B Grade B
Google Willow: Logical error rate
7.72 × 10⁻⁴ error per cycle
As stated in the source: “7.72(9) × 10^-4 per cycle (distance-7 surface code)”
Uncertainty: 0.000763 to 0.000781 (Stated in parenthesis notation as 7.72(9)×10^-4; the type of interval is not stated in the part read.).
Caveats
Preprint, not peer reviewed (no journal reference listed on arXiv at v4, 19 Jun 2026). Quantum memory only, not logical computation. Result depends on a neural-network decoder and RL tuning.
- Set-up
- Distance-7 surface code on Willow, decoded with AlphaQubit2; control parameters fine-tuned by reinforcement learning on top of conventional calibration.
- Method
- Surface code memory, logical error per cycle, AlphaQubit2 neural-network decoder, after reinforcement-learning control fine-tuning
- Compared only with
- results in the group “surface code d7 memory”
- Measured
- Not stated
- Published
- 11 November 2025
- Measured by
- Google Quantum AI and Google DeepMind
- Independent of the maker
- No
- Peer-reviewed
- No
- Repeated by others
- Not yet
Source
Reinforcement Learning Control of Quantum Error Correction, arXiv. Preprint. DOI: 10.48550/arXiv.2511.08493
Comparable results
| Machine | Value | Set-up | Published | Measured by | Grade | Record |
|---|---|---|---|---|---|---|
| Google Willow | 1.43 × 10⁻³ | surface code d7 memory | 9 Dec 2024 | Maker | B | r-000001 |
History
- 27 October 2026: added
Spotted a mistake? Report an error, quoting r-000002.