Understanding Python's Integer Caching and Reference Counting

In CPython, integers are not stored in fixed 2, 4, or 8‑byte slots; they are represented as arrays of base‑2³⁰ digits, which allows arbitrarily large values. Each integer is a C struct, and because there is no explicit length limit, Python can handle very large numbers without overflow, though arithmetic operations are relatively slower.

To improve performance, CPython pre‑allocates singleton objects for all integers in the range –5 to 256 during interpreter initialization. When a value in this range is needed, the interpreter reuses the existing object instead of creating a new one, saving memory allocations and speeding up common operations.

The singleton behavior can be verified with the built‑in id() function: two variables holding the same small integer have identical IDs, while larger integers produce distinct IDs. Visual examples show that expressions like 2 + 4 and 3 + 3 both yield the same object ID for the result 6.

Reference counting, accessed via sys.getrefcount() , reveals that small integers are referenced far more often than larger ones. A distribution chart for integers –5 to 300 shows high reference counts for the smallest values, with peaks at powers of two (32, 64, 128, 256). The analysis estimates that about 1,993 small‑integer objects are avoided during interpreter startup.

These findings confirm that Python’s integer caching mechanism reduces both memory usage and allocation overhead, especially for frequently used small integers, which is a fundamental optimization in CPython’s implementation.