Bittensor is a decentralized network where machine learning models meet, trade useful outputs, and get paid in a native token called TAO. The idea is to turn model insight and compute into a shared marketplace that anyone can join, rather than keeping AI locked inside a single company.
Bittensor launched in 2021 and is developed by the OpenTensor Foundation. Its blockchain layer is called Subtensor, built with the Substrate toolkit. The project first appeared as a Polkadot parachain known as Finney, then moved to its own chain named Nakamoto in March 2023.
The network is organized into subnets. Each subnet focuses on a task and runs a competitive marketplace. Two main roles keep things moving. Miners host and serve models that answer prompts or provide signals. Validators query those models, score the responses, and help decide who earned rewards.
Bittensor uses a performance-based approach to choose who earns and who writes blocks. Instead of hashing puzzles, nodes are evaluated on the usefulness and accuracy of the model outputs they produce. Better outputs lead to more weight and more rewards.
TAO is the network’s currency for rewards and coordination. The supply is capped at 21 million. New TAO is minted on a schedule of 1 TAO roughly every 12 seconds, which is about 7,200 TAO per day. The issuance rate halves about every four years, similar to Bitcoin.
To register keys for mining or validating, participants must recycle TAO. Recycling removes tokens from circulation and returns them to the issuance pool. Because registration and deregistration change how much TAO gets recycled, the timing of halvings can shift.
Resource flow across the network adjusts dynamically. Each subnet can have its own token paired with TAO in embedded liquidity pools. Market performance of those subnet tokens influences how newly issued TAO is split between subnets. Subnet tokens are released through fair-launch style distributions that favor teams who keep building over quick sales.
Validators stake TAO to earn the right to set weights in a subnet, which raises the bar for bad actors. A common setup is a minimum stake of 1,000 TAO for validators, and users who do not run infrastructure can delegate TAO to a validator to share rewards.
Projects can plug in models for inference or other AI services, compete for rewards, and bootstrap specialized subnets. The setup supports cross-network integrations and makes it easier for teams to attract attention and liquidity if their subnet performs well.
The network can be demanding for newcomers because running high quality models and validators takes capital and expertise. TAO’s value is tied to the health of the Bittensor ecosystem, and like any crypto asset it can be volatile. Decentralized AI is still early, so long-term outcomes are uncertain.