Cardano's Leios upgrade: how it aims to multiply throughput, and when it could land

Cardano's next big protocol change is called Leios. It is a change to Ouroboros, the consensus protocol that decides who makes the next block, and its goal is narrow: let the network carry many more transactions without weakening the security rules it runs on today. Leios is planned to arrive as part of the Dijkstra hard fork, which Intersect, the member organisation that coordinates Cardano's development, is targeting for between December 2026 and March 2027.

This post explains the problem Leios is meant to solve, how it works, what a public testnet has shown so far, how the upgrade gets to mainnet, and what is still uncertain.

The ceiling Cardano has today

On Cardano today, a stake pool wins the right to make a block roughly once every 20 seconds on average, and each block's body can hold at most 90,112 bytes, about 90 kilobytes. Multiply those two numbers and you have the ceiling: Input Output (IO), the engineering company behind much of Cardano's core software, puts it at 4.51 kilobytes of transactions per second.

Raising that ceiling is harder than it looks. Bigger or faster blocks take longer to travel across the network, and the security analysis of Ouroboros Praos, the current protocol, assumes that a block reaches honest nodes before the next one is made. Push the numbers too far and that assumption breaks.

The case for more capacity is partly economic. Stake pools are paid from fees and from a reserve of ADA that shrinks over time. CIP-164, the proposal that specifies Leios, argues that once the network sustains more than about 50 transactions per second, fee revenue can take over as the reserve runs down. IO makes the same point in its funding proposal: fees that grow with use are how pools stay viable later on.

How Leios works

Leios leaves the Praos chain as it is and adds a second, larger kind of block beside it. CIP-164 calls the normal blocks ranking blocks and the new ones endorser blocks. In plain steps:

1. A pool that wins a slot makes a normal ranking block, as today. It may also announce an endorser block: a larger block that points to many more transactions.
2. The endorser block spreads across the network.
3. A committee of stake pools, chosen by stake, checks it and votes on it within a fixed time window.
4. When enough votes are in, they are combined into a compact certificate.
5. A later ranking block includes that certificate, and the endorser block's transactions become part of the ledger.

If an endorser block does not collect enough votes in time, nothing is lost: the next ranking block can carry transactions directly, as Praos blocks do now.

The votes use BLS signatures, a kind of signature that many voters can merge into one short proof. In an October 2025 demonstration, IO reported compressing about 12 kilobytes of votes into a certificate of about 5 kilobytes in milliseconds. Pools will need to register an extra BLS voting key next to the keys they hold today.

Three timing settings carry the security argument. They set how long nodes wait to spot a pool announcing two conflicting blocks, how long the committee has to vote, and how long a certified block gets to spread. Together they are meant to make sure a certified endorser block reaches every honest node before anyone has to process the ranking block that includes it.

Why "Linear"

The version heading for mainnet is called Linear Leios. The original Leios research design had three kinds of block in a pipeline, including separate "input blocks" made in parallel. CIP-164 drops the input blocks and ties each endorser block to a ranking block. The proposal's authors describe the trade as less complexity and less disruption for the rest of the ecosystem, while keeping a substantial share of the throughput gain.

What changes for users

According to CIP-164, wallets and apps should see little change: the node hands them blocks with the endorsed transactions already filled in. Fees do not change structure. The expected difference shows up when the network is busy, when transactions should wait less to be included. That is the design intent; it has not yet been measured on mainnet.

Stake pools carry more of the change. Besides the new voting key, CIP-164 says disk, bandwidth and CPU needs rise with throughput, though it expects today's infrastructure to absorb the first step.

What testing has shown

IO opened a public Leios testnet, MusashiNet, on 23 June 2026. Its first phase, Earth, ran for 41 days and produced more than 127,000 blocks and 30,000 endorser blocks, with 63 registered pools, 40 of them producing blocks at peak. Under a synthetic load, IO measured a peak of 26.8 kilobytes of transactions per second, against 4.51 for Praos alone: about six times today's ceiling.

The testnet is planned in five phases named after Miyamoto Musashi's Book of Five Rings: Earth, Water, Fire, Wind and Void. The second, Water, is running with a rewards programme for pool operators, and a red team is attacking the prototype. IO's stated goal is "mainnet-ready Leios by the end of the year".

Simulations suggest more headroom than the testnet showed. IO reported in late 2025 that throughput rose with demand and then levelled off near 300 kilobytes per second, which it put at about 200 average-sized or 1,000 small transactions per second. These are simulation results, not mainnet figures.

Four multipliers, four meanings

Numbers for Leios get quoted loosely, so it helps to separate them:

• 6x: what the first testnet phase measured (IO).
• 2x to 30x: IO's planned mainnet rollout, starting low and raising capacity in steps through protocol parameter changes that the community has to approve (IO).
• 5x to 20x: the initial target in Intersect's Dijkstra plan, as reported by FinanceFeeds.
• 10x to 65x: the capacity of the design itself, per IO's funding proposal.

Which step of that ladder Cardano starts on is not public yet.

How it gets to mainnet

Leios ships inside the Dijkstra hard fork, which moves Cardano to protocol version 12. Intersect has split Dijkstra into two phases. The first brings Linear Leios together with Nested Transactions (transactions that can contain child transactions, each with its own signatures), a new script version, Plutus V4, and a set of smaller changes. The second, targeted for the second quarter of 2027, switches on Peras, a separate change aimed at faster settlement, through a smaller upgrade inside the same era.

Dijkstra follows the Van Rossem hard fork of 18 July 2026, the first Cardano hard fork ratified through on-chain governance. CoinDesk reported that delegated representatives (DReps) backed it with 78.97% and stake pools with 53.02%, the narrowest margin. IO said at the time that Van Rossem "lays the foundation" for Dijkstra.

The road from here, per Intersect, runs through node releases and test networks:

• Node 11.2, expected soon, carries most of Dijkstra (the new block format, Plutus V4) but not Leios, and is not ready for a hard fork.
• Node 11.3 is planned as the hard-fork release candidate and is where Leios joins.
• Each phase then goes through a hard fork on the preview test network (about two weeks of testing), then on preprod (one to two weeks), then a mainnet governance action that DReps, stake pools and the Constitutional Committee vote on, and only then the mainnet hard fork.

On 5 September Intersect gave two windows for enactment: a moderate-confidence window of 5 December to 4 January, and a high-confidence window of 24 February to 26 March. Its planning page notes that its dates are targets for finishing the code and leave out governance and community testing.

Two more items add to the workload. Intersect's Parameter Committee is reviewing about 100 new guardrails for Leios (and about 50 for Peras): limits that bound what future parameter changes may set. And Dijkstra is the first hard fork that also needs a change to Cardano's constitution; an amendment has been submitted for review.

Risks and open questions

• Timing. The schedule rests on a release candidate that does not exist yet, two test-network hard forks and three votes. The second window is there for a reason.
• Approval. Van Rossem passed among stake pools with just over half. Leios asks more of pools: new keys, more bandwidth and storage. How they vote is not known.
• Network delay. Larger blocks have to travel. In IO's October 2025 prototype demonstration, the time for a normal block to spread rose from about 0.6 seconds to about 3 seconds once endorser-block traffic was added. The timing settings exist to keep this inside safe bounds; the adversarial testnet phases are where that gets tested.
• The starting setting. A rollout from 2x upward means the first mainnet step may be modest. The higher numbers depend on later parameter votes.
• Demand. Extra capacity changes fee revenue only if people use it. Whether traffic grows into the new ceiling is something no upgrade can decide.

Leios is one of the larger changes to how Cardano makes blocks since Praos itself. The design is public, a testnet has shown a sixfold gain under load, and the route to mainnet is mapped. What is still open is when it lands, how much capacity it starts with, and how much of it gets used.