BENCHMARKS
How XCX performs
Three suites. Twenty-two languages and runtimes. Every result measured on the same box, the same day, no cherry-picking.
ProcessorAMD Ryzen 7 5800X (3.8 GHz)
Memory32GB DDR4 @ 3200 MHz
Operating SystemWindows 11
←Back to Home
01 · REAL-WORLD KERNELS
Main Suite
Fibonacci recursion, a linear congruential generator, the Sieve of Eratosthenes, and JSON encoding, ranked by geometric mean.
| # | Language / runtime | FIB (30) | LCG (100M) | SIEVE | JSON |
|---|---|---|---|---|---|
| 01 |  ZigAOT | 1.66ms | 5.28ms | 11.97ms | 0.04*ms |
| 02 |  RustAOT | 1.81ms | 21.36ms | 12.70ms | 0.06*ms |
| 03 |  CAOT | 1.19ms | 85.34ms | 8.72ms | 0.07*ms |
| 04 |  C++AOT | 1.26ms | 86.33ms | 18.27ms | 0.09*ms |
| 07 |  GoAOT | 3.10ms | 242.37ms | 16.16ms | 0.17*ms |
| 08 |  JavaJIT | 2.64ms | 217.88ms | 24.38ms | 0.18*ms |
| 11 |  XCX 4.4.1JIT | 6.324ms | 104.507ms | 28.736ms | 0.519ms |
| 16 |  C#JIT | 5.66ms | 208.82ms | 15.94ms | 5.58ms |
| 17 |  Node.jsJIT | 6.70ms | 1363.96ms | 29.42ms | 0.63ms |
* No built-in JSON parser in the standard library; result includes the Stdlib Penalty.
† XCX 3.0 / 3.1 json values come from the pre-4.2 benchmark suite (different methodology) and are not directly comparable with current results.
METHODOLOGY
How we measure
All benchmarks follow strict rules for computational equivalence, dead-code-elimination prevention, idiomatic implementations, and environment-specific execution procedures (AOT / JIT warmup / interpreted). The JSON column applies a "Stdlib Penalty" for languages without a built-in JSON parser (marked with *), and Main Suite results are aggregated using the Geometric Mean.
XCX 4.4.1 JIT results are medians from 50 warmup runs followed by 250 measured runs for Main Suite and Func & Arith, and 15 warmup runs followed by 35 measured runs for each Loop Suite benchmark. The loop total is 1544.30 ms, down 1.6% from XCX 4.4 (1570.17 ms). Cross-function calls improved by 48.5%.
XCX 4.4.1 interpreter (--no-jit) results use LONG mode: 5 warmup runs followed by 15 measured runs for Main Suite and Func & Arith, and 1 warmup run followed by 5 measured runs for each Loop Suite benchmark. Results are medians. The interpreter loop total is 93440.10 ms, up 0.5% from XCX 4.4 (92989.04 ms).