On October 16, 2026, Anthropic announced that its Claude Mythos 5 model autonomously designed quantum circuits. The model outperformed previous state-of-the-art designs in seven out of eight benchmark tasks across fields including chemistry, physics simulation, and number theory.
The system utilized a closed-loop evolutionary framework where Claude proposed circuit designs based on plain-language specifications. An automated evaluator verified the accuracy of the results and scored the circuits based on the number of Toffoli gates and qubits required.
In one specific case, the modular inversion architecture achieved up to 17 times fewer gates than previous approaches, although it required a higher number of qubits. The project was conducted by researcher Andi Gu under the Anthropic STEM Fellows Program.
The research involved collaboration with the Harvard Quantum Initiative and QuEra Computing. The findings suggest that AI models can assist research teams by iterating on circuit designs with minimal human intervention, helping to optimize resource usage on current hardware.
