Insider Brief

• QC Design’s Meridian AI system reported lower logical error rates than published methods and a general-purpose AI agent across more than 100 fault-tolerance design tasks.
• The evaluation covered 10 quantum error-correction code families, different hardware layouts and error models representing major quantum computing platforms.
• Meridian achieved a median logical error-rate reduction of more than 10x against five published algorithms and more than 40% against a general-purpose GPT-6 Astra-based agent, with improvements reaching 98.4%.

PRESS RELEASE — QC Design today published “Over 10x reduction in logical error rates with Meridian,” a white paper showing that Meridian, its purpose-built AI system for fault-tolerant quantum computing design, achieved lower logical error rates than published methods and general-purpose AI agents across its entire suite of over 100 fault-tolerance design tasks. Logical error rate measures how often protected quantum information fails.

A quantum computer’s architecture determines its hardware needs and runtime. Better choices can save years of development and millions in capital expenditure. Finding the best architectures requires expertise spanning computer science, quantum error correction, device physics and engineering. Today, only a small number of people have this kind of expertise.

Meridian expands what quantum hardware teams can explore. Its specialized AI scientists combine frontier models, human-curated knowledge and validation tools from Plaquette, QC Design’s quantum design-automation platform. Plaquette serves as an accurate world model, testing candidates under realistic hardware imperfections so Meridian can distinguish genuine improvements from designs that score well for the wrong reasons.

“Every quantum manufacturer we meet is working on designing the best architecture for its hardware,” said Dr. Ish Dhand, co-founder and CEO of QC Design. According to co-founder Prof. Martin Plenio, “Meridian combines AI-driven exploration with the modeling and validation needed to solve that design problem. It raises the ceiling of what a hardware team can design and thus build.”

The white paper evaluates one part of architecture design: the circuits that detect errors in protected quantum information. The 100+ task suite spans 10 error-correction code families, varied hardware layouts and error models representing the major quantum computing platforms.

Meridian outperformed established methods from the scientific literature, achieving a median reduction of over 10x in logical error rates across the evaluation suite. For each task, QC Design compared Meridian with the best-performing circuit produced by five leading published algorithms. Compared with a general-purpose agent based on OpenAI’s GPT-6 Astra, Meridian achieved a median reduction in logical error rate of over 40%, with improvements reaching 98.4%, equivalent to an approximately 63x lower logical error.

QC Design believes the approach behind Meridian could apply beyond quantum computing. Many scientific problems require both the ability to generate candidate solutions and the expertise to judge whether they work. Meridian is an early example of an AI system that brings these capabilities together, helping researchers investigate promising solutions that might otherwise go unexplored.

Read the Meridian white paper: https://qc.design/meridian/white-paper.