On October 15, 2026, Google Quantum AI announced a demonstration of magic-state cultivation on its Willow processor, reporting an error reduction of about 40-fold compared with uncultivated states. A preprint dated December 15, 2025, gives the experimental detail: near θ = π/4, the logical error probability was 4.4(2) × 10⁻³ with direct state injection and 1 × 10⁻⁴ to 2 × 10⁻⁴ for the full cultivated circuit. The researchers also reported a T-state fidelity of 0.9999(1), while about 8% of attempts were retained at that angle. (Google Quantum AI; research preprint)
What magic-state cultivation does on Willow
A magic state is a quantum resource used to implement a T gate, a non-Clifford operation needed for universal fault-tolerant quantum computing. Cultivation starts with an injected state that contains errors. Researchers then use fault-tolerant measurements and post-selection—keeping attempts with the desired result and discarding those with detected errors or the wrong outcome—to prepare a higher-fidelity T state.
The error reduction comes with heavy post-selection
The reported error values describe two different experimental conditions: direct injection alone and the full circuit with cultivation. At θ = π/4, the paper reports that about 8% of cultivation attempts were retained. The high-fidelity result therefore came from selecting a small share of attempts, rather than retaining every one.
| Experimental condition | Logical error probability near θ = π/4 | What was measured |
| Direct state injection | 4.4(2) × 10⁻³ | Injection-only logical error |
| Cultivation | 1 × 10⁻⁴ to 2 × 10⁻⁴ | Logical error for the full cultivated circuit |
The fidelity figure and the error probabilities describe different quantities: the preprint reports T-state fidelity of 0.9999(1), while the values in the table are logical error probabilities. Neither is a generic measure of a quantum computer’s overall accuracy.
A code transition, with more engineering ahead
The experiment used a distance-3 color code to cultivate the state, then demonstrated grafting it into a distance-5 surface-code encoding. The color code supports the cultivation procedure; the surface code is the destination for later fault-tolerant processing. The researchers used Willow, a 105-qubit processor with tunable transmon qubits arranged in a square lattice.
Google says cultivation still needs to be demonstrated at larger scale. The paper identifies real-time T-gate implementation, larger code distances and further decoder work as next technical steps.
