Starknet (STRK): the STARK rollup on Ethereum, its token and its risks

Starknet is a validity rollup that runs on top of Ethereum: it executes transactions on its own network, proves with a STARK proof that it did so correctly, and lets Ethereum check that proof. StarkWare, the Israeli company behind the STARK proof system, built it. Starknet has been live on Ethereum mainnet since 29 November 2021, and its STRK token pays the network’s fees, secures it through staking and carries governance votes.

The record is uneven. Its proofs have never let a wrong state reach Ethereum, not even during the outages of September 2025 and January 2026, when the network stopped for hours. Yet StarkWare alone still orders and publishes the blocks. DeFi deposits on the chain are about half their March 2024 peak, and investors and early contributors can receive up to 127 million more STRK each month until March 2027.

Then there is the question of where Starknet lives. A non-binding Snapshot poll opened on 15 October 2026 asks whether to research leaving Ethereum and becoming an independent, post-quantum layer 1. Voting closes on 22 October 2026 at 09:32 UTC; according to the proposal text, a yes backs research and nothing more. Our news piece follows the Starknet quantum poll and who holds the votes; this page explains the network itself.

What Starknet is: a STARK validity rollup settling on Ethereum

A rollup moves the work of running transactions off Ethereum, then posts a compact record back to it. There are two families. Optimistic rollups assume batches are honest and leave a window for anyone to challenge them. Validity rollups such as Starknet attach a cryptographic proof to every batch, so Ethereum only accepts the new state once the maths checks out.

The Starknet documentation calls the network a decentralized, permissionless layer 2 validity rollup. That holds for users and developers, who can deploy a contract or send a transaction without asking anyone. It does not yet hold for block production, which StarkWare still controls (see the risk section below).

The proofs are STARKs, a type of zero-knowledge proof developed by the research group around StarkWare’s founders. Two properties matter to users. STARKs need no trusted setup ceremony, which L2BEAT lists as a “transparent setup”. And they rest on hash functions rather than on elliptic curves, which is why StarkWare says the proofs would survive a large quantum computer.

How Starknet works, step by step

A Starknet transaction goes through four stages, as the documentation describes them:

• The sequencer collects transactions from its mempool, orders them and builds blocks.
• The Starknet operating system program (SNOS) runs over the block, and a proof of that run shows the block leads to the stated end state.
• StarkWare’s shared prover, SHARP, aggregates many blocks into one proof, which contracts on Ethereum verify.
• Compressed state differences are published to Ethereum, so anyone could rebuild the Starknet state from Ethereum data alone.

Since 13 March 2024 those state differences travel as blobs, the cheaper data format Ethereum introduced that month, rather than as calldata. Version 0.13.4, live on 25 March 2025 according to L2BEAT, added stateful compression to shrink them further.

Users see two kinds of confirmation. Pre-confirmations arrive in about half a second, according to the project’s 2025 review. Finality on Ethereum takes far longer, because proofs are built and aggregated in batches: Starknet’s release table sets targets of three hours, then under 1.5 hours, as goals rather than results.

S-two, the prover that replaced Stone

In late 2025 Starknet switched provers. L2BEAT dates the change to 19 October 2025, and the Starknet blog announced S-two (also written Stwo) as live on mainnet on 3 November 2025. S-two uses Circle STARKs over a small prime field called M31, and the blog describes it as up to an order of magnitude faster than Stone, the prover it replaced. The code is open source under the Apache 2.0 licence in the starkware-libs GitHub organisation.

Cairo, and why Starknet is not EVM compatible

Contracts on Starknet are written in Cairo, a language designed so that every program run can be proven. Cairo began in 2020 as a low-level, assembly-like language. Cairo 1.0, whose first public version came out on 5 January 2023, brought a Rust-like syntax and Sierra, an intermediate layer that makes sure every Cairo run can be proven. Starknet Alpha v0.11.0 then opened the network to Cairo 1 contracts and started a transition away from the old ones.

Here lies the main practical difference with Arbitrum, Optimism or Base. Starknet does not run the Ethereum Virtual Machine, so a Solidity application cannot be copied across as it is: it has to be rewritten or go through a translation layer. The porting work and a smaller pool of Cairo developers are one reason the app catalogue and liquidity on Starknet are thinner than on the large EVM rollups.

Every account is a smart contract

Starknet has no plain key-pair accounts. A StarkWare post of 20 July 2026 sums up the design: “Every Starknet account is natively a smart contract, and account abstraction is the only account model.” In practice, wallets can offer multisig, session keys for games, passkey logins and new signature schemes without any change to the protocol. Transactions can also bundle several calls, such as an approval and a swap, into one.

Fees on Starknet: paid in STRK, kept by the sequencer

Since version 0.14.0, rolled out on 1 and 2 September 2025, fees are paid in STRK only; ETH was accepted before, according to the STRK page of the documentation. A fee covers three resources: L2 gas for computation and data on Starknet, L1 data gas for the state changes posted to Ethereum, and L1 gas for messages sent to Ethereum. The L2 gas price adjusts block by block in the style of Ethereum’s EIP-1559, targeting 80% of block capacity, with a floor of 3 gFri, or three billionths of a STRK.

Costs are low. Starknet’s own year-end review puts the average gas fee under $0.001 through 2025 and a transfer under one cent. The fees do not shrink the token supply, though: the docs say no fee is burned and the sequencer receives all of them. A Starknet blog post of 30 September 2025 lists a fee burn as coming next, with the date marked “TBD”.

Starknet history: from StarkEx to the Grinta upgrade

StarkWare was founded in early 2018 in Netanya, Israel. Its four founders, as Globes listed them when the company raised its Series A that October, were Eli Ben-Sasson, Uri Kolodny, Michael Riabzev and Alessandro Chiesa; StarkWare’s own site still names Ben-Sasson, Kolodny and Chiesa as co-founders. The company first sold scaling as a service. StarkEx, on Ethereum mainnet since June 2020 according to StarkWare’s documentation, began with the DeversiFi exchange and later ran dYdX, Immutable X and Sorare, each on its own engine.

Starknet was the general-purpose version. Its alpha reached Ethereum mainnet on 29 November 2021, and the StarkGate bridge followed on 9 May 2022. The token came later. STRK distribution through the “Starknet Provisions” programme began on 14 February 2024, and trading opened on 20 February 2024. The airdrop drew criticism over its eligibility rules, and the first insider unlock, about 1.3 billion STRK due on 15 April 2024, drew more. On 22 February 2024 StarkWare replaced it with the gradual monthly schedule still in force, The Block and Blockworks reported.

The first serious glitch followed in April 2024. Starkscan’s block data showed a gap of about four hours between two blocks, and Starknet said on X that a rounding-error bug had caused a reorganisation of blocks, Cointelegraph reported on 5 April. The team said production had continued and that new transactions were refused for a few minutes; Cointelegraph could not confirm whether blocks had stopped. We found no formal incident report for that episode.

2025 was a year of upgrades. L2BEAT moved Starknet to Stage 1 on 15 May 2025, once a Security Council, upgrade delays and a path for censored transactions were in place. Version 0.14.0, named Grinta, went live in early September 2025 with three sequencers taking turns in consensus instead of one, though its announcement says their operation and control stay centralised under StarkWare. Bitcoin staking followed on 30 September 2025, then S-two.

In 2026 the focus moved to privacy and Bitcoin. Version 0.14.2, live on 20 April 2026, added in-protocol proof verification and STRK20, a framework for tokens with private balances. On 12 May 2026 strkBTC launched as the first STRK20 asset: a bitcoin-backed token with public and shielded modes, where the viewing keys of shielded balances go to an independent auditor and the bitcoin bridge is run, for now, by a federation of five firms. Version 0.14.4 reached mainnet on 5 October 2026, with node operators asked to update to Pathfinder 0.24.0 or Juno 0.16.6.

The STRK token: supply, allocation and monthly unlocks

StarkWare created 10 billion STRK in May 2022, and they were minted on Ethereum on 30 November 2022, according to the documentation. Ten billion is the starting supply, not a cap: staking rewards mint new tokens on a schedule set with the community, and L2BEAT put the supply at about 10.15 billion on 17 October 2026. The staking page says inflation will stay under 1.6% a year under the initial parameters, and the project’s 2025 review reported about 33 million STRK minted by staking since launch.

The initial allocation, as published in the docs:

Holder	Share of the 10 billion STRK
Early contributors	20.04%
Investors	18.17%
Grants, including development partners	12.93%
StarkWare	10.76%
Foundation strategic reserves	10.00%
Community provisions (airdrops)	9.00%
Community rebates	9.00%
Foundation treasury	8.10%
Donations	2.00%

Insiders hold a large block: 38.21% of the initial supply for early contributors and investors, plus 10.76% for StarkWare. CoinGecko counted 7.59 billion STRK in circulation on 17 October 2026, about three quarters of the starting supply.

127 million STRK a month until March 2027

Only the investor and early-contributor tokens vest on a fixed calendar. From 15 April 2024 to 15 March 2025, up to 64 million STRK were released on the 15th of each month. Since 15 April 2025 the monthly token unlock has been up to 127 million STRK, and it continues until 15 March 2027. Locked tokens cannot be transferred, sold or pledged, and the docs say they could not be staked either at their last update, though they can be delegated for votes.

As of 17 October 2026, five tranches remain: 15 November and 15 December 2026, then 15 January, 15 February and 15 March 2027. By our arithmetic, that is up to 635 million STRK still to come. CoinGecko splits the November tranche into 66.61 million for early contributors and 60.39 million for investors.

The price history explains why each tranche gets watched. CoinGecko lists STRK’s all-time high at $4.41, a level price trackers date to its first trading day, 20 February 2024; by early March the daily close sat below $2.30. Coindaro shows the live price on this page rather than in the text, and none of this is a view on where the token goes next.

How to stake STRK and Bitcoin on Starknet

Staking went live on mainnet on 26 November 2024, The Block reported, after token holders ratified the SNIP 18 proposal on 13 September 2024 in Starknet’s first community vote, held on Snapshot X. The rules on the official staking page today:

• A validator locks at least 20,000 STRK and runs a full node.
• Delegators have no minimum and pick a validator, who takes a commission.
• Unstaking takes seven days, down from the 21 days set in SNIP 18.
• Staked STRK carries 75% of consensus weight and staked Bitcoin 25%.

The Bitcoin side arrived on 30 September 2025. Holders of wrapped bitcoin tokens such as WBTC, tBTC and SolvBTC can stake them, and a post on the Starknet blog sets the limit: “The consensus power of staked Bitcoin is limited to 25% of the network.” A Monetary Council decides which wrappers earn rewards. On 4 June 2026 it said it intended to drop LBTC from that list after a seven-day comment period, leaving WBTC, SolvBTC, tBTC and strkBTC untouched. SNIP 39, a draft posted in April 2026, asks to make strkBTC stakable.

The project reported more than 1.1 billion STRK staked at the end of 2025, over 23% of the circulating supply, and more than 1,700 BTC staked. We found no official total for October 2026. The staking page quotes no fixed yield, and we do not either: rewards depend on the minting curve, the total staked and the validator’s commission.

For now stakers watch blocks more than they make them. Since Staking v2 in June 2025, validators attest to blocks, which lets anyone measure their performance, but they do not decide which blocks become final. Version 0.15, with Staking v3, would have blocks verified by stakers. Starknet’s release table lists it for September 2026; as of 17 October 2026 it had not been announced as live.

Who runs Starknet: StarkWare, the Foundation and a Security Council

Three groups share control. StarkWare, led by co-founder and chief executive Eli Ben-Sasson, writes most of the core software and operates the sequencers and the prover. The Starknet Foundation, set up in 2022, holds treasury tokens, runs grants and governance, and appoints the Security Council. James Strudwick was appointed its executive director in August 2024, after Diego Oliva stepped down, The Block reported at the time.

The Security Council has 12 seats. The documentation lists the member organisations: StarkWare, Nethermind, Zellic, Hypernative, Pragma, AVNU, Cartridge, Ready, Karnot, Space Duck, Flying Muffins and Protolabs. The individuals holding the keys are not published. According to L2BEAT, nine of the twelve can push an emergency upgrade with no delay, any three can post a state root that includes censored transactions, and the Foundation can remove members as well as appoint them.

Governance votes, and how concentrated they are

STRK holders vote on Starknet Improvement Proposals (SNIPs) through Snapshot X on the Governance Hub, using tokens on Ethereum, staked STRK or voting STRK on Starknet. Most voting power is delegated. In the quantum poll that opened on 15 October 2026, about 94% of the voting power counted on the morning of 17 October came from STRK the Foundation had delegated to whitelisted delegates, as our coverage of the vote details. That whitelist covers 109 addresses with fixed weights. Anyone holding STRK can vote, but those delegates carry the result.

Is Starknet safe? Stage 1, one operator and two outages

L2BEAT rates Starknet a Stage 1 ZK rollup, a middle grade. Stage 1 means users keep some protection if the operator turns hostile, but the Security Council can still override the system. As of 17 October 2026, L2BEAT listed these risks:

• Only a whitelisted proposer can post new state roots; if it stops, withdrawals freeze until a council minority steps in.
• Users can send transactions through Ethereum but cannot force the operator to include them.
• The single operator could reorder transactions to extract value (MEV).
• Regular upgrades come with an 8-day delay, too short for the Stage 2 standard of 30 days, and emergency upgrades come with none.
• The core bridges let out at most 5% of locked funds in any 24 hours.

None of these has cost users money at the protocol level so far: we found no case of funds lost through a fault of Starknet’s own contracts or proofs. Apps built on it are another matter. In February 2025 the zkLend lending market lost about $9.5 million to an exploit of its own contracts, Decrypt reported, and later wound down. On 20 August 2026 Atomiq, a third-party service for bitcoin swaps into Starknet, took all its routes offline after what it called AI-assisted attacks; no user funds were lost, according to Crypto Briefing, the only outlet we found with an original report.

2 September 2025: two reorgs after the Grinta upgrade

One day after version 0.14.0, the network was intermittently unavailable from 02:20 to 13:40 UTC. The Starknet incident report describes two chain reorganisations that reversed about an hour and a half of activity in total. The report gives three causes: Ethereum node failures that left the sequencers seeing different Ethereum states, a manual fix that skipped some validations, and a bug in the blockifier, the component that executes transactions. The proving layer could not prove the inconsistent blocks, so the chain’s integrity held.

5 January 2026: a revert bug halts sequencing

A blockifier bug in how nested calls were reverted produced incorrect execution. Activity between 09:24 and 09:42 UTC was dropped and reverted, and sequencing stayed paused from 09:42 to 14:17 UTC, about four and a half hours, according to the January incident report. Again the faulty execution never reached finality on Ethereum, because the proving layer stopped it. The follow-ups listed: fuzz tests comparing the blockifier with SNOS, an internal audit of the revert logic, and an automatic halt whenever execution and proving disagree.

Both failures started in the execution layer, not in the cryptography. The proofs did their job; the sequencer, the part users wait on, is where Starknet has broken.

Audits: the staking contracts are covered, the core is less visible

Three auditors’ reports on the staking system are public in StarkWare’s staking repository: Zellic on Staking V1 (22 November 2024), a CodeHawks competitive audit on Staking V2 (14 June 2025), and Pashov Audit Group on Bitcoin staking (review from 31 July to 15 August 2025). We found no public audit report for the sequencer, the blockifier or the Starknet OS published after 2022. StarkWare also says its post-quantum Falcon-512 account contracts have not been audited yet.

Starknet and quantum computers: the proofs, the bridge and the L1 poll

Whether Starknet is quantum resistant depends on the layer. The STARK proofs rest on hash functions. StarkWare calls them post-quantum secure, and the ethereum.org page on quantum resistance (updated 8 September 2026) says STARKs are already quantum-resistant. The rest of the stack is not there yet. StarkWare’s roadmap of 30 June 2026, as The Block reported it, has three phases: BLAKE2 in place of Pedersen hashes plus post-quantum consensus signatures such as Falcon-512; tools to migrate older contracts; and the parts tied to Ethereum, namely bridge syscalls and blob data availability, which wait on Ethereum’s own transition.

That last phase is the root of the poll. On 8 October 2026 Starknet said on X that it was considering becoming its own layer 1, with 2027 as a target. “Quantum-proofing crypto is urgent, so we must move at lightning speed,” Ben-Sasson said, as quoted by Cointelegraph on 15 October. The proposal itself says a yes vote supports further research and does not approve a move to layer 1. Any real exit would need further proposals and votes, and no timeline exists.

Starknet in numbers: TVL, value secured and activity

Two measures of the money on Starknet differ by a wide margin, so each figure carries its date:

Measure	Value	Source and date
DeFi total value locked	$159.2 million	DefiLlama, 17 October 2026
DeFi TVL peak	$330.8 million	DefiLlama, 26 March 2024
Total value secured (all assets on the rollup)	$564.3 million, 48.3% of it in STRK	L2BEAT, 17 October 2026, about 19:30 UTC
Stablecoins on Starknet	$148.4 million	DefiLlama, 17 October 2026, about 19:40 UTC
Activity	4.71 user operations per second over 24 hours	L2BEAT, 17 October 2026
Largest app by TVL	Extended (perpetual futures), $115.3 million	DefiLlama, 17 October 2026, about 19:40 UTC

Almost half of the value secured is STRK itself, so the figure moves with the token price. Activity is far below capacity: the project reported a peak of 2,630 user operations per second in 2025, against fewer than five on 17 October 2026. Behind Extended, the main apps on DefiLlama that day were Starknet’s Bitcoin staking ($50.0 million), the Ekubo exchange ($26.7 million), Endur liquid staking ($13.5 million) and the Vesu lending market ($11.2 million).

How much StarkWare raised: about $262 million by 2022

The money went to StarkWare, the company; STRK itself was never sold to the public. The Block counted $262 million raised by May 2022, which includes about $6 million raised before the Series A, according to Globes. The four main rounds:

Round	Amount	Date	Lead
Series A	$30 million	29 October 2018	Paradigm
Series B	$75 million	24 March 2021	Paradigm
Series C	$50 million ($2 billion valuation)	16 November 2021	Sequoia Capital
Series D	$100 million ($8 billion valuation)	25 May 2022	Greenoaks Capital and Coatue

Other backers included Sequoia, Founders Fund, Pantera, Coinbase Ventures and Tiger Global, as well as Three Arrows Capital and Alameda Research, both of which collapsed in 2022. Investors received 18.17% of the initial STRK supply, and their tokens are part of the allocation released each month until March 2027. We found no funding round announced after May 2022.

Starknet dates to watch: the poll, the unlocks, v0.15

The quantum poll closes on 22 October 2026 at 09:32 UTC; its outcome only decides whether research goes ahead. The next 127 million STRK release falls on 15 November 2026, and the last scheduled one on 15 March 2027. Version 0.15 and Staking v3, which would have stakers verify blocks, sit on the official table for September 2026 but had not been announced as live by 17 October 2026. Still undated: a STRK fee burn, a trustless bridge for strkBTC, and the move to Stage 2 that would require longer upgrade delays and a Security Council limited to provable bugs.