Performance of WebAssembly Runtimes in 2026
7 hours ago
- #Benchmarking
- #Performance
- #WebAssembly
- Benchmark results for WebAssembly runtimes (2024-2026) show varying performance improvements in libsodium crypto workloads.
- Wasmtime consistently improved each year, from 2.67x native slowdown in 2024 to 2.41x in 2026 (baseline).
- Node showed gradual improvement, from 8.60x to 7.95x native slowdown over the three years.
- Bun made a significant leap in 2026, reducing slowdown from 26.42x in 2025 to 8.77x in 2026 (baseline).
- Wazero performance remained relatively flat across the years (~4.7x native slowdown).
- WAMR in AOT mode was fast, with 1.59x native slowdown in 2025 and 1.57x in 2026 (baseline).
- Wasmer's baseline improved modestly, but with wide_arithmetic support in 2026, it achieved the best result at 1.33x native slowdown.
- wasm2c remained a strong option, with slowdowns around 2x native and improving slightly in 2026.
- WasmEdge performed well but required explicit AOT mode (--run-mode=aot) for optimal results.
- Experimental wide_arithmetic WebAssembly instructions provided substantial speedups for crypto code where supported (e.g., Wasmtime and Wasmer).
- Runtime performance depends heavily on enabled WebAssembly features, runtime choice, and deployment model.
- Benchmarking actual workloads is recommended due to variability across runtimes and feature sets.