A 3.75 GHz model gives lower F10.7 errors than a sunspot model in 64 cycle 25 months within the cycle 24 range

This study tests whether a cycle 24 radio model gives better same-month F10.7 estimates than a sunspot model in the available part of cycle 25. The target is a daily median of Canadian scans adjusted to one astronomical unit. Monthly means use the same daily dates for the target, Japanese nominal 3.75 GHz flux, and SILSO V2 sunspot number. Each month needs at least 20 common days. A linear radio model and a quadratic sunspot model are fitted to 132 months from December 2008 through November 2019. The test covers December 2019 through September 2026. It keeps 64 months within both training predictor ranges. The radio root mean square error is 4.14 solar flux units (sfu). The sunspot error is 9.63 sfu. The radio error is 57.0% smaller. The paired error difference is 5.49 sfu. Its conditional 95% interval from 12-month calendar blocks is 2.96 to 9.43 sfu. Tested changes in polynomial degree, data processing, month selection, and target source keep the radio error smaller. Both models have larger test errors than training errors. This is an exploratory study; the test data were inspected during development. Much of the 2024 maximum is outside the tested ranges. Instrument changes and solar phase can affect the result. The study establishes a matched-sample benchmark, not a new physical cause. Code, figures, residuals, coefficients, source hashes, and reproduction instructions are preserved. Source archives remain external under their own terms.

F10.7 · solar radio flux · solar cycle 25 · sunspot number · Nobeyama · calibration · matched dates · reproducible research