On Friday, September 18, 2026, at around 05:01 UTC, the Solana network tightens its own beat: a slot will then last 250 milliseconds instead of 300. The switch for it is already set and takes
On Friday, September 18, 2026, at around 05:01 UTC, the Solana network tightens its own beat: a slot will then last 250 milliseconds instead of 300. The switch for it is already set and takes effect automatically at the boundary into epoch 1037. For you as a holder of Solana, that mainly means three things: delegations and unstaking take effect faster, staking rewards arrive more often and in smaller portions, and the window in which a signed transaction stays valid shrinks to roughly 38 seconds. If your SOL simply sits on an exchange, you need do nothing. If you sign offline, delegate, or run a validator yourself, read this to the end.
The change is no surprise but the third step of a roadmap that began in August. The only thing new is that the 250 millisecond step now has a concrete date. Solana's official overview page still listed this step without a mainnet date when we checked on September 17, while the development team Anza had already flagged the activation as imminent on September 16. That gap between announcement and documentation is why the figure has barely surfaced so far.
A single number in the protocol is being changed: the target time for a slot. That value drops from 300 to 250 milliseconds. The number looks small, but almost everything in the network that looks like time hangs off it. Epoch length follows it, the validity period of a blockhash follows it, and the ceilings for compute per block follow it too.
The improvement proposal responsible is SIMD-0525. It describes a staircase from 400 through 350, 300 and 250 milliseconds to a target of 200 milliseconds. Two steps have already been taken on mainnet: the move from 400 to 350 milliseconds on August 19, 2026, at the start of epoch 1019, and the move from 350 to 300 milliseconds on August 25, 2026, at the start of epoch 1023. The third step is the one at issue here. A fourth is still to come after it.
What explicitly does not happen: there is no token swap, no migration, no deadline by which you would have to pull something out of a contract, and no confirmation you would have to grant in your wallet. Anyone asking you to approve a Solana upgrade wants something from you other than your consent to a protocol parameter.
A slot is the time window in which exactly one validator may produce a block. Once it expires, the next one is up, whether the previous one delivered or not. Slot time is therefore not a speed limit for transactions but the clock rate at which the network puts out its blocks.
An epoch is Solana's accounting period and covers a fixed 432,000 slots. At its boundary the network does the bookkeeping: delegations take effect, rewards are settled, prepared protocol switches are armed. Because the number of slots is fixed while the duration of a slot falls, the epoch gets shorter. At 400 milliseconds it worked out at 48 hours, at 300 milliseconds it is 36 hours, at 250 milliseconds it is 30 hours. At the final target of 200 milliseconds it lands at 24 hours.
This is the point at which the figures in circulation deserve a look: several reports give an epoch length of 24 hours for Friday's change. That is the value for 200 milliseconds, meaning the step after this one. For 250 milliseconds the same calculation gives 30 hours, and the official formula based on 432,000 slots arrives at that value too.
A feature gate is a switch in the validator code that keeps an already shipped function dormant until enough stake weight is behind it. When it flips, it never does so mid-epoch but at an epoch boundary. For this change the chain ran across three epochs: in epoch 1035 the switch stood at pending, in epoch 1036 it became active at protocol level, and with the start of epoch 1037 the new timing rules apply.
For the third time since mid-August, the network's beat is being tightened. Solana has no clock. The network counts slots, and a time of day can only be estimated from slots. That is precisely why every serious announcement puts an "around" in front of the time: the boundary into epoch 1037 falls on a particular slot, and when that slot is reached depends on how fast the network actually runs in the hours beforehand.
That imprecision is why two sources can say 05:01 UTC and a third 05:06 UTC without any of them being wrong. If you want something done by the exact moment, plan with an hour of buffer. If you only want to know whether you are affected, the calendar day will do: Friday, early morning European time.
cryptoticker.io collected this analysis itself on September 17, 2026. Method: ten consecutive performance samples of 60 seconds each via Solana's public mainnet node, retrieved shortly after 00:50 UTC, plus a query of the current epoch status. That covered 1,893 slots across ten minutes of network operation.
The result: on average 317.0 milliseconds elapsed per slot, with individual samples between 312.5 and 326.1 milliseconds. The value therefore sits above the current target of 300 milliseconds. That is normal and no sign of a problem. What is measured is elapsed clock time divided by the number of slots actually filled, and every slot a validator does not serve stretches that average.
The date could be recalculated from the same query. The network stood in epoch 1036 at slot 111,540 of 432,000, leaving 320,460 slots to go. At the measured 317 milliseconds that gives 28.2 hours of remaining runtime, and therefore September 18 at around 05:06 UTC. Our own calculation confirms the announced time of 05:01 UTC to within a few minutes. The node queried reported client version 4.3.0-rc.0.
What we could not check: how large the share of validators already running the recommended version is, and whether the switch actually takes effect on Friday. Both only show up after the fact. The compute ceilings mentioned further down also rest on a specialist report rather than on the official overview page.
For anyone who has delegated SOL, epoch length is the practically most important quantity in this whole change. Delegations and their withdrawal take effect at epoch boundaries, not immediately. If the epoch shortens from 36 to 30 hours, the typical wait until a new delegation earns rewards, or until deactivated SOL is freely available again, shortens with it. How that process works in detail, and why a waiting period is not a penalty, we took apart in our article on lock-up periods and unstaking duration.
With rewards, the rhythm changes, not the amount. Payouts happen per epoch. More epochs in a year therefore mean more credits, but smaller ones. The annual yield on your delegation does not rise as a result. Anyone logging earnings by epoch gets more rows in the same spreadsheet from Friday, and anyone comparing providers by payout frequency will find the differences between the platforms in our overview of staking providers.
One common misunderstanding should be cleared up here: Friday's date has nothing to do with Alpenglow. That is a different and considerably larger overhaul of the consensus mechanism with a roadmap of its own, which we described in our article on the Alpenglow activation at the end of September. Two switches, two dates, two different sets of consequences.
Transactions signed offline have fewer seconds after the change before the blockhash expires. The blockhash is the timestamp of a Solana transaction: a reference to a recently produced block that prevents the same instruction from being submitted again later. The lifetime of that stamp is measured not in seconds but in blocks. According to Solana's documentation, 151 blockhashes are valid. The documentation still converts that using the old target time of 400 milliseconds and thus arrives at roughly 60 to 90 seconds.
That span falls with the beat. At the 317 milliseconds measured today, roughly 48 seconds remain; at 250 milliseconds, roughly 38. A specialist report on the upcoming change puts the window at 37.5 seconds, which is the same calculation using 150 blocks instead of 151. The order of magnitude is the same either way: a good half minute.
This matters wherever a human being or a device stands between creating and sending a transaction. Signing with a hardware wallet burns those seconds on unlocking, paging through the display and confirming. So does preparing a transaction, then checking the address once more at leisure, and only sending afterwards. The consequence is not a lost payment but a rejected one: the network discards a transaction with an expired blockhash, and the wallet reports an error. How to tell whether such an attempt really failed or went through after all is covered in our guide to failed Solana transactions.
In practice that means: unlock the device before sending, check the address beforehand rather than inside the running window, and do not treat the confirmation dialogue as reading time. Which devices have short confirmation paths and which send you through several menus is shown in our hardware wallet comparison.
Shorter blocks mean less compute time per block. So that throughput per second stays constant, the ceilings fall in the same proportion as the slot time. A specialist report on the change puts the block limit after the switch at 37.5 million compute units instead of the previous 45 million, and the ceiling for a single writable account at 15 million instead of 18 million. Those values were not on the official overview page on September 17, which is why we explicitly mark them as that report's figures.
For you as a user, the number matters less than its consequence. When a single account in heavy demand gets fewer compute units per block, transactions compete more tightly for the same pool. In quiet phases you will notice nothing. At peak times, say at the launch of a new token, the priority fee more often decides whether your swap makes it into the next block or waits. Anyone who has permanently set their fee ceiling in the wallet to the lowest possible value should know that setting before noticing it for the first time during a rush.
Anyone running a validator has the only real deadline in this whole business. The recommended version is Agave v4.3.0-rc.1, released on September 11, 2026. Anza has asked operators to install the update by the end of Friday, meaning by the close of the same day on which the switch takes effect.
That is not an alarm but routine. Running an older version does not immediately cut you off, but it risks missing blocks once the timing rules change. For delegators that creates a quiet task: if your validator skips a conspicuous number of slots in the days after the change, the likely reason is a version that has not been installed, and your rewards fall with it.
For the great majority of investors the honest answer is: nothing. If your SOL sits with an exchange or a broker, their operator takes care of client versions, blockhashes and epoch boundaries. You do not have to confirm an update, change an address or meet a deadline.
Two edge cases remain. First, around a change like this a provider may pause deposits and withdrawals on the Solana network for a few hours. That is a precaution taken by the individual house and will then appear in its status notice. Second, a date like this is a well-known occasion for attempted fraud. Neither Anza nor an exchange will message you asking to enable an upgrade in your wallet.
The roadmap on Solana's site still listed the move from 300 to 250 milliseconds on September 17 as active on devnet and testnet, with a mainnet date still to be determined. The two steps already taken, by contrast, are recorded there with date and epoch. Anza's announcement of September 16 is therefore more current than the documentation page describing it.
A simple rule follows for you: go by the epoch number, not by a date in an article. The number 1037 is verifiable, and any wallet with network details, as well as any public block explorer, will show you which epoch the network is currently in. If it says 1037 or higher, the switch has flipped.
Sources to read up on: the official page on reduced slot times with the staged plan, and the developer documentation on transaction confirmation with the figure of 151 valid blockhashes.
(As of September 17, 2026. This article is not investment advice. Prices and fee structures change; check the terms with the provider before you buy.)
