Claude Beat Quantum Physicists by 17x? Read the Fine Print

Source: Designing more efficient quantum circuits with Claude (Anthropic) · 2026-10-29; Discovering state-of-the-art quantum circuits with large language models (arXiv:2610.22367) · 2026-10-29

What Anthropic actually claims

Anthropic gave Claude Mythos 5 plain-language specifications for eight computational building blocks covering chemistry, optimization, physics simulation, number theory and numerical computing. The model worked in a closed loop. It proposed a circuit, an automated evaluator checked that it was correct and scored it on qubits multiplied by gates, and Claude revised the design with no human input between rounds. The headline task, elliptic-curve point addition, ran for 53 hours from a one-paragraph brief. According to Anthropic, the only human input was that brief plus occasional prompts to think more broadly.

Source: Designing more efficient quantum circuits with Claude (Anthropic) · 2026-10-29

Andi Gu led the work through Anthropic's STEM Fellows Program, working with researchers from Harvard and QuEra Computing. The paper's authors are Gu, Alek H. Kemeny, Susanne F. Yelin and Mikhail D. Lukin, and it was posted to arXiv on October 15. The abstract says that in seven of the eight tasks, the discovered circuits significantly reduced both Toffoli gate counts and qubit counts compared with the prior state of the art. In the eighth, the search produced a new reversible architecture for prime-field modular inversion.

Source: Interesting Engineering: Claude's new quantum circuit design uses 17 times fewer gates · 2026-10-29; Discovering state-of-the-art quantum circuits with large language models (arXiv:2610.22367) · 2026-10-29

Where the 17x comes from, and why it's narrower than it sounds

The key idea was a reversible version of divsteps, a classical algorithm for modular inversion. Instead of storing the history of every branch decision, Claude's design reconstructs most of them from intermediate values. The researchers checked the decoding formula with large-scale numerical simulations and formalized it in the Lean proof assistant. A threshold between storing and decoding produces a whole family of designs. According to Quantum Computing Report's summary of the paper's tables, other points in that family make the opposite trade: fewer qubits in exchange for more Toffoli gates. So '17x fewer gates' is one point on a trade-off curve, not a free win. Quantum Computing Report also notes that the comparisons are against hand-crafted baselines and that the figures come from a classical simulator.

Source: Designing more efficient quantum circuits with Claude (Anthropic) · 2026-10-29; Quantum Computing Report: Harvard, Anthropic and QuEra use LLMs to discover quantum circuits · 2026-10-29

Our take: this is a real result, but the press has been far too generous. Most stories so far repeat Anthropic's own post, and Crypto Briefing says plainly that it quotes no outside experts. Anthropic picked the eight tasks itself, and the multipliers vary a lot: the chemistry gain is about 4x, while one decoder figure goes past 9,100x. When one paper's numbers span three orders of magnitude, the headline is whichever number gets clicks. Wait for someone outside the team to rerun the circuits before calling this 'beating physicists.' Also keep in mind that physicists built the evaluator, chose the problems and wrote the paper.

The crypto question everyone will argue about

Anthropic notes that point addition is the core step in quantum algorithms for the elliptic-curve discrete logarithm problem, which a sufficiently large quantum computer could solve for curves where classical computers can't. The post also says the gains concern circuit designs and resource estimates. They don't mean large fault-tolerant quantum computers exist.

Source: Designing more efficient quantum circuits with Claude (Anthropic) · 2026-10-29

Expect 'AI just brought Q-Day closer' takes from crypto Twitter. That's overrated. Cheaper circuits do shrink the hardware you would eventually need, and that's a reasonable reason to keep up post-quantum migration plans. But nothing in this paper breaks a key today. The underrated story is the method: a model grinding for 53 hours against an automated checker. Most AI 'discoveries' fail on verification, and this one at least had verification built in.

What it means if you pay for Claude

Anthropic's Mythos page says Mythos 5 was 'currently only available to a small group of vetted partners' at its June 2026 announcement. The page lists Mythos 5.1 as limited to vetted cyberdefenders and life scientists through trusted access programs. The same page describes Claude Fable 5.1 as the same underlying model as Mythos 5.1, with safeguards for cybersecurity and biology. The page doesn't mention Pro or Max plans.

Source: Claude Mythos (Anthropic) · 2026-10-29

Source: Anthropic Consumer Terms of Service · 2026-10-29

Why this clause matters here: the quantum project worked because every design went through an independent evaluator, plus Lean proofs for the key formula. That is the terms' advice turned into a working method. Your subscription doesn't include the evaluator, so you have to supply it yourself.

Optional: see current Claude plan listings from a seller. This is a reseller listing, not official access to Mythos and not an endorsement.. That link carries our partner code Sk52g, which is how a purchase is attributed to this site. Promotion code UPD3S.

Claude · Compare price & terms →