Bittensor (TAO): 2,500 slots per subnet instead of 256, and what TAO stakers need to know

Bittensor has changed the rule by which a subnet distributes its rewards. Since the release of Runtime 475 on October 7, 2026, the owners of a subnet may choose between two procedures: the existing Yuma Consensus and a new mode called Null. Under Null, no vote among many validators decides the weights any more; exactly one validator does, namely the one with the largest stake. At the same time the number of slots per subnet rises from 256 to as many as 2,500.

For holders of TAO this is no footnote from a developer channel. Anyone delegating their tokens to a validator receives their dividend on a different calculation. And anyone wanting to enter a subnet as a miner no longer needs TAO to do it under certain conditions. The price stands at $273.87 on Monday, October 19, 2026, or 244.29 euros, which is 7.3 percent below the level of a week ago.

Null Consensus: the largest stake scores alone

A subnet at Bittensor is a market of its own for one kind of computation, for language models or for image processing, say. Miners deliver results, validators score them, and the distribution of the emission arises from those scores. The score is called a weight in the protocol, and whoever may set weights thereby decides about money.

This is exactly where Null takes hold. The mode grants precisely one validator permit, to the identifier with the largest unrounded stake; in a tie it goes to the lower identifier. Only that one identifier may set, commit or reveal weights, and that also covers the routes by which an owner scores itself. It is set out that way in the corresponding pull request of the Subtensor project, which was merged into the main branch on October 6, 2026.

The consequence is a markedly shorter calculation. Under Null an epoch loads only the single weight row, skips the consensus and skips the bonds, that is the accounting of trust over time. The frozen bond stores are retained but no longer updated. If the chosen validator supplies no valid weights, the protocol distributes the miner rewards evenly across all registered identifiers.

Yuma Consensus: what the existing procedure delivers

Yuma Consensus is the procedure Bittensor has used for years to score the participants of a subnet and fix their rewards. It takes the weights of all eligible validators, forms a stake-weighted consensus from them and penalises outliers: whoever deviates sharply from the majority view loses influence. The Bittensor protocol documentation still describes it as the procedure that translates weights into rewards.

That mechanism costs computing time, and it costs it quadratically in the number of validators. Hence a subnet's capacity under Yuma sits at 256 identifiers. Null buys the higher capacity by leaving out the most expensive part of the calculation. That is a trade, not an improvement on every axis: the deviation penalty that reins in a single validator under Yuma falls away with the consensus.

2,500 slots, divided by the number of mechanisms

The 2,500 slots are not a fixed figure per subnet but a shared budget. It is divided by the number of emission mechanisms: with one mechanism 2,500 identifiers are available, with two it is 1,250, with four 625. An owner can raise the capacity only after switching to Null. The project's tests draw the line sharply; 2,500 identifiers are accepted, 2,501 rejected.

The way back is more laborious than the way out. Anyone switching from Null back to Yuma has to prune their subnet explicitly to the shared ceiling of 256 identifiers beforehand. That pruning deletes at most 64 identifiers per operation, aborts commitments already running and clears orphaned entries in limited steps. A subnet with 2,000 participants therefore does not come back in a single step.

The upshot for the projects: Null is built for large subnets with a division of labour, in which many miners deliver a clearly measurable service and the scoring leaves little room for argument. For small subnets with contested scores, Yuma remains the more robust procedure.

Dividends under Null: inactive identifiers are served too

Here lies the point that concerns delegators directly. Although only one identifier may set weights, all identifiers with a positive stake keep their pro-rata dividends, expressly including the inactive ones and those without a validator permit. The permission to score and the entitlement to a dividend are decoupled under Null.

Anyone who has delegated TAO to a smaller validator therefore does not lose their entitlement through the switch. What changes is influence: in a Null subnet the validator you follow no longer has a say in how the miners are scored, as long as it is not the largest. The payout then depends solely on the stake share, no longer on the quality of the scoring.

The allocation itself is computed by the protocol in whole units with wide integers, and it never exceeds the funded epoch budget. That sounds technical but has a practical side: rounding remainders that arise with many thousands of recipients do not run out of the budget.

Fee-free PoW registration: computing work instead of a TAO burn

The second part of the upgrade is the more tangible one for newcomers. Until now, entering a subnet as a miner cost a registration fee, the so-called burn: you destroy TAO to obtain a slot. The upgrade places a second route alongside it, one that the owners of a subnet can enable. Through the pow_register call, a slot arises through computing work, a proof of work.

A direct call without a tip requires neither a TAO transaction fee nor the registration burn nor the prior purchase of a security deposit. The root subnet is excluded from this. New or missing settings default to off, and expressly stored enablements are retained. The old registration routes keep their burn behaviour unchanged.

The proof is tied to the slot and not transferable. It binds the subnet, a recent historical block hash, the hotkey, the signing coldkey and a nonce. It is accepted at an age of one to five blocks. If the search takes longer, the command line fetches the latest block hash and the difficulty anchored to it every twelve seconds. A proof that has been found and then sits too long is rejected.

Burn price and PoW difficulty rise separately

The two routes do not affect each other in price. Registrations via computing work raise only the PoW difficulty, registrations via the burn raise only the burn price. Both draw on the same settings for multiplier and half-life, but they have their own price limits and their own demand.

A subnet owner can offer the burn alone, computing work alone, or both. The protocol refuses to switch off whichever route is the last one remaining; so a subnet cannot accidentally close itself to new participants. The existing rules continue to apply to the owner's cool-down and lock-up periods.

Economically this shifts entry from capital to hardware. Whoever has graphics cards and electricity pays with computing work. Whoever does not keeps paying with TAO. For the price the direction is ambiguous: less burn means less demand for TAO for registrations, more participants mean more demand for stake. Which effect prevails will only show once enough subnets have enabled the route. Anyone looking for an exchange with a European licence for that will find one in the crypto exchange comparison.

Runtime 475 to 478: how a Bittensor upgrade goes live

Between October 5 and October 13, 2026 the project published five runtime versions: Runtime 473 on October 5, Runtime 475 on October 7, a proposal for 476 and Runtime 477 on October 8, and Runtime 478 on October 13. Null Consensus and the PoW registration sit in Runtime 475; the pull request itself carries specification number 474, which received no release entry of its own. cryptoticker.io compiled this review itself on October 19, 2026, on the basis of the eight most recent release entries of the Subtensor project and the pull request named.

The interim entry for 476 is interesting because it lays open the route by which a new runtime lands on the Finney mainnet. The entry names a deployment via a two-of-two multi-signature: the build automation sets its half, after which the three-member panel has to produce the second approval. Alongside that it carries the commit, the call hash, the checksum of the runtime file and a deployment time, so that anyone can check the published file against a rebuild of their own.

For holders that is a useful yardstick. An entry with the addition “proposed” is code waiting for signatures, not a change already in effect. Anyone wanting to know whether an announced change has arrived in the network looks at the specification number the network reports, not at the headline.

What stays open in the pull request

The upgrade is merged and shipped, and it carries two open points in its own record. The project's owner expressly admitted a single review instance named Skeptic as a non-green check; its five findings on resource accounting remain unresolved and visible. All other checks had to pass.

Plainer still is the sentence at the end of the rationale: the accounting of the reference weights and the resource envelope of the deployment remain unchecked, and a runtime update on a shared testnet had not been submitted at that point. The owner settled on 2,500 identifiers with atomic payouts and deferred further optimisations of the reward accounting. Historical comparative costs for 4,096 identifiers are noted in the project as a historical measurement, not as a reason for the smaller capacity.

Our assessment: the capacity of 2,500 slots is documented and tested, operation at that size is not, according to the project's own record. In favour are the passed boundary tests and the successful registration without a balance; against are the five open findings on resource accounting and the missing testnet run. Anyone holding TAO should therefore read the first wave of large Null subnets as a stress test and not as a settled matter. No recommendation to buy follows from that; with TAO a total loss is possible.

TAO price week on week: $273.87, with $269 as the weekly low

The price stood at $273.87 on Monday, October 19, 2026 at around 9:50 a.m. German time, according to CoinGecko data. Over 24 hours that is 1.5 percent more, over seven days 7.3 percent less and over 30 days 6.5 percent more. The weekly high of the daily values was $295.17 on October 12, the weekly low $269.16 on October 18. Within the past 24 hours the range ran from $254.35 to $275.54.

Market value comes to $3.11 billion, which corresponds to rank 35. In circulation are 11,339,646 TAO out of a maximum of 21 million, so 54 percent. Daily turnover is $162 million and therefore a good 5 percent of market value; in that respect TAO is considerably more liquid than many tokens of comparable size. The price is 63.9 percent away from the peak of $757.60 of March 7, 2024.

As levels on the downside, the weekly low at $269 and the round mark at $250 suggest themselves, on the upside the weekly high at $295 and the $300 mark. That is an observation of the range the price has moved in over the past seven days, and not a price target. The hard ceiling of 21 million tokens makes TAO scarcer than most protocol tokens; release schedules and the running emission determine how quickly the remainder reaches circulation.

Buying TAO in Germany: the route via a licensed exchange

TAO is not a token every German platform carries. Since the end of the MiCA transition period, only licensed providers may supply crypto-asset services in Germany, and the range of thinner altcoins differs markedly from provider to provider. Before buying, first check whether your provider trades TAO at all and whether it holds a licence under the European crypto regulation; the list of licensed institutions is kept by the supervisor, not by the exchange itself.

On custody the same arithmetic applies to TAO as to other protocol tokens. If the holding sits on the exchange, you carry the provider's default risk. Anyone holding it themselves takes on responsibility for the keys. Delegated staking is the normal case at Bittensor and not a side business, because stake decides influence in the network. Anyone delegating through an exchange should establish beforehand who holds the keys in that case and what share of the reward stays with the provider.

For tax, the one-year holding period remains unchanged for 2026: anyone holding TAO for longer than twelve months and then selling pays no tax on the gain. Staking income is to be kept separate from that and runs as other income. With several exchanges and wallets the allocation of acquisition dates quickly becomes confusing; how the order counts there is set out in our piece on the FIFO method across several exchanges.

Leveraged products on TAO exist, and with a token that runs through a range of $21 in a single day they are correspondingly risky. A position with fivefold leverage is liquidated on a counter-move of 20 percent, and those 20 percent sit inside the monthly range with TAO.

Bittensor upgrade: 2,500 slots change the arithmetic for stakers

Three steps make sense now:

1. Check which mode the subnets holding your stake run in. The mode is a setting of the owner, and it determines whether your validator's scoring still counts at all. If your stake sits with a smaller validator in a Null subnet, the dividend remains, the influence does not. If you want to change provider, compare the terms in the staking comparison first.
2. Settle your buying route before you top up. TAO is not traded at every licensed provider, and fees for smaller altcoins are frequently above those for Bitcoin. Which institutions hold a European licence is set out in the comparison of regulated exchanges.
3. Keep the holding period and staking income apart in your records. The disposal gain after twelve months and the running staking income are treated differently. Which tools keep the two separate is set out in the comparison of tax tools.

(As of October 19, 2026. This article is not investment advice. Prices and fee structures change; check the terms with the provider before you buy.)