Key points

  • Solana's official upgrade page lists Alpenglow as active on testnet and devnet, while mainnet remains unactivated.
  • The upgrade replaces TowerBFT voting with direct validator votes and certificates, targeting roughly 150-millisecond finality.
  • Applications that only submit transactions or read account state should not need migration, but indexers and streaming infrastructure face data-model changes.

Solana has activated its Alpenglow consensus upgrade on both devnet and testnet, placing the redesign on the network's two public testing environments before any mainnet switch. Solana's official upgrade page lists both test clusters as active and mainnet as not activated, while CoinDesk reported that devnet changed over on September 25, one day after testnet.

Public testing expands before mainnet

The milestone gives validators, infrastructure providers and application teams a broader setting in which to observe the new consensus process. Devnet is commonly used by developers testing applications, while testnet is oriented toward network and validator testing. Running Alpenglow on both does not mean the upgrade is live for normal mainnet transactions, and Solana has not presented the roughly 150-millisecond target as a guaranteed production result.

Related reporting: Solana cuts target slot time to 250ms without raising throughput

Alpenglow changes how validators vote

Alpenglow replaces TowerBFT with Votor, a protocol in which validators send votes directly to one another and aggregate them into certificates. Under the existing design, validators submit vote transactions onchain and finality takes about 12.8 seconds. Solana says Votor can finalize a block after one or two voting rounds and targets finality of roughly 150 milliseconds. Rotor, a later phase intended to replace the Turbine data-propagation system, is planned separately and is not yet scheduled.

Execution remains unchanged

The upgrade changes consensus rather than transaction execution. Solana says the Solana Virtual Machine, transaction formats, programs and fee mechanics remain intact. Applications that only send transactions and read balances or account state should not require migration. The distinction matters because the public test activation is a test of how the network agrees on blocks, not a replacement of the application environment or a change to the way users sign transactions.

Data infrastructure must adapt

Indexers, explorers and services that stream validator data face more work. Alpenglow can expose more than one candidate bank for a slot, so downstream systems must distinguish them and promote the bank that becomes confirmed. Vote transactions also disappear from blocks because voting moves outside the transaction stream. As a result, dashboards that counted vote traffic may show lower transaction totals even when user activity has not changed. Solana advises those systems to reset baselines rather than interpret the accounting change as a sudden loss of network demand.

Testing is the next evidence point

The dual-network activation makes live test data available, but it is not proof that mainnet will consistently reach the target latency. Solana describes 150 milliseconds as an expectation shaped by validator stake distribution, network topology and whether a block finalizes in one round or two. The transition dashboard currently exposes separate views for devnet, testnet and mainnet, allowing operators to follow the rollout while the production network remains on its existing consensus system.

What comes next

The immediate focus is operational testing: validator participation, certificate formation, fork handling and the behavior of data consumers under the new model. A later mainnet activation would require a separate feature-gate decision and coordinated software adoption. Until that occurs, the verified development is limited to devnet and testnet. Performance measurements from those clusters, any defects uncovered during the transition and an explicit mainnet activation plan will be the next material milestones across both environments.

Sources

AI-generated editorial image; not a photograph of the reported event. Prepared with AI assistance and source verification.