For every fixed inverse temperature $\beta>1$, we prove that the expected free
energy per spin of the zero-field Gaussian Sherrington--Kirkpatrick model
differs from its Parisi limit by at most $C_\beta N^{-7/10}$. This improves
the exponent $2/3$ of our earlier work (arXiv:2609.09103). We also compare the
bound with the finite-size correction of order $N^{-3/4}$ predicted by Franz,
Parisi, and Ricci-Tersenghi, and with the strict improvement over $2/3$ that
would follow from the ideas in OpenAI's recently released manuscripts.