On August 24, 2026, IBM announced Qiskit Fermions, an open-source toolkit intended for researchers working on the simulation of fermionic systems in quantum computing. The package enables the representation of fermionic operators and circuits, the construction of transformations between fermionic systems and qubits, and the compilation of the workflow within the Qiskit transpiler.
Many quantum chemistry and materials physics problems involve interacting fermionic particles, while quantum computers execute circuits using qubits. IBM argues that converting a problem to a qubit representation early may result in the loss of some of the problem’s physical structure. Qiskit Fermions therefore preserves the fermionic representation during parts of the compilation process and performs synthesis into qubits only at a later stage of the workflow.
Deferred Conversion and Customizable Design
The package introduces the concept of a “fermionic circuit,” a representation similar to a quantum circuit but operating on fermionic modes rather than qubits. Researchers can select the encoding component that determines how the conversion to qubits is performed, as well as the method for approximating time evolution, or write their own components.
This approach relies on the multi-representation compilation framework introduced by Qiskit in version 2.5. The workflow begins with a fermionic operator and circuit, then synthesis into qubits takes place midway through the compilation process, before the workflow is completed using conventional quantum circuits. IBM says that keeping fermionic-symmetry information available for longer may enable optimizations that are not possible after the problem has been fully reduced to the qubit level.
An Example of Constant Depth
IBM demonstrates a one-dimensional Fermi-Hubbard model to explain the concept. In the example, hopping terms are grouped into “flow groups” before conversion to qubits. With the addition of one auxiliary qubit, an entire group of particle-transfer terms can be converted into single-qubit operations instead of being implemented through multiple entangling gates.
According to the measurement reported in the announcement, the two-qubit gate depth remained constant at 12 as the model grew from four sites to 100 sites. By comparison, the conventional approach based on Jordan-Wigner and naive time slicing reached a depth of 407 at 100 sites. IBM explains that this example requires a custom encoding, while the package also provides a Jordan-Wigner-based preset pass manager to simplify common use cases.
Its Place Within the Qiskit Ecosystem
Qiskit Fermions integrates with ffsim, a high-performance simulator that can execute some circuits at the fermionic level more efficiently when the circuit preserves particle number. The resulting circuits can also be used with the SQD add-on, which processes samples taken from quantum hardware to estimate eigenvalues, such as a molecule’s ground-state energy.
The package currently includes a Python interface for fermionic circuits and synthesis features, while IBM plans to support C in a future release. The company also points to plans to connect the tool more deeply with chemistry and materials-science software, and to enable the circuit library to be ported to Rust and then to C as Qiskit’s support for custom gates and circuit types develops.
Why Does This Announcement Matter?
The practical change is not merely the introduction of a new simulator, but the provision of a software layer that separates the description of a fermionic problem from the way it is implemented on qubits. This may give researchers more room to test new encodings and algorithms without rebuilding circuits manually. However, the constant-depth results presented apply to the Fermi-Hubbard example and an encoding based on flow groups, and do not automatically establish a comparable improvement for all chemistry or materials problems. In addition, some features remain limited to Python, meaning that the scope of integration with existing scientific software is tied to the development of the planned C interfaces.
The package is now available as an open-source project, along with documentation and examples for one- and two-dimensional Fermi-Hubbard models, a GitHub repository, and research papers related to flow-group encodings.