I don't think people have given up on model based RL, it is just that describing a proper model is (like you are saying) very difficult.

in the case you haven't seen or read the following: https://bair.berkeley.edu/blog/2019/12/12/mbpo/