ML-DSA is NIST's lattice-based post-quantum signature standard, published as FIPS 204 on August 13, 2024 and derived from CRYSTALS-Dilithium. It comes in three parameter sets, ML-DSA-44, -65 and -87, at NIST security categories 2, 3 and 5. The price is size: an ML-DSA-65 signature is 3,309 bytes, against 65 bytes for an Ethereum ECDSA signature.
On Ethereum, a transaction that carries an ML-DSA-65 public key and signature and checks them with a precompile would pay about 210,000 gas for its calldata under current rules. The check itself, priced at 9,000 gas in a July 2026 draft proposal, is small enough to fit inside that figure.
What ML-DSA Is
ML-DSA stands for Module-Lattice-Based Digital Signature Algorithm. NIST's FIPS 204 specifies it, and the standard words its own claim carefully: ML-DSA "is believed to be secure, even against adversaries in possession of a large-scale quantum computer."
It does the job ECDSA, EdDSA and RSA signatures do today, which NIST's final transition report classes as quantum-vulnerable. If you want the background on what a quantum computer would break on a chain, start with the Q-Day threat.
Two related names come up often. SLH-DSA, also on NIST's list of post-quantum signatures, is a hash-based scheme. ML-KEM, specified in FIPS 203 and derived from CRYSTALS-Kyber, is a key-establishment scheme that two parties use to agree on a shared secret key.
The NSA's guidance labels it "ML-DSA (previously CRYSTALS-Dilithium)". FIPS 204 says ML-DSA is derived from version 3.1 of CRYSTALS-Dilithium, and it changes details along the way. Signing is "hedged" by default in ML-DSA, where Dilithium 3.1 defaulted to deterministic signing, and the final standard restored a check for malformed input that the draft had left out. When you pick a library, check that it implements final FIPS 204.
How It Works, Briefly
ML-DSA has three algorithms: key generation, signing and verification. Its security rests on the Module Learning With Errors (MLWE) problem, a lattice problem, together with a variant of the related Module Short Integer Solution problem. The scheme works with a matrix of polynomials, and the numbers in each name are that matrix's dimensions: in ML-DSA-65, the matrix is 6 by 5.
Signing follows a pattern called Fiat-Shamir with Aborts, and apart from the abort step it is shaped like a Schnorr signature such as EdDSA. The signer produces a commitment, derives a challenge from the commitment and the message, and computes a response. If releasing that result would leak information about the private key, the signer discards it and tries again. FIPS 204 gives the expected number of tries: 4.25 for ML-DSA-44, 5.1 for ML-DSA-65 and 3.85 for ML-DSA-87.
ML-DSA-44, -65 and -87
| Set | Security | Public key / signature |
|---|---|---|
| ML-DSA-44 | Category 2, 128 bits | 1,312 / 2,420 bytes |
| ML-DSA-65 | Category 3, 192 bits | 1,952 / 3,309 bytes |
| ML-DSA-87 | Category 5, 256 bits | 2,592 / 4,627 bytes |
The sizes come from Table 2 of FIPS 204. The bit figures are the classical security strengths in Table 3 of NIST IR 8547, published October 27, 2026.
NIST's categories are pegged to familiar primitives. Category 2 means breaking the scheme takes at least as much work as finding a collision in SHA-256. Category 3 matches a key search on AES-192, and category 5 a key search on AES-256.
Private keys are larger again, at 2,560, 4,032 and 4,896 bytes, but FIPS 204 lets you store only the 32-byte seed and regenerate the rest. One caveat applies to the smallest set: NIST IR 8547 says ML-DSA-44 drops to category 1 if it is used with an approved random bit generator of less than 192 bits of security.
What the Size Means on a Chain
Ethereum charges 16 gas per non-zero byte of calldata (EIP-2028) on top of the gas a transaction spends on execution. EIP-7623 adds a floor of 40 gas per non-zero byte, and the transaction pays whichever total is higher. Keys and signatures look like random data, so almost every byte is non-zero, and for a transaction carrying an ML-DSA-65 key and signature the floor wins unless execution costs more than about 126,000 gas. Applying both rates to the FIPS 204 sizes:
| Set | Key + signature | Calldata gas, floor / standard |
|---|---|---|
| ML-DSA-44 | 3,732 bytes | 149,280 / 59,712 |
| ML-DSA-65 | 5,261 bytes | 210,440 / 84,176 |
| ML-DSA-87 | 7,219 bytes | 288,760 / 115,504 |
The rows count every byte as non-zero, which overstates the cost by a fraction of a percent. For scale, every Ethereum transaction starts at 21,000 gas, and an ordinary account's 65-byte ECDSA signature is covered by that base. A smart-contract account that passes an ECDSA signature in calldata pays about 1,040 to 2,600 gas for it.
Some designs keep the public key on chain and send only the signature. EIP-8164 stores an ML-DSA-44 key in the account itself, carries the signature in its own transaction field and adds a fixed 50,000 gas to verify it, in place of the usual ECDSA check. Fireblocks deploys its ML-DSA-44 verifier's key once as contract code; that verifier spends so much on execution that its 2,420-byte signature pays the standard rate, 38,720 gas. Whichever design you use, moving from ML-DSA-44 to ML-DSA-65 adds 889 bytes per transaction when only the signature travels, and 1,529 bytes when the key travels too.
Checking the signature costs less, if the chain has a precompile for it. EIP-8355, a draft from July 2026, proposes precompiles for all three sets at 6,500, 9,000 and 13,500 base gas, and says the figures still need benchmarking before they are final. The October 2025 draft EIP-8051 covers ML-DSA-44 only, at 4,500 gas. Both are still drafts, and a 9,000-gas verification on its own leaves the transaction well under the point where the floor stops applying. Without a precompile, verification runs as contract code: in September 2026 Fireblocks reported a FIPS 204-compliant ML-DSA-44 verifier at 1.23 million gas per verification, plus about 4.1 million gas to deploy each public key once.
Put together, these numbers suggest that once a precompile exists, carrying the bytes costs many times more than checking the math. The gap may widen: EIP-7976, in Last Call as of November 2026, would raise the floor to 64 gas per byte. All of this is Ethereum mainnet pricing. Other EVM networks, QDay Network among them, set their own fees.
Choosing a Parameter Set
The case for ML-DSA-44 is size. EIP-8051 uses it for verification and EIP-8164 for converting Ethereum accounts to post-quantum keys. Cloudflare's July 2026 analysis of signature candidates calls ML-DSA-44's security a "generous 160 bits" by its own estimate.
The case for the larger sets is time. The author of EIP-8355 argues that "an account authenticator is long-lived and holds funds for years, so it wants margin above 128 bits." The same Cloudflare post notes that attacks on lattices cut more security over the last 25 years than they have recently, and that production lattice cryptography uses parameter sets well above 128 bits as a hedge.
Some policies make the choice for you. The NSA's CNSA 2.0 FAQ, in its October 2026 version, requires ML-DSA-87 for all classification levels in US national security systems.
Our recommendation: use ML-DSA-44 where every byte is paid for and keys rotate often, ML-DSA-65 for account keys and agent keys that will hold value for years, and ML-DSA-87 when a policy you must follow asks for category 5.
Signing Options Developers Run Into
FIPS 204 leaves a few choices to the implementer, and they matter when two systems have to agree:
- Hedged signing is the default. It mixes fresh randomness with a precomputed value from the private key, which helps against side-channel attacks and against a faulty random number generator. A deterministic variant is allowed, but NIST says it should not be used where side-channel attacks are a concern and cannot be otherwise mitigated. The same verification algorithm accepts signatures from either variant.
- HashML-DSA is a separate, domain-separated variant that hashes the message before signing.
- Signing takes an optional context string of up to 255 bytes, empty by default. EIP-8355's draft precompiles always verify with the empty context, so a contract that needs domain separation has to put it inside the message.
Deadlines: What ML-DSA Replaces, and by When
NIST's final transition report, IR 8547, published October 27, 2026, deprecates ECDSA, EdDSA and RSA signatures after 2030 and disallows them under NIST standards and guidelines after 2035. It also says organizations may keep using them while they migrate, so ECDSA remains allowed in the meantime.
For US federal systems the date is earlier. Executive Order 14412, signed June 22, 2026, directs guidance that requires federal high-value assets and high-impact systems to use post-quantum digital signatures by December 31, 2031. Other governments' dates are listed in What Is Q-Day?.
Many systems will run both kinds of signature for a while. NIST says hybrids are commonly built as dual signatures, in which every component must verify, and expects them to be temporary.
Where QDay Fits
QDay Network accounts sign with ECDSA today. Its technical docs put native ML-DSA-65 accounts in Phase 2, planned for Aevum Testnet in Q2 2027 and for mainnet in Q4 2027. For the transition, the docs describe hybrid signatures in which an ML-DSA-65 signature and an ECDSA signature must both be valid, with hybrid keys for agent wallets planned for Q4 2026. A single BIP-39 mnemonic will derive both the existing ECDSA account and the new ML-DSA-65 account.
