LLM-designed quantum circuits with fewer gates and qubits than published designs in seven of eight tasks
Claude, run in a loop with an automated evaluator, produced reversible circuits that cut both Toffoli and qubit counts below the best published designs in seven of eight quantum-computing primitives and found a new reversible architecture for prime-field modular inversion in the eighth, including an elliptic-curve point addition with 17 times fewer gates at 12% more qubits.
- Lab
- Anthropic
- Model
- Claude (the announcement names Claude Mythos 5)
- Field
- Computer science
- Date
- 2026-10-15
- Human collaborators
- Andi Gu, Alek Kemeny, Susanne Yelin, Mikhail Lukin
What was found
The tasks span numerical computation, chemistry, optimization, quantum simulation and number theory, and each was posed as a plain-language specification with an external program that checks a circuit's correctness and scores it by qubits times gates. Claude proposed designs, received the score and revised, with no human input between rounds. For elliptic-curve point addition, the core step of quantum discrete-logarithm attacks, it found a reversible implementation of the divsteps modular-inversion algorithm that stores no branch history, checked the key identity by simulation and in Lean, and extended the Pareto frontier against Schrottenloher's and Luo and colleagues' 2026 designs. For the Reed-Solomon decoder inside Decoded Quantum Interferometry, the announcement reports about 12 times fewer gates than Google's construction and a locator step needing 19 qubits instead of a worst-case 665, tested on 403 million sample decodings with a failure rate of about one in 10,000.
Novelty check
The baselines are the ones the announcement and paper name: Schrottenloher (arXiv:2606.02235) and Luo et al. (arXiv:2607.13816) for point addition, and Khattar et al.'s 2025 decoder for DQI. Those comparisons were not re-derived here, and the other six tasks' baselines were not checked. A web search on 2026-10-20 found earlier work using language models to search for quantum circuits, including evolutionary state-preparation searches, but none claiming these primitives.
Caveats and known objections
A lab announcement with an unrefereed arXiv preprint, so claimed. The evaluator checks each circuit's correctness on test inputs, which is strong evidence for these deterministic tasks but not a proof for every input, and only the divsteps identity is said to be Lean-checked. The gains are not uniform: the headline point-addition design trades 12% more qubits for fewer gates, and the DQI decoder accepts a small failure rate. Autonomy is search-scaffold: the paper describes an evolutionary framework in which language models propose designs and an independent evaluator scores them, and the humans wrote the specifications and the evaluator, required several qualitatively different approaches in parallel, and occasionally prompted broader search. The announcement links no code.
Nobody outside the lab has checked this yet.
Reading the primary source closely enough to say whether it supports the claim counts as a check, and you are credited on the entry.
Or on GitHub: submit a check challenge the grade send a correction or send a pull request
Flag this for triage
Signals order the review queue and nothing else. They are never published, and they never move a grade: that takes a citation.
Entry history (1 event)
- AddedEntered the registry graded Claimed and Search scaffold.
Entries are never deleted. A grade that does not hold up is downgraded on the record, with the reason beside it.
Graded claimed for verification and search scaffold for autonomy. What these mean.
Cite this entry
whataifound.org. (2026). LLM-designed quantum circuits with fewer gates and qubits than published designs in seven of eight tasks. whataifound.org: A Registry of AI Scientific and Mathematical Discoveries. https://whataifound.org/finding/2026-10-15-llm-quantum-circuit-designBibTeX
@misc{whataifound-anthropic-2026-design,
title = {LLM-designed quantum circuits with fewer gates and qubits than published designs in seven of eight tasks},
author = {{whataifound.org}},
year = {2026},
howpublished = {whataifound.org: A Registry of AI Scientific and Mathematical Discoveries},
note = {Result by Anthropic. Verification: Claimed. Autonomy: Search scaffold.},
url = {https://whataifound.org/finding/2026-10-15-llm-quantum-circuit-design}
}