And all that with 4-bit CPU (!) which worked for a very long time on very small batteries.
I'd extend that to the whole HP calculator lineup through about 1994 or 95.
There are a lot of aspects I could single out, but William Kahan's involvement in their numerical algorithms is a particular standout for me. They wanted high quality numerics out of battery powered 1970's era micro-electronics, so they brought Dr. Kahan in to design the algorithms necessary for the task.
(Dr. Kahan is also noteworthy for his involvement in IEEE-754, which is still the standard for floating point today.)
Serious, serious engineering went into these calculators.
For me, at least, probably the combination of the 48GX and 42S. Both represent the end of the line of HP's traditional hardware engineering, which is such a significant part of using these devices. The software gaps between the 48GX and the later 49 can be made up by installing software into a 48, if that turns out to matter for you.
One of the more significant new parts of the 49 software is the "Meta Kernel", which was externally developed by Jean-Yves Avenard as a 48 extension. IIRC, Avenard ultimately wound up working for HP on their calculators, which I believe is how MetaKernal wound up as part of the core 49G ROM. IIRC, this was also a bit of a reconstitution of HP's calculator engineering effort, as the previous team had been disbanded a few years prior. (ie; William Wickes, who was heavily involved in the 41 and 28/48, wound up at some point working on video conferencing systems for HP.)
The reason I'm also including the 42S is that it's a lot more traditional in the way that it operates. If the 28/48/49/50 series feels like a small handheld computer with a custom programming language, the 42 feels more like an actual calculator (with a lot of capabilities).