Anthropic reports a Claude-designed quantum circuit with 17 times fewer gates for 12 percent more qubits
Anthropic says Claude designed a quantum circuit with 17 times fewer gates after a 53-hour experiment with Harvard and QuEra researchers. The cost is 12 percent more qubits than the smallest earlier design, so the saving helps only teams that can spare them.
The Product Desk
Estimated Toffoli gates to simulate the ground-state energy of FeMoco: roughly 50 trillion in the 2017 estimate, approximately 340 million after subsequent advances, a reduction of about 150,000 times.
| Measure | Value | As of | Claim |
|---|---|---|---|
| 2017 estimate, roughly | 50,000,000,000,000 Toffoli gates | 2017 | 9 |
| Later estimate, approximately | 340,000,000 Toffoli gates | After advances | 9 |
What happened
- Anthropic gave Claude eight computational building blocks to optimize, and an automated evaluator checked each design for correctness and scored its resource needs.
- Claude derived a way to recover divsteps branching decisions from intermediate numerical values, so the circuit no longer stores the full decision history.
- A separate optimization cut the costliest step of a leading quantum optimization algorithm by about 12 times.
- Researchers from Harvard University and QuEra Computing collaborated on the work, which Andi Gu led through Anthropic's STEM Fellows Program.
Why it matters
- What's now possible capability Hybrid designs that store some branching decisions and rebuild the rest let a team with a fixed qubit count choose how much memory to spend.
- What doesn't add up contradiction The outlet frames the trade as memory against processing time, but the figure it reports counts gates, so elapsed run time is not established.
- What it limits constraint Error correction, hardware controls and qubit management over long calculations remain open, so fewer gates alone does not make the algorithm runnable.
Anyone building a circuit for elliptic-curve arithmetic meets a storage bill at the first conditional step. Divsteps, the modular-inversion method Claude worked on, runs efficiently on a classical computer [13]. A quantum circuit cannot discard temporary information [3], so a divsteps circuit has to keep a record of its branching decisions, and that record can take enormous numbers of extra qubits [6]. Modular inversion matters here because it plays a central role in quantum algorithms aimed at the elliptic-curve discrete logarithm problem [5].
Claude tested its approach in numerical simulations and formalized its correctness in Lean, according to Interesting Engineering's account of Anthropic's experiment [7]. The reported saving, 17 times fewer gates than an earlier approach [15], comes from one hybrid design that uses 12 percent more qubits than the smallest earlier design [16]. Read as a straight 17-fold cut, that is about 6 percent of the gates [14]. The source does not name the earlier design that the 17 is measured against.
Here's what teams tell themselves about a result like this: fewer gates means a better circuit. Here's what the source supports: fewer gates for a team that can spare about 12 percent more qubits. Existing machines have only hundreds to thousands of qubits [12], so a team already near its limit cannot spend an extra 12 percent.
Anthropic argues that automated circuit optimization could help speed progress like the FeMoco reduction [10]. A 2017 estimate put the FeMoco ground-state simulation at roughly 50 trillion Toffoli gates, and later advances brought it to about 340 million, a drop of about 150,000 times [9]. Claude's run took 53 hours [1]. The FeMoco figures and the divsteps figures describe different circuits, so the 150,000 and the 17 cannot be set side by side.
We'd sort the claim with two facts about the reader's own project. The first is whether the team has about 12 percent more qubits than its smallest current design [16]. The second is whether gate count is what limits the circuit. With spare qubits and a gate-bound circuit, the reported point applies and the circuit is worth re-costing. With spare qubits and a circuit limited elsewhere, there is no reason to move. With no spare qubits, the reported point does not fit, and the only hybrid with published numbers is the 12 percent one [16].
What to watch
- Whether Anthropic or its Harvard and QuEra collaborators publish the circuit and name the earlier design the 17-fold figure is measured against.
- Whether elapsed run times or hardware results appear for the hybrid alongside gate counts.
- Whether figures for other points on the memory-versus-gates curve are reported.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence40
- Adoption
- Not scored
- Hype gap+30
- Incentives
- Not scored
- Confidence45
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
The design emerged from a 53-hour experiment led by researcher Andi Gu through Anthropic's STEM Fellows Program.
- [2]
Researchers from Harvard University and QuEra Computing collaborated on the work, in which Claude developed a reversible arithmetic algorithm for elliptic-curve calculations.
- [3]
Quantum circuits cannot simply discard temporary information, since their operations must remain reversible; storing intermediate results requires additional qubits.
- [4]
Anthropic tasked Claude with optimizing eight computational building blocks across chemistry, physics simulation, optimization, number theory and numerical computing, and an automated evaluator checked each design for correctness and scored its resource requirements.
- [5]
Modular inversion is a mathematical operation used in elliptic-curve arithmetic and plays a central role in quantum algorithms targeting the elliptic-curve discrete logarithm problem.
- [6]
A reversible circuit must preserve enough information to reconstruct every computational step, so divsteps requires a record of its branching decisions, potentially consuming enormous numbers of additional qubits.
- [7]
Claude developed a way to recover divsteps branching decisions directly from intermediate numerical values; it derived mathematical identities to establish the method, tested the approach through numerical simulations, and formalized its correctness using Lean. The resulting circuit eliminates the need to store the complete history of branching decisions.
- [8]
Claude explored hybrid designs that store some information and reconstruct the rest.
- [9]
Researchers estimated in 2017 that simulating the ground-state energy of FeMoco would require roughly 50 trillion Toffoli gates; subsequent advances brought the estimate to approximately 340 million gates, a reduction of about 150,000 times.
- [10]
Anthropic argues that automated circuit optimization could help accelerate progress similar to the FeMoco reduction.
- [11]
Quantum computers still need better error correction, reliable hardware controls, and systems that can manage fragile qubits during extended calculations.
- [12]
Existing quantum machines have only hundreds to thousands of qubits, and environmental disturbances can introduce errors during computation.
- [13]
Divsteps simplifies numerical values through repeated conditional operations, and classical computers can efficiently execute the method.
- [14]
If 17 times fewer gates means a 17-fold reduction, the circuit needs about 5.9 percent of the gates of the comparison design, roughly 6 percent.
- [15]
Anthropic's Claude Mythos 5 developed a quantum circuit design that uses 17 times fewer gates than an earlier approach.
- [16]
One design uses 12 percent more qubits than the smallest previous design but requires 17 times fewer gates.
- [17]
One optimization reduced the costliest step of a leading quantum optimization algorithm by approximately 12 times.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- interestingengineering.comClaude’s new quantum circuit design uses 17 times fewer gates than earlier approach
1 article · October 16, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI for ScienceFollow
- Quantum ComputingFollow
- Quantum circuit optimizationFollow