In general, MLKEM is indeed faster than X25519. By a small amount yes, but I am
not aware of any X25519 implementations that are faster than MLKEM. MLKEM KeyGen
+ Decaps is on the order of ~70,000 cycles (Encaps is approx 35000
cycles-ish) whereas the equivalent for X25519 (KeyGen + DH) is closer to 100,000
cycles. This does not control for variations that you will see in actual high
performance deployments, but the 10,000+ cycle difference gives you an idea of
the pecking order. These numbers were pulled from a hat (eBATS) and will of
course vary.

I think a better way to phrase is is that MLKEM + X25519 is slightly
better in performance than P-256, which indicates that the slowdown from
hybridization is well outside of the range of fatality for adoption in any
sense. Anyone able to afford P-256 can afford the hybrid (and will probably see
a speedup, unless they pick the P-256 hybrid)

If performance is really an issue, I’d be curious to see if there’s consensus to
adopt eg. 4Q-MLKEM768 as a high performance alternative. I don’t think there
will be, but while we’re on the topic of new things and performance, I think
these two might go together nicely.

Joshua

-------- Original Message --------
On Saturday, 02/28/26 at 02:03 Nadim Kobeissi <[email protected]> wrote:

> Dear Deirdre,
> 
> This isn’t very important, but putting aside ML-KEM’s massive key size
> increase over X25519, the claim that ML-KEM is almost twice as fast as X25519
> is highly dependent on hardware optimizations, as far as I’m aware, and is not
> always true.
> 
> It could become increasingly universally true over the coming years though, as
> hardware optimizations to accommodate ML-KEM become more common. For example,
> I think Apple hardware recently implemented SHA3 as a CPU instruction, and
> that certainly would give ML-KEM a big speed boost!
> 
> Nadim Kobeissi
> Symbolic Software • https://symbolic.software
> 
> 
> > On 28 Feb 2026, at 1:22 AM, Deirdre Connolly <[email protected]>
> > wrote:
> > 
> 
> > 
> > > In particular when the use of hybrid crypto comes with negligible
> > overhead, as for ML-KEM + ECC.
> > 
> > X25519 is almost twice as slow as MLKEM768
> > (https://blog.cloudflare.com/pq-2025/#ml-kem-versus-x25519);  P-256 is about
> > the same
> > 
> > On Fri, Feb 27, 2026, 5:25 PM Tibor Jager <[email protected]> wrote:
> > 
> > > 
> > > 
> > > > Am 27.02.2026 um 21:16 schrieb Ilari Liusvaara
> > > <[email protected]>:
> > > > - There does not seem to be any evidence that ML-KEM is weak. I think
> > > >  that if ML-KEM gets badly broken, it will be for unforeseeable reasons
> > > >  (which is a risk for any cryptographic algorithm, including prime-
> > > >  field ECC).
> > > 
> > > Except that for a hybrid mode, both ML-KEM and ECC must be broken
> > > simultaneously.
> > > 
> > > I think it is unwise to rely *only* on ML-KEM (or any other scheme based
> > > on relatively new hardness assumptions), and currently do not support any
> > > draft that does not use hybrid cryptography. In particular when the use of
> > > hybrid crypto comes with negligible overhead, as for ML-KEM + ECC.
> > > 
> > > For almost every broken cryptosystem there was a time when there seemed to
> > > be no evidence that it is weak. ML-KEM still needs to stand the test of
> > > time.
> > > 
> > > Best regards,
> > > Tibor
> > > _______________________________________________
> > > TLS mailing list -- [email protected]
> > > To unsubscribe send an email to [email protected]
> > 
> > _______________________________________________
> > TLS mailing list -- [email protected]
> > To unsubscribe send an email to [email protected]

Attachment: signature.asc
Description: OpenPGP digital signature

_______________________________________________
TLS mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to