RSA-896 factored with CADO-NFS ported to GPUs, and no new threat to deployed keys

Sources: Weis post, author's own account, HN discussion
Summary: Stephen A. Weis published the two 135-digit factors of RSA-896 on his own site in a post dated 2026-09-19, and his page and the account @sweis, which links that site, now carry the method rather than leaving it to the thread: an existing implementation of the general number field sieve, CADO-NFS, ported to GPUs and scheduled onto idle capacity that was already paid for. He states there is no new algorithmic factoring improvement and no new threat to deployed keys, so what the run demonstrates is the cost of assembling the compute rather than a change in the difficulty of the problem. The factors were not multiplied out for this page, no third party has published a confirmation, and neither source claims a factoring record, so none is claimed here.
Comments: A commenter corrected the account's description of the number field sieve from exponential to subexponential, and the account named sweis replied that he misspoke and corrected it, while keeping the point that the runtime did not improve.
Why it matters: The result moves on engineering and compute scheduling rather than on mathematics, and the author states key-size guidance is unchanged by it.