Intro to Subnet 2

The world's fastest zkML proving cluster

Subnet 2 is a decentralized zero-knowledge machine learning (zkML) proving network built on Bittensor. It provides verifiable AI inference through cryptographic proofs, enabling any computation to be proven and verified trustlessly.

Network Performance

Metric
Value

Proofs Generated

300M+

Unique Miners

1,500+

Proof Time Improvement

10x

Subnet 2's incentive mechanism rewards faster, smaller proofs—driving continuous optimization without central coordination. In 2025, the network improved average proof times by an order of magnitude through competition alone.

What is Proof-of-Inference?

Proof-of-Inference is the core mechanism that powers Subnet 2. When an AI model runs an inference, the network generates a zero-knowledge proof that mathematically guarantees:

  • The specified model was actually executed

  • The input data was processed correctly

  • The output is authentic and untampered

This transforms AI from a black box into a verifiable system where outputs can be cryptographically proven to be correct.

How It Works

REQUEST → SLICE → PROVE → RESPOND
  1. Request: Clients submit inference requests to the network

  2. Slice: Computation is distributed across miners

  3. Prove: Miners generate zero-knowledge proofs in parallel

  4. Respond: Proofs are aggregated and verified, output is returned as cryptographically verified

Network Architecture

Subnet 2 operates on Bittensor's incentive network with two key participants:

Miners receive input data from validators, run inferences through zk-circuits, and return outputs with accompanying proofs. Miners compete on proof generation speed, proof size, and output accuracy. The network is CPU-intensive, providing opportunities for non-GPU miners.

Validators distribute inference requests to miners, verify the returned proofs, and score miners based on performance metrics. Validators can accept external queries, routing organic traffic through the network.

Applications

Studio

Build and deploy your own zero-knowledge circuits. Convert AI models into verifiable circuits that can be proven on the Subnet 2 network.

Stats

Real-time analytics dashboard showing network performance, miner leaderboards, proof generation metrics, and model statistics.

Last updated