That is a very good question. At this point, a hash function in the SHA-2 family is generally considered secure.
MD5 has been known to be cryptographically insecure since about 2008. Collisions can be reliably reached in sub-second timeframes on hardware that is over a decade old. It also has many other attack vectors. The only place that it really could reasonably be used is when checking for file integrity for an rsync or the like but even then, with modern hardware, there’s little reason to not use a secure hashing algorithm.
For SHA-1, successful collisions were hit in under 2^69 ops as early as 2005.
In 2017, Big G (when they were still trying but to be evil) announced the SHAttered attack that that reliably reached collisions with 2^63.1 ops. SHAttered required 6500 CPU-years and 110 GPU-years to implement but that’s a number well within reach for a well-funded adversary. Several other attacks from other directions have been proven out with the barrier to entry getting significantly lower. It doesn’t even take a state actor anymore with costs being estimated as low as $45k USD in 2020.
SHA-2 has not yet had any publicly disclosed success in defeating all hashing rounds. Last year, there was success in collision in 31/60 rounds for SHA-256 and 31/80 rounds for SHA-512. So, it’s generally thought to still be secure (noone has had yet disclosed a practical collision or pseudo collusion that is close to defeating ALL rounds).
EDIT: Newlines to avoid formatting (how do I escape formatting characters?)