move_to_end(key) is the operation a hand-rolled LRU cache needs every time a key is read or written: bump it to the most-recently-used end. The companion popitem(last=False) evicts the least-recently-used entry in O(1). Plain dicts do not expose either method, which is the main reason OrderedDict still earns its keep in 2025. The same pair powers Python's functools.lru_cache under the hood.

Two OrderedDicts are equal only if they have the same keys, the same values, AND the same insertion order. Two regular dicts are equal even with different orders. Mixed comparisons (OrderedDict vs dict) drop down to the dict semantics. This stricter equality is exactly what you want when testing JSON serialisation, comparing parsed YAML, or asserting that a stable order is preserved through a transformation. It is also a sharp edge: never assume OrderedDict == OrderedDict matches dict == dict.

The 12-line LRU above is the smallest production-ready cache you can write in Python, and it is impossible without move_to_end and popitem(last=False). For most other cases (configs, request payloads, parsed query strings) plain dict is faster, more memory-efficient, and IDE-friendlier. Reach for OrderedDict when the order itself is part of the contract or when the LRU primitives are required. Otherwise default to dict.