Every optical satellite splits incoming light into bands. What separates a multispectral instrument from a hyperspectral one is how many bands there are, how wide each one is, and whether they sit next to each other without gaps.
Broad bands versus contiguous narrow bands
A multispectral sensor carries a handful of bands, typically between four and about a dozen, each covering a wide slice of the spectrum. A familiar open archive example carries thirteen bands spanning the visible, near infrared and shortwave infrared, with individual bandwidths ranging from roughly fifteen to nearly two hundred nanometres. Those bands are placed deliberately, with gaps between them, to sample regions where useful contrast exists.
A hyperspectral sensor, also called an imaging spectrometer, carries hundreds of bands that are narrow and contiguous. A current spaceborne spectrometer samples roughly 380 to 2500 nanometres in about 285 channels at around 7.4 nanometre spacing. Because the channels touch each other, every pixel produces a near continuous reflectance curve rather than a short list of values.
Why it matters
The practical consequence is the difference between detection and identification.
Broad bands average everything inside them. A narrow absorption feature ten or twenty nanometres wide gets diluted into a band two hundred nanometres wide and largely disappears. Multispectral data is very good at telling you that something changed, that vegetation is stressed, that a surface is bright or dark, that water is present. It is poor at telling you which specific material caused the signal, because many different materials produce similar broadband responses.
Contiguous narrow bands preserve the shape of the curve, including the position, depth and asymmetry of individual absorption features. Those features are what make a material identifiable. Kaolinite and alunite both look like pale rock in a multispectral image. In a hyperspectral image they have distinct absorption features near 2.2 microns that can be separated and named.
The tradeoff
Hyperspectral data costs more per pixel in every sense. Splitting light into hundreds of narrow channels means less energy reaches each detector element, so instruments compensate with larger pixels or longer integration times. Coverage is usually narrower and revisit slower. Files are one to two orders of magnitude larger.
The right question is not which is better but which answers your question. If you need change over wide areas at high cadence, multispectral wins. If you need to know what a material is, you need the curve.