Infleqtion Entangled 30 Logical Qubits, and an AI Found the Shortcut

Infleqtion set itself a target of 30 logical qubits in 2026. On September 24 it said it had hit that number on Sqale, its commercial neutral-atom quantum computer, using just 80 physical atoms. The company calls it a first for a commercial neutral-atom system, and it marks a steep climb from the two logical qubits Infleqtion reported in 2024 and the 12 it reported in 2025.

The more unusual detail is how it got there. A key entangling operation in the experiment was not designed by a physicist. It was found by GPT 5.6 Sol, and it cut the physical gate count for that step in half.

What Infleqtion actually ran

The technical write-up on Infleqtion's blog lays out the setup. The 80 atoms were organized into ten blocks of eight, each block encoding three logical qubits. State preparation used an [[8,3,3]] code, which has distance three. The operations themselves ran in an [[8,3,2]] code, a distance-two code that the company says can "detect any error or correct for a lost atom."

The logical qubits were initialized in superposition and then pushed through an IQP circuit, a family of commuting phase-gate circuits often used as a sampling benchmark. That circuit included four logical CCZ gates, the non-Clifford operations that any useful quantum computer needs, implemented transversally with eight physical qubit rotations. All told, the run used about 1,000 physical operations, which Infleqtion brands a "KiloQuOp," including 136 two-qubit gates to build the entanglement.

The headline performance figure is a hit fraction of about 25 percent, against an ideal random baseline of 0.024 percent. That is the "signal approximately 1000x stronger than underlying noise" in the press release. Infleqtion also used post-processing to recover measurements from runs where atoms went missing, which roughly quadrupled the number of usable shots, though the company acknowledges this came "at the cost of greater error rates."

The gate an AI found

The most novel part of the result is a logical operation Infleqtion calls a double-CZ gate. Entangling logical qubits across blocks with the standard transversal CX approach costs eight physical two-qubit gates. The double-CZ gate, which works across triplets of logical qubits, needs four.

"A core accelerant for our work was the AI-assisted discovery of a new entangling operation between logical qubits that performs a key operation with half as many physical gates," said Pranav Gokhale, Infleqtion's CTO and general manager of computing. On near-term hardware, where every two-qubit gate adds error, halving the count for a core operation is not a rounding improvement. It is the difference between a circuit that returns signal and one that returns noise.

The episode fits a pattern that has been building all year: models are now contributing to the design of the machines and codes that may one day outrun them. Here the contribution was narrow and checkable. The gate either works on the hardware or it does not, and Infleqtion says it did.

Detection is not correction

Infleqtion's framing has drawn pushback. Writing for PostQuantum, Marin Ivezic pointed out that the distance-two code used during computation can flag an error but cannot fix one. Runs where errors are detected are thrown away. That works for a circuit of this size, but acceptance rates fall with every added gate, so post-selection does not scale to the long computations that matter.

Ivezic also noted that Infleqtion's earlier roadmap framed its 2026 goal around error correction, and that real-time decoding and mid-circuit measurement, needed for genuine fault tolerance, remain on the company's list of future priorities. His sharpest line: "A company that has traded publicly since February, and whose chief executive defines logical qubits as error-corrected, should say in its headline which kind it is counting."

The comparison set is also crowded. Per PostQuantum, Harvard researchers ran the same [[8,3,2]] code at 48 logical qubits with IQP circuits in 2023. Quantinuum has reported 94 error-detected and 48 error-corrected logical qubits from the 98 physical qubits on its Helios system, along with a 94-logical-qubit GHZ state at 94.9 percent fidelity. Microsoft and Atom Computing entangled 24 logical qubits with error correction in 2024. Infleqtion's claim is specifically about a commercial neutral-atom system, and the qualifier carries weight.

Why the number still matters

None of that makes the result hollow. The ratio is the point. Getting 30 logical qubits out of 80 physical ones, fewer than three atoms per logical qubit, is the kind of efficiency that neutral-atom companies have promised as their advantage over superconducting systems, which typically spend hundreds or thousands of physical qubits per logical one in surface-code designs.

Infleqtion also has paying users running on the logical layer. The company says three customers are using logical-qubit circuits on Sqale, including the Wellcome Leap Quantum for Bio program, where its team built a quantum neural network approach for biomarker discovery that trains on GPUs and runs inference on the quantum processor. That work began at 12 logical qubits and carried into this result.

CEO Matt Kinsella kept his statement simple: "Getting 30 logical qubits to work together is hard, and our team has done it." The company, which trades on the NYSE as INFQ, now targets 100 logical qubits by 2028 and 1,000 by 2030.

The bar moves next

The next milestone the field will judge Infleqtion on is not a bigger count. It is whether those logical qubits can be corrected in real time, mid-circuit, without discarding runs. Until then, logical qubit headlines across the industry will need a footnote about which kind is being counted.

What this result does establish is that the neutral-atom path can be dense, that the company is shipping on the dates it published, and that AI-discovered gates are already a practical tool for squeezing more out of noisy hardware. For a company whose biggest revenue source is still government sensing contracts, a credible computing roadmap is worth as much as the qubits themselves.