EDENCODE · NEWS RELEASE
APRIL 14, 2026 · RESEARCH
Lower logical error rates, a faster classical decoding stage, and evidence that decoder size must scale with code distance.
SAN JOSE, CALIFORNIA — April 14, 2026
On World Quantum Day, EdenCode Inc. published results from an early-access study of NVIDIA Ising, the open-source AI model set for quantum computing released the same day.
The study connects the Ising training framework to a Tanner-graph-based simulator and evaluates six model architectures ranging from 50,000 to 7.1 million parameters on NVIDIA H200 GPUs, under both simple and correlated noise models that include CNOT hook errors.
The company reported three outcomes. The hybrid of a neural pre-decoder with PyMatching reduced the logical error rate by 1.7 to 2.0 times compared with PyMatching alone, while the sparser residual syndromes accelerated the conventional decoding stage by approximately 7 times. The model also generalized from surface codes to repetition codes without architectural modification.
The study further reports a co-scaling relationship: as the code distance increases, decoder size must grow in proportion for the error-correction advantage to hold. Models of 50,000 parameters helped at distance 3 but degraded performance at distance 9, while the largest models retained an advantage across all tested distances.
EdenCode said the finding implies that scaling quantum hardware will require parallel scaling of classical AI infrastructure, and that the joint performance of the quantum and classical layers will define the reliability of a fault-tolerant machine.
EdenCode Inc. is a San Jose, California-based company building the intelligence layer for useful quantum computing. EdenCode develops AI-native decoders and control systems designed to meet the real-time latency and accuracy requirements of fault-tolerant quantum computing across multiple hardware platforms. For more information, contact hwanda@edencode.ai.
SOURCES EdenCode research note · Quantum Computing Report, April 15, 2026
EDENCODE · BRAND ASSET
The EdenCode wordmark is available for approved company, press, research and investor communications.
Use the supplied wordmark asset. Do not recreate, redraw or compress it into a screenshot.
Use the master at full size for print and high-resolution materials. For digital use, keep the rendered wordmark at least 120 px wide.
Keep clear space around every side of the wordmark equal to at least the height of the orange mark.
Do not stretch, rotate, recolor, add effects, place it on a busy image or separate the symbol from the wordmark.
