Kaspa (KAS): a full reference on the coin and how it is mined

Kaspa is a proof-of-work network launched on 7 November 2021 with no premine, no presale and no founder allocation. The protocol grew out of the work of Yonatan Sompolinsky, author of the GHOST and PHANTOM research papers, and the first version was built by DAGLabs, a company wound down that same 2021. Since then the code has been carried by a community of developers, among them Michael Sutton, Shai Wyborski and Ori Newman.

The core idea is that Kaspa is not a chain but a directed acyclic graph of blocks, a blockDAG. Blocks found in parallel are not discarded as orphans: they join the shared graph and the GHOSTDAG protocol puts them in order. That removes the main limitation of classic PoW, because the block rate can be raised without losing more work and without weakening security.

After the Crescendo fork on 5 May 2025 the network produces ten blocks per second, with a target block time of one hundred milliseconds. The Toccata fork on 30 June 2026 added covenants and zero knowledge proof verification on layer one. A large share of reference sites still describes Kaspa as it stood in 2024 and repeats the figures one block per second and tens of KAS per block: both are long out of date.

This page is written as a reference for hardware owners: network parameters, the mechanics of the monthly reward cut, a ranking of the kHeavyHash ASICs in our catalogue with an honest electricity threshold, pools, wallets, the ecosystem, and a separate list of the numbers that the big aggregators still publish in an outdated form.

• 2021Mainnet goes live on 7 November with no premine and no coin sale. For the first six months the network runs in its pre-deflationary phase with a flat reward of 500 KAS per block. On 22 November the kaspad v0.11.4 hard fork ships and mining resumes on 26 November.
• 2022In May the chromatic emission phase begins: the reward drops to 440 KAS and then falls every month. On 24 May the first exchange listing and a rebrand with a new logo arrive. Mining still runs on graphics cards.
• 2023In spring the first kHeavyHash ASICs reach the market, the IceRiver KS0, KS1, KS2 and KS3 series. Hashrate climbs by orders of magnitude and GPU mining of Kaspa dies out within a few months.
• 2023In April the first alpha of rusty-kaspa is published, a full rewrite of the node in Rust in place of the original Go implementation. The rewrite is meant as the precondition for a future increase in blocks per second.
• 2023In September the Kaspium mobile wallet gets its official release in the app stores, having existed only as a beta since February.
• 2024Bitmain ships the Antminer KS3, then the KS5 and KS5 Pro at 20 and 21 TH/s. The kHeavyHash market stops being an IceRiver monopoly and competition pushes down prices for used first generation machines.
• 2024In September the KRC-20 token standard on the Kasplex protocol goes live on mainnet. A network designed for payments picks up its first wave of application load.
• 2024In October the first third-party firmware for the Antminer KS series appears: it claims a gain of up to 20 percent with a developer fee of 1 to 4 percent depending on how hard the machine is pushed.
• 2025The Crescendo fork activates on 5 May at DAA score 110,165,000. The network moves from one block per second to ten, the block reward is divided by ten, and the GHOSTDAG parameters are recalculated for the new pace.
• 2025Through summer and autumn Kasplex zkEVM launches, the first layer two with ethereum compatible contracts on top of Kaspa. In October the network goes through stress tests peaking at several thousand transactions per second.
• 2026In March the public mainnet of Igra Network opens, a decentralised EVM layer built as a based rollup: the transaction queue is formed by Kaspa miners themselves. More than 3,000 transactions per second are claimed and an external audit was carried out.
• 2026In March rusty-kaspa v1.1.0 ships with better synchronisation, and in June versions 2.0.0 and 2.0.1 arrive carrying the parameters of the next fork.
• 2026The Toccata fork activates on 30 June at DAA score 474,165,565. Layer one gets covenants, zero knowledge proof verification and transaction ordering commitments. Node disk requirements grow.
• 2026By the end of July about 96 percent of the total supply has been mined and the block reward drops to 2.45 KAS. Annual network inflation falls to roughly 2 percent.

In an ordinary chain two miners who find a block at almost the same moment compete: one block survives, the other becomes an orphan and its work is thrown away. That is exactly why classic networks keep the block interval long, because the more frequent the blocks, the larger the share of wasted work and the easier an attack becomes. Kaspa solves it differently: both blocks stay in the graph, each one references several predecessors at once, and the protocol agrees on the order afterwards.

The order is set by GHOSTDAG. It splits blocks into the honest core of the graph and everything else: a block counts as blue when it is well connected to the main body, and as red when its links are weak, which is what covert mining looks like. From there a single linear sequence of transactions is built, and transactions that conflict over the same funds are discarded according to that order. The parameter k, which sets the allowed width of the graph, has been 124 since the Crescendo fork.

None of this changes the miner side of the job: the same nonce search against a difficulty target, the same stratum, the same pools. What changes is the scale. Blocks arrive every one hundred milliseconds, the network issues on the order of 864,000 blocks a day, and the reward for each one is tiny by historical standards. A pool pays for that in infrastructure; for the miner the only visible effect is a smoother stream of payouts.

1. A transaction enters the networkThe wallet signs the spend of a UTXO and broadcasts the transaction to nodes. In quiet periods the mempool is practically empty, so there is almost no queue to wait in.
2. The miner takes a block templateThe node hands over a template with transactions and references to several tips of the graph at once: since Crescendo a block may point at up to 16 predecessors.
3. Nonce search over kHeavyHashThe ASIC runs matrix multiplication and Keccak until the result falls below the target. The block that is found goes back to the node and on into the network.
4. The block joins the graphParallel blocks do not compete, they coexist. The merge set of a single block is capped at 248 blocks since Crescendo.
5. GHOSTDAG orders the graphThe protocol colours blocks blue or red and builds one common transaction order. Conflicting spends are cut off according to that order.
6. Confirmation and settlementPractical certainty arrives within seconds. Protocol finality is reached at a depth of 432,000 blocks, roughly 12 hours.
7. The reward becomes spendableCoins from a block mature after 1,000 blocks, about 100 seconds. That is why pools can credit payouts with almost no delay.

kHeavyHash is not a general purpose hash but a pairing of two different workloads. First the data goes through a matrix multiplication against a fixed 64 by 64 matrix, then the result is processed by a function from the Keccak family. The idea was to make the computation depend not only on logic but on moving data around: the authors called the algorithm optoelectronics friendly, meaning it could in principle run on optical accelerators.

In practice this delivered no ASIC resistance, and it never could: a deterministic matrix operation maps beautifully onto silicon. The first kHeavyHash ASICs shipped in spring 2023 and within a few months graphics cards had lost all purpose here. Today the efficiency gap between a GPU and a purpose built machine is measured in hundreds of times, and there is no way back.

The key difference from SHA-256 for a hardware owner is that the kHeavyHash market is narrow. Machines come from essentially two manufacturers, IceRiver and Bitmain, plus small batch units from Goldshell and iBeLink. Model lines refresh less often and the spread of efficiency inside the catalogue is enormous: from roughly 117 joules per terahash at the top down to 650 joules on the oldest compact boxes.

Within a single model there is almost no firmware variation: kHeavyHash never grew the kind of third-party firmware market that SHA-256 has, and most machines run stock. That makes life easier for a beginner and at the same time takes away a familiar tool from an experienced miner, because pushing the efficiency point down, the way it is done on the Antminer S series, rarely works here and never on every board.

The economics of the algorithm deserve a separate reminder. The Kaspa block reward falls every month rather than once every four years, so at a flat price and a flat network hashrate the same machine earns roughly 5.6 percent less each month. Planning payback over a year without accounting for that slope makes no sense.

Full description of the kHeavyHash algorithm →

The Kaspa supply cap is about 28.7 billion KAS, more precisely 28,704,035,605 coins. By the end of July 2026 some 27.59 billion had been mined, roughly 96 percent of the total. That is a fundamental difference from Bitcoin, which at the same moment still has a century of issuance ahead: almost the entire Kaspa supply is already circulating.

The decay mechanism is unusual. For the first six months after launch the network paid a flat 500 KAS per block, then in May 2022 the chromatic phase began at 440 KAS. Since then the reward is multiplied every month by the twelfth root of one half, that is by roughly 0.943874. Twelve such steps make an exact halving, but without a single cliff: this network has no equivalent of Bitcoin halving day.

The Crescendo fork preserved the emission schedule per second rather than per block. When the network moved from one block per second to ten, the block reward was divided by ten. Since 5 July 2026 that means 2.45 KAS per block, so 24.5 KAS per second and about 2,116,800 KAS a day across the whole network. The next cut is due in the first days of August 2026, taking the reward to roughly 2.31 KAS, then 2.18 KAS in September and 2.06 KAS in October.

Annual inflation under this schedule is around 2 percent and halves every year. There is no burn mechanism in the protocol and fees go entirely to the miner. The flip side is obvious: fees are a negligible part of miner revenue, the network is barely loaded with competition for block space, and as issuance fades the question of what pays for security ten years from now stays open.

The asic.es catalogue lists 19 kHeavyHash models, and below is a full breakdown of the ones worth considering in 2026.

Browse miners in the catalogue →

The ranking follows catalogue efficiency strictly, from the best joule per terahash to the worst, and not model popularity.

First things first: the HashCore Toolkit section and the AsicBoost builds on our site target SHA-256 and Scrypt, and there are no kHeavyHash machines on the support lists. Not one of the 19 catalogue models on this algorithm has firmware from us, and we will not promise that it is coming: the kHeavyHash market is too narrow, and the IceRiver and Bitmain KS platforms differ a great deal from the familiar Antminer S series.

At the same time, the claim found in reference sites that no third-party firmware exists for Kaspa at all has been wrong since autumn 2024. For the Antminer KS3, KS5 and KS5 Pro there is third-party firmware from a large mining service, presented in October 2024: it claims a performance gain of up to 20 percent alongside lower consumption, with a floating developer fee of 1 to 4 percent depending on the overclock level chosen. For IceRiver machines there is an open build with overclocking and voltage control, distributed through a public repository and charging no fee.

The practical conclusion for a kHeavyHash ASIC owner is this: fine tuning options exist, but they are more modest than on SHA-256 and almost always come down to overclocking rather than lowering joules per terahash. Moving to any third-party firmware voids the manufacturer warranty, and on machines drawing 3 kW a mistake in voltage settings kills a board rather than causing a simple reset.

If you run a mixed fleet, it makes sense to split your approach: use our toolkit on SHA-256 and Scrypt and count the efficiency gain there, while leaving the kHeavyHash machines on stock firmware and winning on tariff, cooling and the timing of your coin sales.

The entry bar is simple: a kHeavyHash ASIC, a Kaspa wallet address, a pool account and a stable internet line. A separate node is not required, the pool fully covers your link to the network. Setup takes minutes: enter the pool address in the web interface of the machine, use your wallet address with the kaspa: prefix as the login, and your statistics show up in the pool dashboard within a few minutes.

Infrastructure is sized by power draw. A machine at 3,000 to 3,500 W needs a dedicated line, its own breaker and an industrial socket, and from three machines up you are looking at a three phase supply. Over a day such a machine puts out around 80 kilowatt hours of heat, and that heat has to go somewhere: in a closed room the temperature becomes unacceptable within tens of minutes.

Noise is a chapter of its own. The full size KS5, KS5 Pro and KS7 run at the level of an industrial fan and you cannot live next to them. In a home the only realistic options are the compact models: the KS7 Lite at 500 W and the KS0 Ultra at 100 W. Everything else belongs in a garage, a utility room or a hosting site.

Your own node is only needed if you run a pool, a service or your own tooling. After the Toccata fork the requirements went up: at minimum eight processor cores, 16 GB of RAM, 640 GB of SSD and a link from 10 megabytes per second, with 12 to 16 cores, 32 GB of memory and a terabyte drive recommended. The extra disk space comes from the doubled transient mass limit that supports zero knowledge proofs. The database upgrade is one way, so rolling a node back to an older version is not possible.

One more point about pools and versions: both Crescendo and Toccata required pool operators to update how they build block templates, otherwise blocks are rejected as invalid. For an ordinary miner this turns into a simple rule: before a fork, check that your pool has announced readiness, and do not leave a machine on a site that stays silent.

Miner revenue in Kaspa comes from three quantities: daily network emission, your share of total hashrate and the price. Daily emission is currently about 2.12 million KAS, and with a network hashrate on the order of several hundred petahashes per second that works out to roughly 7.5 KAS per terahash per day on a calculation made at the end of July 2026. Fees play practically no part in revenue.

Then electricity is subtracted, and the honest picture here is unpleasant. The tariff break-even for the best machine in the catalogue sits around 0.07 dollars per kilowatt hour, for the mainstream KS5 and KS5 Pro around 0.058, and for previous generations from 0.024 to 0.04. Household tariffs in Spain and most of the EU are more than twice as high, which means that on grid electricity Kaspa mining in Europe loses money across the entire catalogue. That is not a figure of speech and not an invitation to wait: it follows directly from the current reward and the current price.

The third factor is structural: the reward drops 5.6 percent every month. If the price stands still and network hashrate does not fall, your daily income in six months will be about 71 percent of what it is now, and in a year exactly half. Any payback model built on constant income is wrong by construction for this coin. The only scenario where the numbers work is cheap or self generated electricity plus a bet on the price rising, and that risk deserves to be called by its name.

Open the profitability calculator →

One detail that reference sites often get wrong: there is no native Kaspa support in Ledger Live. The device works with KAS through a separate community built app and the external KasVault web interface. The key does stay inside the hardware wallet, but you will not get the usual account management directly in Ledger Live, and that is worth checking before buying a device specifically for Kaspa.

A Kaspa address looks like a bech32 string with the kaspa: prefix, and the prefix is part of the address, not an explanation. Pools normally accept the address in full, prefix included. The network runs on the UTXO model, so regular small pool payouts build up many small outputs over time: sending a large amount then makes the transaction heavier and more expensive, although in absolute terms the fee still stays at fractions of a cent.

KAS trades on many second tier venues: KuCoin, Bybit, Kraken, MEXC, Gate, CoinEx, Bitget, LBank, Uphold. At the same time the coin has no spot market on Binance or Coinbase, and rumours of a listing there have surfaced every year since 2023. For a miner this is a practical matter: what you mine has to be sold on venues with a thinner order book.

Liquidity deserves a sober look. Daily turnover across all exchanges at the end of July 2026 sits around ten million dollars against a market capitalisation of roughly three quarters of a billion. That means regular selling by a small farm does not move the market, while exiting a large position in one go produces visible slippage. Split your sales and never work a single market order.

The hardware itself is bought in Europe either from authorised suppliers with a warranty and customs handled, or on the second hand market. For kHeavyHash the used market is cheap right now precisely because the machines do not pay off on European tariffs, and an attractive price on the box should not mislead you: what counts is the electricity bill over a year, not the price tag. Before buying used, check the running hours, the condition of the fans and whether the unit was run overclocked.

The original purpose of Kaspa is fast settlement. A transaction lands in the graph within tenths of a second, practical certainty arrives in seconds, and the fee amounts to fractions of a cent. On that count the network does its job: on ordinary days the mempool is close to empty, there are no queues, and daily flow holds at the level of hundreds of thousands of transactions.

A second layer of use appeared in September 2024 with the KRC-20 token standard, and through 2025 and 2026 full layer two networks were deployed: Kasplex with its zkEVM, and Igra Network, an EVM layer built as a based rollup, whose public mainnet opened in March 2026. What makes Igra distinctive is that transaction ordering is set by Kaspa miners themselves, so security is borrowed from layer one.

The Toccata fork of 30 June 2026 added covenants, zero knowledge proof verification and transaction ordering commitments to layer one. It is important to see the boundary: Kaspa has no full smart contracts with accounts and global state on layer one. What arrived are the primitives such systems are built from, while application logic still lives on layer two. Claims at either extreme, that there are no smart contracts at all and that they are running on layer one, are equally inaccurate.

The practical weight of the ecosystem is still moderate: layer two activity is measured in thousands of transactions per hour, not millions. For a miner that means fees will stay a rounding error in revenue for the foreseeable future, and income will be set by emission and price.

Kaspa security rests on the same work economics as any PoW network, adjusted for the structure of the graph. GHOSTDAG is built so that blocks created away from the main body are marked red and cannot rewrite the transaction order: to run an attack you have to do more than find many blocks, you have to stay connected to the main graph, which means competing on hashrate in the open. Protocol finality since Crescendo is reached at a depth of 432,000 blocks, about 12 hours, where before the fork it was 24 hours.

The practical risk lies not in the mathematics but in concentration. At the end of July 2026 the largest pool holds some 30 percent of hashrate, and the top two together approach half. Add to that the fact that hardware comes from essentially two manufacturers and you get a narrower bottleneck than in SHA-256, where vendors and pools are many times more numerous. The practical takeaway for a miner is simple: do not join the biggest pool out of habit, because hashrate distribution depends directly on decisions like that.

The number of public nodes holds around 535 against a historical average near 450. The bar to run one keeps rising: after Toccata the minimum configuration is eight cores, 16 GB of memory and 640 GB of SSD. That is still an ordinary server or a strong home computer, but no longer a laptop, and with the next increases in blocks per second the question of node accessibility will get sharper.

No major incidents involving double spends or history rollbacks have been recorded in the network. Both forks, Crescendo in May 2025 and Toccata in June 2026, went through at the DAA scores announced in advance without splitting the network, which for a network running at this block rate is an achievement in itself.

One block per second. Out of date since 5 May 2025. The Crescendo fork activated at DAA score 110,165,000 and moved the network to ten blocks per second with a target block time of one hundred milliseconds. Reference sites that still say one block per second are describing a configuration two steps back.

A reward of 55 KAS, or any figure in the tens of KAS per block. Numbers like that belong to the era of one block per second. After Crescendo the block reward was divided by ten, and at the end of July 2026 it stands at 2.45 KAS. One large mining aggregator was still showing 55 KAS in the middle of July 2026: that is a value from the middle of 2025.

The reward is recalculated once a year. No, the cut happens every month through multiplication by the twelfth root of one half, roughly 5.6 percent. It takes twelve such steps for the reward to halve exactly, but Kaspa never has a single halving day.

Hashrate quoted in exahashes with a unit error. Some aggregators show values in the hundreds of millions of EH/s, which is a unit conversion error. The real order of magnitude at the end of July 2026 is about 0.28 EH/s, that is roughly 280 PH/s.

Finality takes 24 hours. That is the pre-Crescendo parameter. After the fork finality depth is 432,000 blocks, about 12 hours, and the merge depth bound is 36,000 blocks, about an hour.

Kaspa has no smart contracts and none are planned. Out of date. The KRC-20 token standard has worked since September 2024, layer two networks with EVM since 2025, and the Toccata fork of 30 June 2026 added covenants and zero knowledge proof verification to layer one.

Full smart contracts are live on Kaspa layer one. The opposite extreme, also wrong. Toccata delivered primitives, but there is no account model and no global state on layer one, and application logic lives on layer two.

There is no third-party firmware for kHeavyHash. Wrong since October 2024. For the Antminer KS series there is third-party firmware claiming a gain of up to 20 percent with a fee of 1 to 4 percent, and for IceRiver machines there is an open build with overclocking and voltage control. It is a separate matter that our own toolkit does not cover this algorithm.

Ledger supports Kaspa natively. It does not. There are no Kaspa accounts in Ledger Live, and the device works with the coin through a third-party app and an external web interface. The key stays in the hardware wallet, but you will not get the familiar interface.

The founder of Kaspa goes by a different name. Distorted spellings of the name circulate online. The protocol grew out of the work of Yonatan Sompolinsky, author of the GHOST and PHANTOM research, and development has been carried by the community since DAGLabs was wound down in 2021.

The network processes more than 150 million transactions a day. That is a one off peak recorded during stress tests in autumn 2025, not a working level. Ordinary daily flow at the end of July 2026 runs at some 800 thousand transactions.

A node runs fine on a weak machine with a couple of cores. Out of date twice over. After Crescendo the minimum was eight cores, 16 GB of memory and 256 GB of SSD, and after Toccata the disk requirement grew to 640 GB plus a link from 10 megabytes per second. The database upgrade is one way and cannot be rolled back.