What is OpenGradient (OPG)?
OpenGradient (OPG) is a decentralized infrastructure network that makes AI inference cryptographically verifiable. It operates as a specialized AI coprocessor rather than a standalone blockchain, allowing applications and on-chain protocols to outsource heavy AI workloads to its network. The OPG token serves as the payment, incentive, and security layer of this verifiable intelligence economy.
What functions does OPG serve within the network?
OPG fulfills several roles inside the OpenGradient ecosystem. It is first and foremost a settlement medium: every verified AI call on the network is paid for with OPG, and settlement occurs on the Base network.
Model creators also earn through OPG. When their models uploaded to the Model Hub are used, they receive payments denominated in OPG. Node operators and token holders who want to contribute to network security can stake OPG and earn rewards distributed over a multi-year schedule. Beyond that, OPG grants voting rights on protocol upgrades, fee parameters, and foundation treasury management, giving holders a say in governance.
How does OpenGradient's hybrid architecture work?
Traditional blockchains excel at validating token transfers, but re-running expensive, non-deterministic AI inference on every node is impractical. OpenGradient addresses this with the Hybrid AI Compute Architecture (HACA).
The core idea is separating execution from verification. Inference requests are routed to dedicated GPU and TEE (Trusted Execution Environment) nodes and completed at web2-like speeds. After that, full nodes asynchronously validate the associated proofs and record the result on the EVM-compatible ledger running on Base. The system supports a spectrum of verification methods: zkML (zero-knowledge machine learning) proofs offer mathematical certainty, while TEE attestations rely on hardware-based secure enclaves. This design combines speed with security, and developers can call AI models directly from Solidity smart contracts through a simple function call.
