The previous monolithic active grating bender design met some basic design requirements. However, after a real grating (BM-AGM) had been fabricated and installed for testing, the results showed that the usable length is a mere 60 mm because of the higher-order term error in the surface profile. A method was thus derived to eliminate the higher-order term error by modifying the width of the bender substrate through finite-element method simulation, reducing the residual error from about 100 nm to 6 nm. Owing to the closure of the grating department of Zeiss, ruling the monolithic bender is no longer available and the design has to be modified to a composite-type bender with Si substrate. A prototype was fabricated and assembled to examine all the design situations. The surface roughness of the width-modified Si substrate is around 30 nm before assembly. The residual error after assembly and bending is less than 10 nm. It proves that the design is feasible. However, due to the manufacturing capacity of the vendor, a short-length substrate is required and the design has to be modified. The detailed design modification and testing results are presented in this paper.