Solana’s planned consensus rewrite has moved from a closed rehearsal environment onto the network’s public test clusters, a step core engineers have treated as the start of the formal activation path. Anza, the firm that maintains the Agave validator client, said Alpenglow completed its testnet handoff on September 24, 2026 at slot 444,625,255.
The same migration finished on devnet on September 25 at slot 504,148,999.
In both cases an Alpenglow genesis block formed,
TowerBFT was retired on that cluster, and Votor began finalizing blocks.
Validator votes on those networks now travel off-chain rather than as ordinary transactions packed into blocks.
The cutover followed a script Anza had already described for every cluster. A feature gate opens at an epoch boundary.
A 5,000-slot countdown then runs, lasting about 17 minutes.
At the boundary, validators confirm a last TowerBFT block with 82 percent of stake, after which Votor takes over.
Anza’s testnet note was blunt about sequencing: devnet next, then mainnet-beta only after an observation period.
No live-network date was attached to that last step.
What changes is the meaning of “final.” TowerBFT has long required a long confirmation chain, producing irreversible settlement on the order of 12.8 seconds.
Alpenglow is officially running on testnet ⛰️
The handoff completed at slot 444625255: Alpenglow genesis block formed and TowerBFT is retired.
Testnet now finalizes with Votor and votes have entirely moved off-chain.
Devnet is next. Mainnet-beta after our observation period.
— Anza (@anza_xyz) September 24, 2026
Votor is designed to collapse that wait into a certificate.
One path aims for roughly 100 milliseconds when a large stake share agrees in a single round; a second path aims for about 150 milliseconds if another round is needed.
Alpenglow is officially running on devnet ⛰️🎉
The handoff completed at slot 504148999: Alpenglow genesis block formed and TowerBFT is retired.
App developers: test your programs and integrations now during our observation period leading up to mainnet-beta migration. pic.twitter.com/lNgC0Y4xoF
— Anza (@anza_xyz) September 25, 2026
The Solana Foundation’s own upgrade materials describe the 150-millisecond figure as an expectation measured against a given stake map, not a constant that will hold in every geography or under every load. Execution, programs, and fees are intended to stay the same.
You are not ready for this https://t.co/j8gdB8NexI
— toly 🇺🇸 (@toly) September 25, 2026
A later piece of the design, Rotor, is still scheduled separately and does not change how blocks propagate in this first phase.
That is why the public activations matter to application teams even before any mainnet switch.
Devnet is where most programs are exercised with valueless tokens.
Anza told developers to test integrations during the observation window rather than wait for production.
After the testnet flip, Solana Foundation engineer Ilan Gitter said the handoff looked clean and cited more than 25,000 transactions per second with 150-millisecond finality on that cluster—figures from a test environment, not from the chain that holds user funds.
Solana co-founder Anatoly Yakovenko answered Anza’s devnet countdown with five words:
“You are not ready for this.”
The line reads as both a boast and a warning.
Millisecond settlement would change how wallets, exchanges, bridges, and liquidation engines treat confirmation.
It would also surface assumptions that only appear when real balances, real latency, and a full validator set are in play.
Alternative clients such as Firedancer and Frankendancer still lacked Votor support at the moment of the public migrations, so the first clusters leaned on Agave.
Those speeds remain largely unproven under full mainnet conditions.
The immediate work is observation.
Operators focused on Solana (SOL) will watch certificate behavior, client diversity, and whether “confirmed” and “finalized” can safely collapse into one state. Only after that window will Anza say whether the same handoff is ready for the network that actually settles value.