The journal Nature published a study by Microsoft Quantum and ion quantum computer developer Quantinuum on reducing logical errors in a quantum processor. The teams reported improvements ranging from 11 to 800 times compared to comparable physical circuits.
Separately, IBM Research described an approach to discovering new quantum error correction codes using large language models (LLMs). The OpenEvolve-based system identified 465 candidates, but their practical applicability has yet to be verified.
Microsoft Quantum and Quantinuum’s Work
Error correction remains one of the main barriers to scaling quantum computers. Modern qubits are sensitive to noise and quickly accumulate errors, so long computations require logical qubits, decoders, and circuits that detect and correct failures during operation.
A physical qubit is the hardware unit of a quantum processor. A logical qubit is a more reliable unit in which error correction codes combine several physical qubits. This scheme is necessary for a quantum computer to perform long computations without losing results due to noise.
The article “Improved quantum processor logical error rates via correction and detection” describes the results of the joint work of Microsoft Quantum and Quantinuum. The experiment used two constructions optimized for the Quantinuum ion processor: a 12-qubit code inspired by the Knill scheme and a 16-qubit tesseract color code. The first encodes two logical qubits, the second four.
According to Microsoft, the schemes covered computations involving up to 12 logical qubits. During Bell state preparation, the logical error rate decreased from approximately 0.8% for the physical scheme to 0.001%, resulting in an 800-fold improvement.
Repeated error correction showed a result 51 times lower than the physical baseline per round. Preparing a 12-qubit cat state, a multi-qubit superposition state, resulted in a 22-fold improvement.
“Our results demonstrate that modern quantum devices are already capable of using fault tolerance and error correction to significantly suppress errors in non-trivial quantum circuits,” the article’s abstract states.
Microsoft also recalled previous joint results with Quantinuum: over 14,000 individual experiments without recorded errors, demonstrating 12 reliable logical qubits, and a hybrid chemical simulation using logical qubits, artificial intelligence, and high-performance computing.
IBM Research Utilizes AI for Code Discovery
IBM Research reported using OpenEvolve to search for quantum error correction codes. OpenEvolve is an open-source library that applies large language models for the evolutionary improvement of software code.
The team focused on bivariate bicycle codes. These are a type of quantum code with low-density parity checks, which IBM considers in its roadmap for fault-tolerant quantum computing.
The parameters of such codes are recorded in the format [[n,k,d]], where n is the number of physical qubits, k is the number of logical qubits, and d is the code distance. The higher the d, the more errors the code can withstand before losing utility.
Following initial runs, the system proposed 465 candidates. Among them, IBM highlighted the [[288,50,8]] code with 50 logical qubits, surpassing the previous record of 16 for this family. The company also noted the compact [[72,4,8]] code with 72 physical qubits and variants [[288,16,12]] and [[360,12,≤24]].
According to IBM, some candidates may be comparable to the [[144,12,12]] gross code under certain types of noise, which the company plans to use in fault-tolerant quantum computers. However, IBM emphasizes that the practical applicability of the discovered codes requires further verification.
The project’s source code qcode-discovery is published on GitHub. The OpenEvolve library is also available in an open repository.
In June 2025, IBM announced plans to build the IBM Quantum Starling by 2029—a large-scale fault-tolerant quantum computer with 200 logical qubits and 100 million quantum gates. The system’s architecture also relies on bivariate bicycle codes.
In June, Quantum X Labs and the research platform Quantum Machines IQCC reported plans to test an AI decoder for quantum error correction.
