Quantum X Labs Says Its AI Decoder Beat Google's Published Matching Benchmarks on One Configuration — the Aug. 21 Release Publishes No Error Rates

by Arash Shirazi · August 24, 2026

Quantum Computing Correspondent, Frontier Tech Wire

Quantum X Labs Inc. (Nasdaq: QXL) said in an Aug. 21 release that its AI-driven quantum error-correction decoder had shown improved performance against matching-family benchmarks, including what the company described as Google's published correlated-matching and PyMatching results for the same configuration. The evaluation used Google's public surface-code dataset, which was generated from a real quantum-hardware experiment. Every element of that claim is the company's own; the release is a press release, not a paper. The issuer itself checks out — Quantum X Labs is a Delaware corporation listed on Nasdaq under QXL, and filed with the SEC until recently under its former name, Viewbix Inc.

The detail the company put weight on is how its model was trained. According to the release, the decoder was trained exclusively on synthetic samples and was not trained on real hardware shots from the Google dataset. If that holds up under scrutiny, it is a meaningful engineering result: decoders that generalise from simulated noise to the messier error behaviour of physical devices would be considerably cheaper to develop than ones requiring large volumes of hardware-derived syndrome data.

Prof. Nir Sharon, the company's chief quantum technology scientist, was quoted in the release saying: "These results are important because they bring us closer to the point where AI-driven quantum error correction can be evaluated against real hardware behavior, not only simulation."

What the release does not contain is any number. It reports no logical error rate, no physical error rate, no code distance, no number of syndrome extraction rounds and no percentage improvement over the benchmarks it says were beaten. The comparison is described as covering one benchmark configuration. The company said its next step is to replicate and extend the results across additional device centres and code configurations — which is a description of work still to be done, not work completed.

There is also no indication in the release of peer review or third-party validation. That distinction matters in quantum error correction, because decoder performance is sensitive to the exact configuration tested: code distance, noise model, number of rounds and cross-validation split can each move results substantially. A claim of advantage on one configuration of one public dataset is not the same as an advantage across the operating regime, and the release does not assert that it is. Google has not, as far as the release indicates, reviewed the comparison.

The announcement follows a pattern. On July 24 the company said it had run its Deep Quantum Error Correction workflow on an NVIDIA GPU in an AWS environment and benchmarked its transformer-based decoder against the classical minimum-weight perfect matching decoder across toric-code noise configurations, reporting that its decoder outperformed MWPM in selected simulated regimes and showed stable logical and bit error rates under varying physical error conditions. That release likewise carried no quantitative results. NVIDIA and AWS are described in it as technologies the company used, not as partners or customers. Sharon was quoted then saying: "These results are meaningful because they move our program from cloud deployment into measured decoder performance and a surface-code data pipeline."

On the hardware side, a May 28 release said the company had launched a neutral-atom quantum computer with more than 50 physical qubits in Tel Aviv, built around laser cooling and dynamically reconfigurable optical tweezer arrays, and targeting thousands of qubits by the end of the first half of 2027. That release described architectural features — high-fidelity loading, extended coherence times, Rydberg-mediated two-qubit gates — without publishing a single fidelity, coherence-time or gate-error figure, and cited no third-party benchmarking. Its own language points to a development-stage machine rather than a production one. Roadmap targets of that kind are statements of intent.

The financial context is the part that carries the most weight. In its 10-Q for the six months to June 30, 2026, filed Aug. 13, Quantum X Labs reported revenue from continuing operations of $646,000, down from $899,000 in the year-earlier period, and an operating loss of $1.945 million. The company reported GAAP net income attributable to shareholders of $1.880 million for the half — but the filing attributes that profitability to a $3.831 million gain on the deconsolidation of CliniQuantum, on loss of control and a switch to equity-method accounting, rather than to operations. Basic EPS from continuing operations was $0.15 and diluted EPS $0.10, both GAAP. This publication was not able to retrieve the June-quarter 10-Q directly from SEC EDGAR; the figures in this and the following two paragraphs are drawn from StockTitan's summary of that filing, and readers should treat the filing itself as the authority.

The balance sheet is thin. Cash and equivalents stood at $2.435 million at June 30, up from $1.018 million at Dec. 31, 2025, against negative working capital of $550,000. Management stated in the filing that there is substantial doubt about the company's ability to continue as a going concern, citing declining revenue, operating losses, roughly $1.24 million of bank debt and limited liquidity. Research and development spending for the half was $185,000 — a fraction of what established quantum hardware programmes spend in a quarter.

Dilution is already visible in the share count. The filing lists 18,829,198 shares outstanding at June 30, 2026 and 21,567,826 as of the Aug. 13 filing date, an increase of roughly 15% in about six weeks. A company carrying a going-concern flag and $2.4 million of cash has few funding routes other than issuing stock, and the share count is one of the more concrete things in the filing to track.

StockTitan's write-up of the Aug. 21 announcement reported the shares up 4.03% at $4.65 during the session, on volume around 2.3 times average — a reminder of how far a headline can move a name of this size. Those are market-data figures from an aggregator, not company disclosure. The checkable version of the company's claim would look different: published error rates against stated code distances and noise models, a preprint, replication on hardware-derived syndrome data — in its July 24 release the company described a relationship with IQCC, which it identified as a Quantum Machines company, as the route to generating that hardware-derived data, though it disclosed no contract, no contract value and no indication of whether any arrangement is binding — and acknowledgement from the group whose benchmarks are being compared against. None of that exists yet.