Quantum X Labs Starts Error-Correction Trials on Arbel Processor

A first step onto real hardware

Quantum X Labs, the Nasdaq-listed company that trades as QXL, has begun running experiments for its quantum error-correction (QEC) program on Arbel, a superconducting quantum processor at the Israeli Quantum Computing Center (IQCC). Quantum Machines operates the center.34 The company calls the program "end-to-end." It aims to tie three things together in one workflow on actual hardware: characterizing the processor, executing circuits, and decoding errors.3

The company was unusually direct about what the experiments do not show. Its announcement says this first stage is not a demonstration of real-time quantum error correction, and not a demonstration of fault-tolerant quantum computation. It describes both as goals for later phases.34 Coverage of the news repeated that caveat, adding that the company has given no date for a working closed-loop system and has reported no results from the new experiments.6

That honesty is the most useful part of the announcement. The program's direction makes sense. But this is groundwork, and readers should not mistake it for a breakthrough. It comes in a month when larger rivals reported error-correction results of a very different scale.

What the program is meant to build

Quantum X Labs splits the work into three technical layers.34

• Calibration and device characterization: studying how the processor behaves and identifying the noise specific to that hardware.
• Control and circuit execution: feeding that characterization data back into how circuits are run.
• Noise-aware decoding: reading error information and choosing the right corrections, with real-time operation as the eventual aim.

The experiments so far cover the hardware foundation. The next phase is to plug the company's noise-aware decoder directly into the live workflow.3 The longer-term goal is a closed loop: error signals from the processor are analyzed and turned into corrections fast enough to keep pace with the computation. The company says it plans to build this with partners in control systems and accelerated computing.34

The focus on speed is the right one. One industry overview lists real-time decoding as one of the field's unsolved problems, next to qubit overhead, wiring and cooling, and running logical gates. It notes that this is why GPUs and fast interconnects such as NVIDIA's NVQLink keep coming up in QEC discussions.15 A decoder that is accurate but too slow does not help a running machine much.

How QXL got here

The Arbel experiments are the latest step in a program the company has built in public through 2026. In June, Quantum X Labs signed a cooperation agreement with IQCC. The deal gave it access to Quantum Machines' control systems, including the OPX1000 real-time controller, so it could test its AI-based decoder under realistic conditions.5 At the time, one outlet pointed out that neither side disclosed timelines, technical metrics, financial terms or success criteria. It also noted that no independent test results were available.9

In July, the company said it had run its Deep Quantum Error Correction workflow on NVIDIA GPUs in an AWS environment. It compared its transformer-based QECCT decoder with the standard minimum-weight perfect matching (MWPM) decoder and reported that QECCT did better in some simulated noise settings.2 In August, it moved to Google's public surface-code dataset, which comes from a real hardware experiment. It said its updated decoder beat matching-family benchmarks, including Google's published correlated-matching and PyMatching results, even though the model was trained only on synthetic data.1

In October, Gix Internet, which holds about 13% of QXL, published more detailed figures. It said the decoder made 12 to 13 percent fewer logical errors than PyMatching on Google's hardware data in NVIDIA CUDA-Q, at code distances 3 and 5. In a separate 25-round streaming test, the reduction was about 6 percent at distance 3 and 9 percent at distance 5.10 Market-data coverage noted that QXL shares rose 4.6% the day after a decoder benchmark announcement on October 21.3

The pattern is consistent. Each step moves the decoder closer to real hardware: simulated noise first, then public experimental data, and now a processor the company can run itself. Quantum X Labs' science chief, Prof. Nir Sharon, has presented this staged approach as the way to move from offline benchmarking to practical real-time QEC.2

Where the reporting agrees and where it differs

The outlets agree on the substance. They report the same three layers, the same next step of decoder integration, and the same disclaimer about real-time and fault-tolerant operation.46 The details that differ are minor but worth noticing. Most reports list the company on Nasdaq as QXL, while one report gives an over-the-counter ticker, QUTX.36 Reports also describe the company differently. Some list subsidiaries in digital advertising and enterprise AI, including the ad-software firm Gix Media and the grammar-correction developer Metagramm. Others mention quantum algorithms for transportation, drug discovery and security, plus quantum-based navigation.46

That mix matters. Quantum X Labs is a diversified company with a quantum unit, not a dedicated hardware maker. It does not build the qubits it is testing; it runs its experiments on another organization's processor.3 So the right comparison is not with qubit makers. It is with companies that sell decoders and control-stack software, which aim to work across many hardware platforms. The company describes its own strategy in those cross-platform terms.3

The bar set by others this month

The industry context puts the Arbel news in perspective. On October 19, Quantinuum reported a logical fidelity of 99.9995% on its Helios trapped-ion computer, using a color code. That works out to five errors per million operations. The company said this is ten times better than its own September record and 100 times better than the best published superconducting result, from Google Quantum AI.11 Independent analysis pointed out that logical errors fell by an average factor of 12 each time the code distance increased from 3 to 5 to 7. That is evidence of real scaling, not a one-off result.13 Robin Blume-Kohout of Sandia National Laboratories, who was not part of the study, said a scaled-up version could reach useful fault tolerance with fewer than a million physical qubits.14

IBM has also reported progress. An experiment with the University of Chicago used 70 logical qubits. It ran 2,415 logical two-qubit operations and 468 T gates, with logical error rates ten times below the physical ones, and finished a classically intractable task in about 15 minutes.12 A separate report says IBM kept 12 logical qubits running on 288 physical qubits for nearly a million syndrome cycles. It also said IBM's error processing took less than 480 nanoseconds, with its Starling machine targeted for 2029.18

Against that, Quantum X Labs' 12 to 13 percent decoder improvement on public data is a small, incremental gain.10 Still, the work is connected to what the leaders are doing. Quantinuum's own description of its architecture lists "tailored AI decoders" as a core part.20 That suggests machine-learning decoding is becoming standard across the field rather than a fringe idea. The open question is whether an outside, hardware-agnostic decoder vendor can stay ahead of in-house teams at companies like Quantinuum and IBM.

What this means

The Arbel launch should be read as a credibility exercise, not a technical breakthrough. Until now, Quantum X Labs' public results came from simulations or from data other groups collected.12 Running its own experiments on a superconducting processor, through a facility with Quantum Machines' control systems, lets it test its decoder against noise it can measure directly.35 That is a needed step, and the company deserves credit for not overstating it.

The important milestones are still ahead. Two things would show whether the program is real progress: published, ideally peer-reviewed, logical error rates from the decoder running inside the Arbel workflow, and measured decoding speeds that match real-time needs. The company has given no timeline for either.6 Until those numbers appear, investors and researchers should treat QXL's roadmap as a plan and not yet a result. They should also keep in mind that the rest of the industry is already reporting results in errors per million operations.