Namespace
← Back to Blog
newssolanaalpenglowconsensusweb3

Solana Alpenglow upgrade targets 150ms finality

October 25, 2026

Anza, the engineering firm that maintains Solana's Agave validator client, activated Alpenglow on mainnet on October 21, 2026, according to Anza and the Solana Foundation. The upgrade retires TowerBFT, the consensus system the network had used since March 2020, and is designed to reduce the time until a transaction can no longer be reversed from about 12.8 seconds to roughly 150 milliseconds. On October 25, Solana's official account listed the change among recent network milestones and described it as the network's biggest consensus upgrade yet.

What happened

Anza said validators completed the switch at slot 459,653,000 at about 14:00 UTC on October 21, The Block reported. The Solana Foundation places the start of epoch 1064 at approximately 13:36 UTC that day, with Alpenglow consensus beginning at slot 459,653,000 after TowerBFT ran for the first 5,000 slots of the epoch. The Foundation says the preceding slot, 459,652,999, was certified with votes from 446 of 654 validators.

The component now live is Votor, shipped in Agave 4.3. Under TowerBFT, validators recorded votes as on-chain transactions, and a block became final only after those votes accumulated over 32 slots. Votor moves voting off the chain: validators send votes directly to one another and bundle them into certificates. The Foundation's technical notes say a block can finalize after one round if at least 80% of stake votes to notarize it. If that threshold is missed, a second round can finalize the block using 60% stake thresholds.

Anza chief executive Brennan Watt called the work "the largest, most ambitious blockchain upgrade since 'The Merge,'" referring to Ethereum's 2022 move to proof of stake, according to The Block. Anza said the design, published in May 2025, went through public testing before mainnet. The 150-millisecond figure is a target. Anza says the protocol is designed to deliver finality in roughly 150 milliseconds, and the Foundation lists it as target finality, not a measured time for every transaction. Slot time is separate: the Foundation says slots were cut from 400 milliseconds to 200 milliseconds in stages that finished in October.

Why it matters

Finality is what an exchange, a merchant, or a payment service typically waits for before treating a transfer as settled. Solana already provided a faster conditional confirmation; Alpenglow is meant to compress the irreversible guarantee, according to Anza. For people using wallets and applications, the Foundation says nothing about signing or sending a transaction changed. The Solana Virtual Machine, transaction formats, programs, and fees are untouched.

To keep voting and producing blocks, validators must run version 4.3.2 of an Agave-based client — Agave, Jito-Solana, or Harmonic Salsa — and hold at least 0.8 SOL above the rent-exempt minimum in the vote account to pay a Validator Admission Ticket each epoch, the Foundation says. Firedancer and Frankendancer were not supported for the migration. Vote transactions no longer appear in blocks, so published transaction counts from October 21 onward are not comparable with earlier totals. The protocol is specified to tolerate 20% of stake acting maliciously and another 20% offline at the same time, and the older confirmed and finalized commitment levels now describe the same state, according to the Foundation.

What's next

Alpenglow is shipping in phases. Votor is the piece on mainnet. Rotor, which would replace Turbine as the system that spreads block data across the network, is planned for a later release and is not yet scheduled, according to the Solana Foundation. Anza has also pointed to geographically aware leader rotations, meant to shorten handoffs between block producers, and to a research design called Quantumglow that would swap elliptic-curve cryptography for hash-based signatures while aiming to keep the same finality target. Neither has a committed date.

Developers who only send transactions and read account state do not need a migration, the Foundation says. Teams that stream blocks, index throughput, or read time from ledger ticks do, because a slot can carry more than one candidate block and Proof of History no longer paces production. Whether measured mainnet finality stays near the stated target will depend on network conditions.

Sources

More from the Namespace blog