The harshest thing said about Cardano, in my years of following it, is “1 TPS” — painfully slow. Its long-awaited answer is called Leios, which claims to push throughput up by tens to hundreds of times without lowering security by an inch. Its public testnet went live in June 2026, with mainnet targeted for the end of the year.

Let me explain Leios in plain language: what it actually changes, and why “faster without getting flimsier” holds up.

Why scaling is hard

Every public chain wants to be fast, secure, and decentralized at once — but the three pull against each other. That’s the blockchain trilemma. Push one direction and the other two strain.

The usual on-chain scaling paths each carry a bill. Bigger blocks mean every node has to sync more data, ordinary computers can’t keep up, node count drops, and decentralization suffers. Layer 2 pushes complexity up a level and trust models get murkier. Sharding splits one chain into many, with tricky cross-shard coordination and new attack surface. None of the three keeps all three properties intact.

Leios takes a fourth path: don’t stretch a single block — split “packing transactions” and “deciding the order” into separate pipelines.

Today’s way: one block both loads the truck and takes a number

Look at how Cardano (and most chains) work today: every block both carries transactions and decides their order. It’s a single lane — no matter how many cars, they line up one at a time.

flowchart TB
    subgraph 传统[Traditional Ouroboros]
        T1[One block<br/>packs and orders] --> T2[A single lane<br/>everything queues]
    end
    subgraph Leios[Leios]
        L1[Input blocks pack in parallel] --> L2[Endorser blocks vote by stake] --> L3[Ranking blocks settle the order]
    end

Leios: an assembly line

Leios splits the single lane into three stages, each with its own block type. Input Blocks (IBs) only “pack transactions”: multiple stake pools can produce them at the same time, no queueing — like several waiters taking orders simultaneously. Endorser Blocks (EBs) are stakeholder votes weighted by stake, each one confirming “I’ve seen these input blocks” — a checkmark on every order ticket. Ranking Blocks (RBs) form a linear chain that, weighted by stake, settles the final order of the confirmed input blocks — the kitchen plating dishes in a fixed sequence.

flowchart TB
    A[Stake pools produce input blocks in parallel<br/>roughly every 0.2-2 seconds] --> B[Stakeholders vote via endorser blocks<br/>weighted by stake]
    B --> C[Ranking blocks settle a linear order<br/>weighted by stake]
    C --> D[Final ledger: order is clear<br/>and every step carries stake endorsement]

The core idea in one line: throughput goes to the parallel input blocks; trust stays with the linear ranking blocks.

Why security doesn’t drop an inch

The claim I most wanted to verify was “security doesn’t drop.” After studying it, my take: the design’s core idea holds up, and it has papers behind it.

Leios isn’t pulled out of thin air. It comes from the research paper High-Throughput Blockchain Consensus under Realistic Network Assumptions, plus the implementation spec CIP-0164. Three points matter.

First, every layer has stake endorsement: input blocks must be confirmed by stakeholders’ votes, and ranking blocks themselves are produced weighted by stake. Messing with order or injecting fake blocks still requires controlling the corresponding share of stake — parallelism doesn’t relax the security model. Second, the order is still linear: for a settlement layer, “who goes first” must be unambiguous, so Leios parallelizes only the “loading” stage; ordering remains a single clear chain. Third, the details are optimized: CIP-0164 shrank each block certificate from about 8 KB to about 200 bytes — certificates must be broadcast and verified network-wide, so smaller directly supports higher throughput. That kind of detail work is engineering taste for knowing where the bottleneck is.

Where things stand

The timeline is concrete. On June 23, 2026, the Musashi Dojo public testnet went live; developers and small stake pool operators (SPOs) can already try it. Alongside it, Piranha, an adversarial testing tool, simulates protocol attacks against the implementation. Mainnet still needs a hard fork — it arrives with the upcoming Dijkstra era — with official guidance targeting the end of 2026.

A dose of cold water

After the hype, the honest part. Leios’s “thousands of TPS” figures are mostly simulations; reality depends on transaction size and network conditions, and “tens to hundreds of times” is already a good outcome. Also, the Cardano community has been led on by “always the next version” for years. This time the testnet is real, but until mainnet ships, everything is still on the way.

Looking further out, Leios matters less for a pretty TPS number and more for proving a third scaling route works: push throughput up without shrinking the node set (decentralization) or lowering stake backing (security). For me, that means one thing — once it ships and runs stable, the “1 TPS” label thrown around for years can finally retire.

Article Link: