

Controlled Inputs

Workload sources and attached assets are checked before they reach a host.

Measured Usage

Usage and job outcomes are tracked consistently for renters and GPU owners.

Auditable Agent

The GPU agent is open source so hosts can review the software they run.

FAQ

What models can I run?

The pool catalog lists supported Ollama models and any fully verified image, video, or speech presets currently available.

Do I need a membership?

No. Add Compute Balance whenever you need it and use it for GPU jobs and paid media re-downloads.

How does billing work?

You prepay your Compute Balance, see the price before using a GPU, and never carry a negative balance.

Is it safe to list my GPU?

RunGPU uses managed workload boundaries and owner controls designed to protect hosted machines.

What if a host is interrupted?

The platform tracks job state and can return eligible queued work to the GPU pool.

Is the agent open source?

Yes. GPU owners can review the agent software they install.

Ready to start?

Browse available GPUs or list your own hardware to earn.