https://vislearn.github.io/PrimeDepth/
https://github.com/vislearn/PrimeDepth
https://paperswithcode.com/sota/monocular-depth-estimation-on-eth3d

I saw this months ago, but never got around to trying it. There's a webiste that benchmarks depth map models and this is at the top of one of the categories. PrimeDepth and above it is a combination of PrimeDepth+DepthAnything. The site doesn't have Depth-Anything v2 or Apple's DepthPro though, so I'm not sure how it would stack up against those

PrimeDepth looks worth testing and comparing

HybridDepth ranks #1 in several categories supposedly and #2 in another
https://paperswithcode.com/paper/hybriddepth-robust-depth-fusion-for-mobile-ar

Well ahead of other models:
https://paperswithcode.com/sota/monocular-depth-estimation-on-nyu-depth-v2

I'm not sure how though, because the examples I've seen of HybridDepth don't look so great and accurate.
HybridDepth v2 released in October

https://github.com/cake-lab/HybridDepth