Solana's Alpenglow upgrade is set to activate a new consensus protocol on the blockchain's Mainnet Beta on October 21, with finality expected to fall from roughly 12.8 seconds to about 150 milliseconds. The main network, however, still has to prove that target in practice.

Solana is a layer-1 blockchain on which validators produce blocks and confirm them by voting. The consensus protocol determines when a transaction counts as irreversible — a task handled until now by a mechanism called TowerBFT. Alpenglow premiered in May 2025 at the Solana Accelerate conference. The concept was developed under the leadership of Roger Wattenhofer of ETH Zurich, together with his former doctoral students Kobi Sliwinski and Quentin Kniep, who now work at Solana developer Anza. In September 2025, validators approved the proposal SIMD-0326 with 98.27% of the votes cast. Its core component, Votor, replaces the existing TowerBFT.

Measured against today's figure, the 150 ms target equates to finality roughly 85 times faster. The 150 milliseconds refer to the time to finality, not the block time: transaction execution, the Solana Virtual Machine (SVM) and fees remain unchanged.

Votor replaces TowerBFT in validator voting

Under TowerBFT, validator votes run through the blockchain as ordinary transactions. Votor changes precisely this. Going forward, validators will exchange their votes directly with one another and bundle them into BLS certificates, so a single certificate combines the approval of many validators in one record. Votes will consequently no longer appear as on-chain transactions inside blocks. The speedup therefore comes from the new voting process, not from faster blocks.

Votor has two paths to finality. If at least 80% of stake certifies a block in the first round, the block is final after that round — a fast path expected to take about 100 milliseconds. Up to one fifth of stake can stay absent without blocking this route. By contrast, if round 1 approval falls between 60% and 80%, the slower path takes over at about 150 milliseconds.

Solana itself describes the 150 milliseconds on its official upgrade page as an expectation, not a constant. Moreover, halving the slot time from 400 to 200 milliseconds is not part of Alpenglow; the network completed this separate change earlier, in four stages through October 2026. For exchanges, market makers and payment providers, finality is the key metric, since it determines when a transaction counts as settled. Strong results from test environments therefore do not replace proof under real mainnet load.

Alpenglow migration in epoch 1064 requires tight coordination

The switch begins with the start of epoch 1064 on October 21, when a feature gate activates the new logic. TowerBFT first produces another 5,000 slots — about 18 minutes — before Alpenglow takes over at slot 459,653,000. During this transition, the network briefly stops including user transactions in blocks. Votor only starts once validators holding at least 82% of stake have signed a genesis certificate, making this a one-time, coordinated intervention in a live network.

The question of validator clients is delicate. The migration excludes the alternative clients Firedancer and Frankendancer for now, so their operators must temporarily switch to software based on Agave 4.3 before epoch 1064. Options include Agave itself, Jito-Solana and Harmonic Salsa, with Solana specifying version 4.3.2. Any validator that fails to switch in time will be missing from the genesis certificate — and the more stake misses the switch, the thinner the buffer above the 82% threshold becomes. How much stake ends up in that first certificate will therefore be one of the switchover's earliest visible outcomes.

Nevertheless, the path to activation proceeded in orderly stages. Since July 2026, two preliminary steps have been running on mainnet: BLS key management in the vote account (SIMD-0387) and the Validator Admission Ticket (SIMD-0357). Anza launched a community test cluster in May 2026, followed by a public testnet in September and devnet one day later. In mid-October, the development team announced the mainnet date.

Validators pay 0.8 SOL per epoch instead of vote fees

With vote transactions gone, their fees disappear too. Their place is taken by the Validator Admission Ticket (VAT), whose technical foundation has been live on mainnet since July 2026. It costs 0.8 SOL per epoch, which the network burns, removing the tokens from circulation. Additionally, the vote account balance must sit at least 0.8 SOL above the rent-exempt minimum. The fee first falls due at the start of epoch 1064, covering epoch 1065; validators that do not pay can neither vote nor produce blocks in the following epoch. The first charging cycle will therefore show, in practice, how many operators fund their vote accounts on schedule.

For operators, this shifts the cost structure. Until now, every single vote required an on-chain transaction and with it ongoing fees — costs that arose per validator, largely independent of stake. If the new fixed fee is lower than the previous vote fees, smaller validators benefit most, since fixed costs of this kind weigh comparatively heavily relative to their stake.

A statistical side effect is also lost. Vote transactions previously counted toward the reported transactions-per-second (TPS) figure, so the reported TPS may drop after the upgrade. This is a counting effect, not a network problem.

Lower Byzantine tolerance, open questions on latency and MEV

The security model changes as well. TowerBFT can withstand up to 33% Byzantine stake, meaning stake that acts maliciously. Votor lowers this pure Byzantine tolerance to 20%, while additionally tolerating 20% of stake being offline — the so-called 20+20 model. Overall tolerance to disruptions therefore rises, while tolerance to active attacks falls.

The industry has also raised specific objections. Blockchain project Sei sees pressure toward centralization, arguing that a signal needs about 80 to 100 milliseconds for a round trip across the Atlantic — a stretch that alone exceeds half of the target value. Infrastructure provider Helius points to open questions about Multiple Concurrent Leaders and MEV in off-chain voting, and penalties for equivocation — sending conflicting blocks or votes for the same slot — also remain unresolved. Because these points concern the protocol's design rather than the migration's mechanics, they stay open even after Alpenglow is live.

This activation leaves out some components. Rotor, the planned replacement for the Turbine data distribution layer, belongs only to the second phase and still has no date. Later, Agave 4.4 is expected to bring a Fast Leader Handoff, which Anza anticipates in fewer than 1% of slots. Ultimately, only measurement data after epoch 1064 will show whether Votor reaches 150 milliseconds in daily operation.