Beta Open models, paid per token in USDC on Base

Spare Compute Serious Models

One OpenAI-compatible API for Qwen, Llama, DeepSeek and Mistral, served by idle computers around the world. Developers pay per token. Node owners get paid for every token their machine serves.

Get an API key Earn with your computer

Change one base URL in the OpenAI SDK. No card, no subscription — just a wallet.

Why sparehash

One API, Many Machines

A network of desktops and workstations running open-weight models, behind a router that picks a good one for every request.

Drop-in OpenAI API

Point the OpenAI SDK at /v1. Streaming, embeddings and JSON mode work as you'd expect.

Quickstart →

Priced per model

Small models cost cents per million tokens. You pay for the tokens you receive — failed requests are free.

See live prices →

Failover built in

If a machine drops mid-answer, another one picks up exactly where it left off. Your stream just keeps going.

How it works →
Two sides

Build On It, Or Power It

For developers

  • Open-weight models from 8B to 70B, plus embeddings
  • Prepaid credits in USDC on Base — top up from any wallet
  • Per-key rate limits, usage history and a verified-nodes tier
  • GPT, Claude and Gemini too, discounted, from sellers' spare capacity

Read the quickstart →

For node owners

  • A tray app runs models when your machine is idle, on your schedule
  • Earn a share of every token you serve, paid in USDC to your wallet
  • Apple Silicon and NVIDIA GPUs; the app picks models that fit

How earning works →

Get started

Live In Three Steps

  1. Connect a wallet

    Sign one message to prove it's yours. No password, no gas.

  2. Add USDC credits

    Send USDC on Base from the same wallet. Credits appear after confirmation.

  3. Call the API

    Create a key and swap your base URL. That's the whole migration.

Node.js 
import OpenAI from 'openai';

const client = new OpenAI({ apiKey: process.env.SPAREHASH_KEY, baseURL: 'https://your-coordinator.fly.dev/v1' });
const res = await client.chat.completions.create({
  model: 'qwen3-8b',
  messages: [{ role: 'user', content: 'Hello!' }],
});
Private by default: credentials and personal details are swapped for placeholders before a request leaves sparehash, and put back into the answer for you. The machine running the model never sees them. How it works and its limits

Ready To Build?

Connect a wallet, add a few dollars of USDC and make your first call in minutes.

Open the dashboard Earn instead