XAD is the fastest automatic differentiation library for C++ and Python, built for production-scale systems. Proven in codebases of hundreds of thousands of lines, with 1,200+ tests and ~99% coverage.
| Repository | Description | Language |
|---|---|---|
| xad | Core AD library - forward & adjoint mode, JIT support, Eigen integration | C++ |
| xad-py | Python bindings for XAD | Python |
| QuantLibAAD | Full QuantLib integration - compute all Greeks at once, up to 3 orders of magnitude faster than bump-and-reval | C++ |
| QuantLib-Risks-Py | QuantLib risks from Python via XAD | Python |
| xad-codegen | High-performance native code generation backend for XAD - record-once/replay-many at maximum throughput (commercial) | C++ |
| ad-benchmarks | AD benchmarks on quant-finance workloads (LMM, Heston, CVA, SABR): XAD against CppAD, Adept 2 and autodiff, every gradient checked | C++ |
XAD is the fastest open-source C++ AD library on four quant-finance workloads, benchmarked against CppAD, Adept 2, and autodiff. XAD computes all sensitivities 6.5x to 83x faster than bump-and-reval, and the commercial code generation add-on (xad-codegen) a further 4x to 17x faster than XAD alone.
XAD v2.2.0, AMD EPYC (Zen 5), one core, GCC 15.2, run 10 November 2026. Full methodology and source: ad-benchmarks.
- Fastest AAD library for C++ - independently benchmarked
- Production-proven in large-scale quantitative finance systems
- Full QuantLib integration with 3 orders of magnitude speedup over bump-and-reval
- Eigen compatible - drops into existing C++ codebases with minimal changes
- 1,200+ tests, ~99% coverage
- Python bindings available via
pip install xad
If XAD is useful to you, please star the core repository — it helps others find it.