clifft 0.4.0

Clifft - fast exact simulator for near-Clifford quantum circuits

Clifft is a fast exact simulator for near-Clifford quantum circuits.

Built and maintained by the Unitary Foundation.

Clifft accepts Stim-format circuits, extends them with non-Clifford gates, and compiles them into bytecode executed by a high-performance Schrödinger Virtual Machine. It is designed for circuits whose dominant structure is Clifford, but whose behavior depends on localized non-Clifford operations.

The main simulation cost scales with the active dimension k of the dense state vector, rather than directly with the total number of physical qubits n. Non-Clifford operations can increase k, while measurements can reduce it.

Why Clifft?

• Stim-compatible format and API: parse Stim-format circuits with noise, detectors, observables, and repeat blocks, plus non-Clifford extensions.
• Exact near-Clifford simulation: simulate localized non-Clifford effects without approximating the quantum state.
• Optimizing compiler pipeline: compile once, then sample many shots with HIR and bytecode optimization passes.
• Active-dimension scaling: for low-magic circuits, runtime and memory scale with the localized active state rather than the full Hilbert space.

For QEC workflows, Clifft also supports detector-based post-selection, survivor sampling, and stratified importance sampling for rare-event estimation.

Installation

pip install clifft

Platform / CPU family	PyPI wheel
Linux x86_64 with AVX2	Supported
Linux aarch64	Supported
macOS arm64	Supported
Windows amd64	Supported

All other platforms and CPU families should build from source. See the installation docs.

Quick Start

import clifft

program = clifft.compile("""
    H 0
    CNOT 0 1
    T 2
    M 0 1 2
""")

result = clifft.sample(program, shots=1000, seed=42)
print(result.measurements[:5])

For more details and examples, check out the documentation or take Clifft for a spin in the web-based interactive playground.

Performance

Clifft is designed for near-Clifford circuits where non-Clifford activity remains localized. In this regime, the dominant cost scales with the peak active dimension k, not directly with the total number of physical qubits.

Regime	Representative benchmark	What the results show
Pure Clifford QEC	Surface code d=7, r=7 ▶↗	Stim remains the right tool; Clifft is roughly 10× slower while preserving the same sampling-oriented workflow.
Low-magic FT circuits	MSC d=3 cultivation ▶↗	Clifft reaches 10.4M shots/s, about 370× faster than Tsim on this benchmark.
Larger near-Clifford FT circuits	MSC d=5 cultivation ▶↗	Clifft reaches ~135K shots/s on one CPU core, about 13× faster than SOFT at ~10.6K shots/s on one H800 GPU.
Dense universal circuits	Quantum Volume	In the worst-case dense limit, Clifft remains neck-and-neck with simulators like qiskit-aer and qsim.

Throughput numbers above were measured on cloud instances; the links to the in-browser WASM playground will report lower throughput.

For benchmark details, plots, hardware notes, and guidance on when Clifft is a good fit, see the performance section of the documentation.

The full methodology and scientific results are described in the Clifft paper and companion clifft-paper repo.

Citation

If you use Clifft in your work, please cite the arXiv preprint below.

@misc{chase2026clifftfastexactsimulation,
      title={Clifft: Fast Exact Simulation of Near-Clifford Quantum Circuits},
      author={Bradley A. Chase and Farrokh Labib},
      year={2026},
      eprint={2604.27058},
      archivePrefix={arXiv},
      primaryClass={quant-ph},
      url={https://arxiv.org/abs/2604.27058},
}

Development

See the building from source guide for build instructions.

AI Acknowledgement

We used generative AI tools during parts of the research, software-development, and writing workflow for this project. These tools assisted with code generation and review, implementation analysis, documentation editing, and checks of selected derivations or arguments. All substantive design, validation, and release decisions were made by the human contributors.

Funding

This work was supported by the U.S. Department of Energy, Office of Science, Office of Advanced Scientific Computing Research, Accelerated Research in Quantum Computing under Award Number DE-SC0025336.

This material is also based upon work supported by the U.S. Department of Energy, Office of Science, National Quantum Information Science Research Centers, Quantum Science Center.

License

Apache-2.0