foldhash

The default hasher for hashbrown 0.15+, replacing ahash — a quietly significant ecosystem shift since hashbrown underlies std::HashMap. foldhash is faster than ahash across the board with a smaller HashMap footprint (40 bytes vs 64). With foldhash in place, Rust's stdlib map matches abseil-flat-hash-map within ~5%, closing a longstanding gap to C++.

Why the hasher choice matters more than the table

The deepest insight in 2025's hash map landscape is that hash function choice often dominates table choice. Switching from SipHash (Rust's safe default for hash-flooding resistance) to foldhash yields 2–5× speedup in hash-heavy workloads — larger than any difference between Swiss Table implementations. The Rust compiler itself got a 6% overall speedup from moving to fxhash. By comparison, boost-unordered-flat-map beats abseil-flat-hash-map by ~10–30% on most operations.

This recontextualizes "fastest hash map" benchmarks. For workloads dominated by hashing time (short keys, simple types), the hasher swamps the table. See fastest-hash-map-2025 for the full picture.

Performance profile

foldhash is faster than ahash across the board and faster than rapidhash for integer tuple hashing specifically (0.62 ns vs 0.85 ns). However, rapidhash still leads the overall geometric mean benchmark (4.25 ns vs 4.79 ns). For standard HashMap usage where you don't control the hasher, foldhash is what you get — and it is good.

The ahash succession

ahash was the previous hashbrown default and accumulated massive adoption. foldhash replaced it by being both faster and more memory-efficient. ahash should be considered legacy for new projects.

When to pick something else

• rapidhash — fastest overall geometric mean; right for custom hash structures where you control the hasher.
• gxhash — excels at long-string hashing but requires AES hardware and isn't portable.
• SipHash — keep when adversarial input is a real threat (e.g., user-controlled keys directly hashed). foldhash is not a hash-flooding-resistant hasher in the cryptographic sense.

For more on the broader hashing landscape, see rapidhash.