IBM brings native Qiskit development to Fortran, C++ and Julia

IBM now lets developers use Qiskit natively from Fortran, C++ and Julia through a shared C interface to the quantum software development kit’s Rust core. The new language paths allow them to build and work with quantum circuits in those languages without running a Python interpreter between their applications and Qiskit.

The interfaces are aimed at scientific-computing and high-performance-computing codebases whose compute-intensive components are already written in Fortran, C++ or Julia. IBM said a quantum operation can be linked into an existing application as a subroutine, rather than orchestrated through a separate Python process or embedded interpreter. For broader context on the company’s developer tooling, IBM recently added self-hosted deployment for its Bob development platform.

The underlying Qiskit C API provides a defined application binary interface to the same Rust data model used by Qiskit’s Python API. Developers can build it as a standalone shared library, allowing a compiled-language program to create and manipulate Qiskit objects without a Python runtime dependency. IBM said the bindings share that library and should therefore be interoperable in principle, such as constructing a circuit in Fortran and passing it to C++.

The Fortran interface uses the language’s ISO_C_BINDING mechanism to call the C API directly. Its QuantumCircuit derived type automatically releases memory when it goes out of scope. The project currently requires a Fortran 2018-capable compiler, supports macOS and Linux, and labels itself experimental and unstable.

The C++ interface is header-only and supports C++11 or later. Its documented scope includes circuit construction, transpilation and sampler jobs. Access to IBM Quantum hardware requires a separate supported service interface such as qiskit-ibm-runtime C, QRMI or SQC.

The Julia package wraps the C API in native Julia types and is available through Julia’s package manager. It currently supports circuit and target construction and manipulation, plus transpilation, on Linux and macOS. IBM’s example combines it with QiskitIBMRuntime.jl to construct, transpile, submit and post-process a circuit without leaving Julia.

IBM did not provide comparative latency, memory or throughput benchmarks for removing the Python layer, or adoption figures for the bindings. All three remain under active development, while the C API documentation describes the interface as experimental and subject to change between minor releases.