Start
Overview
Mason is a permissionless network of user-owned GPUs that serves open models through one API, and pays the people who run it.
What Mason is
Mason turns idle GPUs into one inference machine. Anyone with a graphics card can join as a worker, from a browser tab or with a full node. Anyone who needs AI can send a request to the API or open the chat. The orchestrator matches each request to a worker, streams the answer back, and settles the payment.
Users pay in credits priced in dollars and bought with ETH on Robinhood Chain. Workers keep 70% of what each job earns, or 80% if they stake $MASON. The rest funds a treasury that, every day, buys back and burns $MASON and pays stakers.
Why it exists
Open models are free to download but expensive to serve. Running them means renting GPUs from a handful of clouds, which puts the same few companies between every user and every model. Meanwhile, millions of capable GPUs sit idle in gaming PCs and workstations.
Mason connects the two: demand pays supply directly, and the margin goes to the people who hold and stake the network's token instead of to a landlord.
Pick your door
| You are | You get | Start here |
|---|---|---|
| A developer | An OpenAI-compatible endpoint for open models, priced per token | Quickstart, API reference |
| A GPU owner | ETH for every job your GPU completes | Run a worker |
| A holder | Daily buybacks and burns, and a share of revenue when you stake | $MASON, Staking |
The film
38 seconds: the problem, how a request travels through the network, where the money goes, and the product running on a real GPU.
Three principles
- Open. Open-weight models only, served by anyone who passes the hidden checks. No allow-list of operators.
- Private. Prompts and completions are streamed, not stored. Workers see a job, not an identity.
- Owned. Revenue flows to workers and stakers on a fixed, public split, and every number is on the Network page.