TAO is the token of Bittensor, a network where AI work is produced, scored and paid for in the open. It turns up in lists of AI-related coins, but what it is actually used for is rarely spelled out.
This article checks three things against public data and the official documentation: what kind of token TAO is, how its price has moved so far, and where and how it can be held.
Every figure was measured on 22 September 2026 at around 20:30 JST. It is written for people holding crypto with their own money, and it neither recommends a trade nor predicts a price.
What TAO (Bittensor) is
| Name | Bittensor |
|---|---|
| Ticker | $TAO |
| Launched | Trading by 2023 |
| Market cap rank | 33rd ($3.58bn) |
| Price | $315.57 |
| Supply cap | 21 million |
| Developer | Opentensor Foundation |
Bittensor splits AI work, such as generating text or supplying computing power, into many small markets, and pays TAO to the participants whose output is judged to be good.
TAO runs on its own blockchain and has the same hard cap as bitcoin, 21 million coins. The launch row in the table is conservative: CoinGecko records its all-time low in May 2023, so it was trading by then.
The developer row refers to the Opentensor Foundation, which leads development of the protocol. The rules for issuance and distribution are written in the public documentation rather than decided privately.
Why a token like TAO came about, and what it changes
What is awkward about building AI today
The Bittensor whitepaper names two problems with how AI is produced today.
- Individual researchers find it hard to turn the value of the AI they build directly into income
- As a result, a small set of large companies controls access to the best AI
The project’s own site goes further, describing the concentration of control over AI as the greatest risk of all.
What Bittensor changes
Bittensor sets up a market in which AI output is scored by other AI systems. Participants whose output is judged useful are paid in TAO in proportion to what they contribute.
In the site’s own words the approach is “bottom up, rather than top down”: the aim is to spread ownership of AI from a few companies to the people who take part.
How do researchers actually use it to earn money?
The main route for a researcher is to become a miner on a subnet.
What a miner provides is results computed on its own computer. You do not sell the code of the AI you built; you compute answers to the tasks a subnet sets and hand them back.
What actually keeps running is a computer with GPUs, such as a server or a cloud machine. The researcher’s AI model is part of the program running on it.
What you hand back depends on the subnet. Real examples (checked on each project’s official site):
- AI inference: answering questions with AI. SN64 Chutes provides the compute to run openly published AI models
- Compute: lending GPUs and CPUs. SN4 Targon rents out GPUs and CPUs for training and running AI
- Prediction, storage and more: the official documentation lists inference, compute, storage and prediction as examples
Miners are not paid from users’ fees but from tokens the chain newly issues. The more consistently you return good results, the more you receive.
Following the mechanism in the official documentation (read on 26 September 2026), it works in this order:
- choose a subnet whose task fits your AI (inference, prediction, storage and so on)
- pay TAO, which is burned, to register as a miner
- keep a GPU-equipped computer running the program (including your AI model) to the subnet’s rules
- receive the subnet’s own alpha token according to the validators’ scores
- swap the alpha for TAO and, if you need to, sell it on an exchange for yen or dollars
What miners receive is the 41% of each subnet’s issuance set aside for them, divided according to their scores. The higher you rank, the larger your share.
The other route is to create a subnet. You define the task and how it is scored, and as owner you receive 18% of that subnet’s issuance.
This is what the whitepaper means when it says researchers can monetise their work directly. Without being hired by a company or selling their results, the work their AI does is itself what gets paid.
There are catches to keep in mind:
- Registration costs TAO: you pay before you start
- Low performers are removed: when the slots (256 by default) are full, the lowest-ranked neuron past its immunity period is pruned
- You supply the compute: GPUs and other hardware are the researcher’s own cost
- Income varies every time: it depends on your ranking and on alpha and TAO prices, and swapping costs fees and slippage
Where to get spot TAO
Registration is paid in spot TAO on the Bittensor chain itself. TAO on Hyperliquid is a perpetual future, not spot TAO, so it cannot be used to register.
Checking each exchange’s public data on 26 September, spot TAO could be traded on the exchanges below. The Japanese exchanges OKJ and bitbank did not list it.
- Binance (TAO/USDT)
- Coinbase (TAO/USD)
- Kraken (TAO)
- OKX (TAO/USDT)
After buying, withdraw it to your own wallet on the Bittensor network. Before you buy, check that the exchange’s withdrawal screen offers that network.
How the registration fee is actually paid
You pay with the official command-line tool, btcli (steps from the official documentation, read on 26 September).
- install btcli and create a wallet with
btcli wallet create. This gives you two keys: a coldkey that holds funds and a hotkey that identifies you on subnets - withdraw the TAO you bought to the coldkey’s address, sending a small test amount first
- run
btcli tx burned-register, naming the subnet (netuid) and your hotkey - the fee is taken from the coldkey and the hotkey is registered as a miner on that subnet
The fee differs by subnet, and it rises as registrations come in and falls as time passes. Check the current amount before you run the command.
The social impact
- Open participation: researchers, compute providers, data companies and others can join by registering, without being hired by a lab
- Income for researchers: the work an AI model does is paid for directly in TAO
- A market for the building blocks of AI: each subnet trades one kind of output, such as compute, inference, storage or prediction
The limits
- Preventing concentration is a stated goal; no independent assessment of whether it has been achieved was found
- The weight of each validator doing the scoring depends on how much TAO is staked with it
- Subnets vary widely in quality, and not all of them produce AI that is useful in practice
What makes TAO different
Each kind of AI work has its own subnet
Inside Bittensor, work is divided into subnets, each defining its own task. Reading the chain directly, there were 128 subnets registered, which is also the current limit of 128.
Within each subnet, miners produce output and validators score it. How that scoring works is set subnet by subnet, and the rules are published rather than hidden inside a company.
Newly issued tokens in a subnet are split three ways: 41% to miners, 41% to validators and the people staking with them, and 18% to the subnet owner who set the task up.
Issuance halves on the way to the cap
A block is produced every twelve seconds. At first each block issued 1 TAO, roughly 7,200 TAO a day. The first halving came in December 2025, and issuance is now about 3,600 TAO a day.
Halvings are triggered by the amount already issued, not by a calendar date. The next one comes when issued supply reaches 15.75 million; on 22 September it stood at 11.54 million.
That also means nobody can give an exact date for the next halving. It depends on how issuance is counted along the way, so the milestone in coins is the reliable part.
Staking TAO to back a subnet
Holders can stake TAO with a validator and receive part of the rewards. When staked into a particular subnet, TAO is exchanged for that subnet’s own token, called alpha.
Creating a new subnet carries a registration cost of at least 1,000 TAO. The cost doubles each time a subnet is registered and then falls back gradually over about two weeks.
How TAO has moved so far
As of 22 September the price had moved +9.6% over twenty-four hours, +39.4% over seven days and +40.4% over thirty days. Against a year earlier it was up just 1.7%, roughly the same level.
Within a single day it ranged between a high of $324.79 and a low of $278.10, a band of more than ten percent in twenty-four hours.
The all-time high was $757.60 in March 2024, and the price now sits about 58% below it. The all-time low was $30.83 in May 2023.
Looking at daily values over the past year, the highest was $497.8 in November 2025 and the lowest was $152.2 in February 2026: more than a threefold gap inside twelve months.
Those figures describe the past only. A market that has covered a threefold range in a year can do so again in either direction, and nothing in the data says which.
What to watch from here
Sources of demand
The subnet slots are full, and registering a new one means competing on the registration cost. The need for TAO to win a slot is demand built into the rules rather than a matter of sentiment.
Issuance was cut in half in December 2025. With the same demand, fewer newly issued coins now reach the market each day than before the halving.
Who stands behind it
Development is led by the Opentensor Foundation, and the distribution split and halving rules are written in the official documentation. The chain itself can be read by anyone.
Turnover across all venues was about $669 million over twenty-four hours, close to a fifth of the market capitalisation changing hands within a day.
Points of concern
- The price moves widely: about 58% below its peak, with a threefold range within a single year
- Subnet quality varies, and it is hard to tell from outside which ones will last
- Tokens with similar names exist and are easy to confuse with TAO (see the next section)
- Rules covering AI and tokens are still being written in many countries
Where TAO is listed (public data)
On 22 September 2026, neither OKJ (47 markets) nor bitbank (62 pairs) listed TAO. Other exchanges in Japan were not checked for this article.
Hyperliquid lists a TAO perpetual contract, but not the TAO token itself on its spot market. A perpetual tracks the price; it does not deliver the TAO token itself.
At around 20:30 JST on the same day, the public figures were: maximum leverage 5x, 24-hour turnover about $104.6 million, and open interest about $94.1 million. On the same venue BTC allows 40x and ETH 25x.
The funding rate was +0.00489% per hour, about 0.12% per day, paid by buyers to sellers.
With leverage, a small move against a position can take the entire margin.
Holding TAO
When sending, choose a wallet that supports the TAO network, and send a small test amount before the full transfer.
This section records public figures. We do not recommend trading TAO or using any particular exchange.
Frequently asked questions
Q. Can TAO be bought on a Japanese exchange?
Listings differ from exchange to exchange and can change. Of the two checked in this article, neither OKJ nor bitbank listed TAO as of 22 September 2026.
Q. Are TAO1 or HTAO on Hyperliquid the same as TAO?
No, they are different tokens. On 22 September TAO1 traded at $0.0607 and HTAO at $0.7135, orders of magnitude away from TAO at around $315. Check the full name and price, not just the ticker.
Q. What is an alpha token?
Each subnet issues its own token, called alpha. Staking TAO into a subnet exchanges it for that subnet’s alpha, and the exchange ratio moves with how much has been staked into the pool.
In summary
- TAO is the token of Bittensor, a network where AI work is produced and scored, ranked 33rd by market capitalisation
- There are 128 subnets, which is the current limit, and issuance was halved in December 2025
- The price sits about 58% below its peak and covered a threefold range within one year
- Neither OKJ nor bitbank lists it, and similarly named tokens are easy to mistake for it
💡 Comparing TAO with other markets
The markets screen in Hype Trade lists Hyperliquid markets by turnover and price move, so TAO can be compared against the rest on the same scale.
The funding plan table works out a position size from the amount you are prepared to lose on one trade, before any order is placed.
(Based on public CoinGecko, Hyperliquid and Bittensor chain data and the official Bittensor documentation, checked on 22 September 2026 at around 20:30 JST. Not a recommendation to buy or sell; prices move sharply.)
Categories