Short answer: Magic state cultivation is a 2024 proposal for making the non-Clifford resource states that fault-tolerant machines need, by growing one state in place inside a surface code patch instead of running large distillation factories. Its authors estimate it uses about an order of magnitude fewer qubit-rounds than earlier methods. A late 2025 preprint reports a first test on a superconducting processor. All of this is research, and the large-scale claims are estimates from simulation.
The problem
A December 2025 arXiv preprint (2512.13908) sets out the background: fault-tolerant quantum computing needs a universal gate set, and the necessary non-Clifford gates are a significant resource cost in most error correction architectures. Magic state distillation is the established way to supply the states for them, and it is resource-intensive.
The 2024 proposal
The 2024 arXiv paper (2409.17595) refines earlier ideas to prepare good T states. The authors call the construction magic state cultivation because it gradually grows the size and reliability of one state. They say it fits inside a surface code patch and uses roughly the same number of physical gates as a lattice surgery CNOT gate of equivalent reliability. They estimate the infidelity from injection to idling at distance 15 using a mix of state-vector simulation, stabilizer simulation, error enumeration and Monte Carlo sampling. Compared with prior work, they report an order of magnitude fewer qubit-rounds to reach logical error rates as low as 2 x 10^-9 under 10^-3 uniform depolarizing circuit noise, and 4 x 10^-11 when the noise is halved. They suggest further magic state distillation may never be needed in practice, which is their projection from these estimates.
Variants
A 2025 arXiv paper (2502.01743) describes cultivation as using the transversality of a gate on the two-dimensional triangular colour code, with a grafting step that turns the colour code into a matchable code, and notes the resulting code has a longer cycle time and some high-weight stabilizers. It introduces three new protocols, each giving a different magic state, which avoid grafting by using transversal operations on the surface code with non-local connectivity, allowing much lower post-selection rates. Through numerical simulation the authors report state-of-the-art infidelities and acceptance rates, and say that on platforms such as cold atoms and trapped ions, where idle error rates are lower than two-qubit gate errors, cultivation shows a further order-of-magnitude reduction in resource requirements.
A first experiment
The December 2025 preprint reports an experimental study of cultivation on a superconducting quantum processor, including code-switching into a surface code and a fault-tolerant measurement protocol to bound the state fidelity. The authors report error reduced by a factor of 40, a state fidelity of 0.9999(1), and 8% of attempts retained after post-selection. They conclude that the work experimentally establishes cultivation as a viable solution. That is their own conclusion, from a single small experiment.
Limits
The resource comparisons come from the authors' own simulations at stated noise levels, and the experiment is a preprint at small scale. They do not show cultivation running inside a full fault-tolerant computation.
FAQ
What is magic state cultivation?
A protocol that grows one magic state gradually inside a surface code patch, per the 2024 paper.
How does it compare with distillation?
The authors estimate an order of magnitude fewer qubit-rounds for the reliabilities they study.
Has it been tested?
A December 2025 preprint reports an experiment on a superconducting processor with a state fidelity of 0.9999(1) and 8% of attempts retained.
Does it remove distillation entirely?
The 2024 authors suggest it may never be needed in practice. That is a projection.
Which platforms might gain most?
The 2025 variants paper says cold atoms and trapped ions, where idle errors are lower than two-qubit gate errors.
All field notes