Up to 10× cheaper than big cloud

Run your AI on gamers’ GPUs - up to 10× cheaper.

Gamers plug in their GPUs and get paid. Businesses run AI at prices the cloud can't touch. One network - everybody wins.

Earn with your GPU Run AI in 5 minutes

Supported: Windows and Linux · macOS on the way

POWERING THE WORLD'S FAVORITE OPEN-SOURCE MODELS

Llama 3

Mistral

Qwen

DeepSeek

Gemma

NemotronWhisperSDXLFLUX.1Phi-3MusicGenCommand RStable DiffusionCodeLlamaGranite

Any model you want. No allow-list.

The popular ones run with a single line of code. Need something else? Point us at any model on Hugging Face - or bring your own weights. No approvals, no waiting list, no per-model pricing games.

Thousands of models on Hugging Face - ready to run

FOR GAMERS

Your GPU earns while you sleep.

Plug in your card in 5 minutes. Collect tokens for every job it completes and swap them for Steam, Netflix, V-Bucks and 50+ more gift cards. Sitting down to play? One click and the card is all yours again.

Running in 5 minutesPause with one clickTokens never expire

See how much you'd earn How it works →

Last night, while you slept +6,410 t

BUSINESS

Got a whole fleet? Even better.

Server rooms, render farms, post-mining rigs. Connect dozens of cards at once - every card earns on its own, and it all lands on a single invoice.

One node, many cards.: Every GPU works its own jobs in parallel.
Datacenter rates.: H100 and A100 play in a higher pricing league.
Payouts built for business.: Bank transfer, VAT invoice, monthly statement.

Let's talk about your fleet

FOR DEVELOPERS

From idea to AI in 5 minutes.

One API for every model. Tell us what to run and the most you're willing to spend - the network does the rest.

01Create an account. Takes a minute.

02Send a job. One request, one hard cost cap.

03Get the result and the receipt. To the second, to the token.

Create an account API docs ↗

quickstart.py

import griddly result = griddly.infer( model="qwen2.5-coder-7b", prompt="SQL: total sales by month", max_cost_tokens=500,)print(result.text)

# assigned: RTX 4090 · gold node

griddlyapigpu

› SELECT month, sum(total) FROM orders GROUP BY month

# 3 tokens · $0.0003 · 2.4 s

quickstart.py

import griddly result = griddly.infer( model="flux.1-schnell", prompt="Olive oil bottle packshot, white background", max_cost_tokens=500,)result.save("packshot.png")

# assigned: RTX 4090 · platinum node

griddlyapigpu

› output/packshot.png · 1024×1024

# 4 tokens · $0.0004 · 3.4 s

quickstart.py

import griddly result = griddly.infer( model="whisper-large-v3", audio="customer-call.mp3", max_cost_tokens=500,)print(result.text)

# assigned: RTX 3090 · silver node

griddlyapigpu

› Hi, I'm calling about the invoice from March…

# 8 tokens · $0.0008 · 10.3 s

PRICING

Seconds of GPU time. Nothing else.

No subscription, no minimum, no per-model price list. You buy tokens, a job spends them for as long as it actually runs, and you set the ceiling before it starts.

10,000 tokens = $1

1 token = $0.0001

RTX 4090 $0.45/h

Billing starts when the job starts - pulling the model and waiting in the queue are free.

Every job carries a hard budget. The moment it would go over, we stop it.

Tokens do not expire, and the rate is the same whether you spend ten dollars or ten thousand.

Work out your job Full pricing

SECURITY

Someone else’s computer, sealed shut.

A marketplace of strangers’ GPUs only works if you do not have to trust the strangers. Three of the six locks every job goes through:

One door out.

A job reaches Griddly and nothing else - not the internet, not the machine it runs on, not the owner’s home network.

Fresh keys every time.

Each job gets its own encryption keys, created at start and destroyed the moment it ends.

Watching is fatal.

A host that so much as touches a running job is cut off immediately and banned for good.

The whole protocol

WHERE WE'RE HEADED

One network. Ever more power.

Stage	Status	What lands
GPU marketplace	LIVE	AI on demand: text, image, audio, video, music.
Rendering & macOS	NEXT	Film frames across hundreds of cards; Apple Silicon joins the network.
Model training	PHASE 2	Fine-tune your own models on gamer GPUs.
Business scale	PHASE 3	Confidential workloads and hard guarantees for the big players.