Infleqtion Says It Hit 30 Logical Qubits on 80 Atoms. Its Own Blog Says They Ran in a Distance-2 Code That Detects Errors Rather Than Correcting Them, and About a Quarter of Shots Came Back Right

by Arash Shirazi · September 25, 2026

Quantum Computing Correspondent, Frontier Tech Wire

Infleqtion (NYSE: INFQ) said on Sept. 24 that it had entangled 30 logical qubits using 80 physical qubits on its Sqale neutral-atom quantum computer, meeting the 2026 milestone on its published roadmap. The company furnished the release and a presentation to the Securities and Exchange Commission the same day on a Form 8-K under Item 7.01, Regulation FD Disclosure, and said in the filing that it would give supplemental information during its presentation at the Quantum World Congress in College Park, Md., that day. According to the release, the achievement 'makes Infleqtion the first neutral-atom quantum computing company to reach 30 logical qubits on a commercial system', and was 'experimentally confirmed by a signal that is approximately 1000x stronger than underlying noise'.

The company's own technical account, a blog post by chief technology officer Pranav Gokhale published the same day, is where the qualifying numbers live. The experiments were completed in August, it says, using 80 atoms grouped into ten blocks of eight. Each block is prepared into the distance-3 [[8,3,3]] code, which encodes three logical qubits in eight physical ones. Then comes the sentence that matters: 'In our downstream experiments, we operate the qubits in the [[8,3,2]] distance-2 encoding, which can detect any error or correct for a lost atom.'

A distance-2 code detects a single error; it cannot tell which qubit the error was on, and so cannot correct it. What it lets an experimenter do is throw away runs in which an error was flagged. That is error detection with post-selection, a standard and legitimate technique, but it is not the fault-tolerant error correction the company's roadmap language points toward. The exception the blog names, a lost atom, is a special case: the experimenter knows which qubit is missing, and the blog says parity constraints let the team 'reconstruct a missing X-basis measurement outcome in a block with one lost atom' in software after the fact.

The 80 physical qubits for 30 logical is the arithmetic of the [[8,3,x]] family: ten blocks, three logical qubits each, an 8:3 ratio the blog states explicitly. That is far below the surface-code overheads usually quoted for fault tolerance, and the company's Sept. 14 release on integrating its qLDPC library with NVIDIA's CUDA-Q Logical made a similar point with a different code, which it said used approximately six physical data qubits per logical qubit and had been constructed and validated in software. Low overhead at distance 2 or 3 is a property of the code family, not evidence that logical error rates fall as the code grows, which is what fault tolerance requires and what none of the Infleqtion documents reviewed here reports.

The benchmark circuit is an IQP circuit, diagonal gates sandwiched by Hadamards, with entangling gates connecting all 30 logical qubits, four logical CCZ gates, 136 physical two-qubit gates and about 1,000 physical operations in total, which the company calls one KiloQuOp. The blog sets out the success metric plainly. Of more than a billion possible 30-bit outcomes, the ideal circuit produces only 262,144, about 0.024% of the space; uniformly random outputs would land in that set 'only once every 4,096 samples on average'. Infleqtion writes: 'In our experimental dataset, we observed hit fraction of approximately 25%, roughly 1,000 times the uniform-random baseline.' That is the 'approximately 1000x' of the release, a ratio against random guessing rather than a fidelity. A quarter of the shots in the dataset landed in the ideal set; three quarters did not.

The blog also discloses how post-processing moved that number. Applying loss correction 'enabled us to quadruple the number of good shots', it says, and 'This technique does come at the cost of greater error rates'. It does not state the hit fraction before loss correction, the fraction of shots discarded by error detection, or the number of shots in the dataset.

The entangling operation the release credits to AI is described in the blog as a 'double-CZ' between triplets of logical qubits, found with what the company names as the GPT 5.6 Sol model, which halves the physical cost of a logical entangling step from eight two-qubit gates to four. The blog presents it as a circuit identity, two transversal CX operations expanded and cancelled down to four physical gates. That is a compilation result, checkable on paper. Whether a language model found it or a person did does not change what it is.

The comparison the release invites, and then fences off with the phrase 'on a commercial system', is with the Bluvstein et al. arXiv preprint of December 2023, which reported 'computationally complex sampling circuits with up to 48 logical qubits entangled with hypercube connectivity with 228 logical two-qubit gates and 48 logical CCZ gates', using the same [[8,3,2]] code blocks. Infleqtion's blog itself cites 'previous work from Harvard' for the transversal CCZ. Thirty logical qubits with four CCZ gates on a company-owned machine in 2026 is a different claim from 48 with 48 on a university apparatus in 2023, and readers should hold both in view when weighing the word 'first'.

The chief executive's quotation in the release is reproduced in full. 'Getting 30 logical qubits to work together is hard, and our team has done it,' said Matt Kinsella, CEO of Infleqtion. 'Co-design between our hardware and software enabled this demonstration with just 80 physical qubits. We're moving quickly toward our target of 100 logical qubits in 2028, and we're already developing applications with customers. The goal is to give them a quantum computer that can take on problems they can't solve today.'

The trajectory the company draws is two logical qubits in 2024, twelve in 2025 and thirty now, and the blog records that the 2024 result carried 'a median two-qubit gate fidelity of 99.48% after post-selection for atom loss'. No gate fidelity, physical or logical, is given for the 2026 experiment in either document. The blog's priorities are candid about where the gap lies: 'real-time control and mid-circuit measurement to go beyond post-selection', and 'much stronger error suppression' on the way to the MegaQuOp regime of roughly a million reliable logical operations. Today's circuit is a thousand.

Infleqtion, which came to the NYSE through a business combination with Churchill Capital Corp X, reported in its Aug. 12 release reported second-quarter revenue of $12.6 million, up 116% year over year, a GAAP operating loss of $30.6 million against $10.1 million a year earlier, and $582 million in cash, cash equivalents, restricted cash and available-for-sale securities with no debt, including a $27.4 million payroll-tax timing benefit the company expected to remit in the third quarter. It raised full-year 2026 revenue guidance to approximately $43 million and said it 'remain[ed] on track for 30 logical qubits this year'. The Sept. 24 release says the company has three customers for logical-qubit circuits on Sqale, naming the Wellcome Leap Q4Bio programme; no contract values are attached.

The 8-K states that the release and presentation are furnished and not deemed filed for purposes of the Exchange Act, the standard treatment for a Regulation FD furnishing. The company's forward-looking statement lists 'the ability to achieve 100 logical qubits in 2028 and 1,000 logical qubits by 2030' among the statements subject to risk.

What has not yet been published is the paper. 'We plan to release a full paper for our 30 logical qubit results in the coming weeks, with details on experimental methods, resource counts, and statistical analysis,' the blog says. Until it appears, the shot counts, the acceptance rate under error detection, the pre-correction hit fraction and any error bars on the 25% figure are unknown outside the company. What comes next on the roadmap is a system under contract for the Illinois Quantum & Microelectronics Park in 2027, described in August as designed to scale to more than 50 logical qubits, and the 100-logical-qubit target for 2028. Whether those systems correct errors rather than detect them is the question this week's result leaves open.