Skip to main content
The Crunch Protocol distributes AI and machine-learning workloads across a global network of independent contributors, connecting domain experts with thousands of ML engineers to solve real-world prediction problems. As a Coordinator, you define a prediction challenge (called a Crunch), provide data, and set the rules. The protocol handles everything else — model submission, execution, scoring, and reward distribution — so you can focus on the problem you want to solve.

​
The Problem: The Illiquidity of Intelligence

Despite the explosion of AI, most organizations still rely on small, centralized R&D units. This creates two critical bottlenecks:
  • Silos — Teams operate within technical “bubbles,” limited by internal biases
  • Cost — The R&D cost to achieve marginal model improvements keeps rising
Companies also struggle to onboard new talent in a market where demand outstrips supply. This structural scarcity makes the current job market for AI intelligence fundamentally illiquid and inefficient. Crunch Protocol’s clients are already spending millions of dollars to solve problems they delegate to our network.

​
The Solution: Validated Decentralized Model Markets

Crunch solves the top-talent bottleneck by creating a performance market. It provides permissionless ecosystems that coordinate modeling intelligence at global scale. Market pressure drives continuous selection: strong contributors rise, weaker ones are phased out, and the network advances through competitive improvement. Our platform has proven that this approach provides:
  • Continuous innovation — A supply of models from independent contributors that outperforms internal teams
  • Institutional rails — Infrastructure that securely connects decentralized talent to corporate data
  • Instant settlement — Incentives paid strictly on performance, not hours worked

​
How it works

The protocol brings together two participant groups:

Coordinators

Domain-expert teams who define prediction challenges, provide data, set scoring rules, and distribute rewards. You run a Crunch Node that orchestrates the competition.

Crunchers

Machine learning engineers who develop and submit models. They earn rewards based on prediction quality, measured against the rules you define.
Each Crunch:
  • Locks rewards — An escrowed USDC pool funds successful participants
  • Sets rules — Evaluation criteria, duration, and payout schedule
  • Enforces quality — Models are continuously evaluated against real-world outcomes
  • Sets constraints — Resource limits like maximum GPU time, RAM, and minimum performance thresholds
This isn’t theoretical. Major institutions — including ADIA Lab, the Broad Institute of MIT and Harvard, and high-frequency trading firms — are already using Crunch to outperform their internal benchmarks. See Use cases for details.

​
Becoming a Coordinator

The onboarding process breaks down into five steps:
  1. Set up your local environment — Scaffold a workspace and run a complete Crunch locally
  2. Define your prediction task — Specify input data, output format, and the model interface
  3. Configure scoring — Build the scoring function that evaluates and ranks predictions
  4. Register on the protocol — Create a Solana wallet and register through the Coordinator Platform
  5. Deploy — Push your Crunch to testnet, then mainnet through the Coordinator Platform

​
Next steps

Core concepts

Understand the protocol architecture before building.

Getting started

Jump straight into building your Crunch Node.

​
Community

Forum

Discuss strategies and get support

Discord

Real-time chat with the community

Twitter

Latest updates and announcements