A complete overview of the Bittensor network — 128 subnets across 17 categories, plus infrastructure, wallets, exchanges and developer tools.
Each subnet is categorized based on its primary function: what it actually does for the network. Categories are assigned manually based on the subnet's documentation, GitHub repository, and on-chain activity. Some subnets span multiple areas (e.g. a training subnet that also serves inference), but we pick the dominant function.
Serving AI models to users
Training & fine-tuning models
Collecting & curating datasets
Raw GPU, infrastructure, mining
Liquidity, lending, trading
Forecasting & oracle networks
Auditing, detection, privacy
Autonomous AI agents
Research & scientific compute
Video, audio, content creation
Games & virtual worlds
Drones, autonomous vehicles
Decentralized file storage
Sports analytics & signals
SDKs, testing, tooling
Every subnet gets a health score from 0-100 based on four pillars. Scores use percentile ranking, so a subnet in the top 5% for a metric scores ~95 on that component. This means the score reflects how a subnet compares to all others, not arbitrary thresholds.
TAO locked in the pool, market cap, and trading volume. Higher liquidity means less slippage and more confidence from stakers.
Validator count, miner slot utilization, and miner-to-validator ratio. A well-populated subnet signals real demand for its services.
TAO emission received, registration cost demand, market confidence, subnet age (older = more proven), and tempo efficiency.
Buy/sell ratio (money flowing in or out), 7-day price trend, trend consistency, and trading volume. Stable or growing subnets score well.