You'd want to look into either the optical end or electrical end. Probably you are mentioning the electrical end here, and may already have a setup, but mentioning some points either way.
1. The first step is to have the original spec. And the compliance test document MOI (Method Of Implementation) either from Agilent, Tektronix, TD Lecroy or R&S etc. Unfortunately it is the only commercial companies who have dedicated expertise in assessing the complexities in high speed measurements. And MOIs are published by these companies for most High Speed Serial/HSS standards.
(Well, semiconductor and Test & Measurement have a strange symbiotic relationship, and both ends contribute to the original standard! It's not like a semicon company proposes a standard and these companies build boxes for them. And specs are adjusted from both ends. it's all sensible business!)
There are 4 channels of apprx 2.5Gbps each, together forming 10G speed.
The choice of oscilloscope depends on high speed serial characteristic of the signal-
A. The spectral content of interest (Scope bandwidth)
B. The Rise-Time / Fall-time response (Scope or probe rise time)
2. Though bandwidth and rise-time are interrelated, the commercial spec on the scope bandwidth does not always promise high rise-time response. However high rise-time response always gives promising bandwidth.
Better bandwidth spec would offer accuracy in vertical measurements and complements the available dynamic range (set by vertical scale on panel, in turn on ADC hardware and amplifiers).
Though high bandwidth offers basic timing measurements like UI/ unit interval or period measurements, this alone is not good-enough for constructing EYE diagrams and decomposing the Jitter components.
(Random Jitter + Deterministic Jitter)
For cable testing / characterization and Jitter measurements, you'd need lot of spectral response and rise-time response too. We need to go with another beast called Sampling type oscilloscope. It captures just ONE complete waveform/s on multiple triggers and gives a super-high time resolution capture. It actually does some math magic inside with a lot of waveform captures at regular sampling rate (like a comparable DSO) and reassembles it to appear super-high sampling rate.
3. Plus one would need ready-made compliance package (like SFP+ compliance application for Infiniium of Agilent/Keysight or the fancier TekExpress package of Tektronix) Each costing a few grands.
Or develop your own compliance measurement analysis algorithms in Matlab, LabVIEW or Python for each measurement, which is taller order!
(Because each of those measurements are not simple and direct. They are not directly available on scope display. There needs further detailed analysis on the captured waveform, spectrum or EYE etc. with proper modeling technique)
The commercial ads you mentioned are for a reason. Unfortunately none of these measurements can be achieved by simple lab oscilloscopes for a pi testing. There is a category high performance scopes. And they cost a lot.
I'd suggest DSA91204A or DSA8200 type sampling scope that will cover most measurements. Or DPO/DSA 70000 to cover a few of them