# Solana Begins Public Testing of Alpenglow, Replacing Its Consensus Protocol

Votor removes the long chain of on-chain votes in favor of direct validator voting, with Anza estimating median finality near 150 milliseconds against 12.8 seconds today.

- Published: 2026-09-23T11:09:57.416Z
- Canonical: https://polylog.news/crypto/2026-09-23/solana-begins-public-testing-of-alpenglow-replacing-its-cons
- Publisher: Polylog (Crypto desk)
- Section: crypto
- Sources: [CoinDesk](https://www.coindesk.com/tech/2026/09/23/solana-starts-testing-upgrade-that-could-cut-finality-from-12-8-seconds-to-150-milliseconds), [CryptoSlate](https://cryptoslate.com/solana-flips-ethereum-in-fees-while-eth-holds-the-burn-lead/)

Solana's Alpenglow upgrade [moved onto the public testnet on Tuesday](https://www.coindesk.com/tech/2026/09/23/solana-starts-testing-upgrade-that-could-cut-finality-from-12-8-seconds-to-150-milliseconds), beginning live testing of the largest change to the network's consensus rules since launch. Alpenglow retires TowerBFT and installs Votor, under which validators send votes directly to one another and can settle a block after one or two rounds instead of writing a long sequence of votes to the chain.

Anza, the software firm that maintains the Agave validator client, estimates median finality of about 150 milliseconds, against roughly 12.8 seconds under the current design. Earlier simulations produced figures closer to 100 milliseconds under favorable network conditions. Finality is the point at which a payment can be treated as irreversible, and it is the number exchanges, bridges and merchants actually use to price their risk.

Two limits matter for reading the announcement. Rotor, the planned replacement for Turbine, is deferred to a separate proposal, so block propagation is unchanged for now. The initial testnet migration requires Agave 4.3 and does not support Firedancer or Frankendancer, the alternative clients that exist precisely to reduce dependence on a single piece of software. A mainnet software update expected on Sept. 28 is tentative and is not an Alpenglow activation date.

The upgrade arrives as Solana's usage metrics compete directly with Ethereum's. CryptoSlate, working from DefiLlama's Sept. 22 snapshot, reported that [Solana generated more fees than Ethereum while Ethereum retained the lead in fee burn](https://cryptoslate.com/solana-flips-ethereum-in-fees-while-eth-holds-the-burn-lead/), a reminder that fee revenue, staking receipts and value returned to holders are three different measures. Capital still sits overwhelmingly on Ethereum, which holds $54.28 billion in total value locked against Solana's $6.51 billion, according to DefiLlama.

The remaining question is what the speed gain costs. Direct validator voting concentrates correctness in the client software and the vote-aggregation path rather than in a long on-chain record, so a consensus bug becomes harder to reconstruct after the fact.

## What this means

Time to irreversibility is working capital. Exchanges, bridges and payment processors hold balances idle through confirmation windows, and cutting that window from seconds to fractions of a second reduces the float they must fund and the inventory market makers must post on both sides of a transfer. The exposure runs the other way too: if Alpenglow ships before Firedancer support lands, Solana's settlement assurance rests on one client implementation, so a defect in that client becomes a network-wide event rather than a partial outage.

## What to watch

- Whether Firedancer and Frankendancer gain Alpenglow support before validators vote on mainnet activation, which decides if the upgrade arrives with client diversity or without it.
- Whether large exchanges and bridges shorten Solana deposit confirmation requirements after activation, the clearest evidence that the latency gain is being priced into real operations.
- Whether the Sept. 28 mainnet release stays a routine update or becomes an activation vehicle, which would compress the testing period.