Results & proof

What we measured, at what size, against which method.

Every number below has a source, a status and a condition. Hardware means measured on a quantum processor; computation means classical; projection means forward from a measured law.

Measured results with status
Result	What it means	Status
Award winner · GIC 2026 · 2nd worldwide	Judged among the best in the world on industrial materials chemistry Source Challenge judging · GIC 2026 Advanced Materials track, judged result. GIC challenge page (pqic.org) ↗	Judged result
40 qubits	Twice the published chemistry scale, every step within chemical accuracy Source Published paper + our computation · The ~20-qubit baseline is the published work on this method; the 40-qubit figure is our GPU computation with hardware sampling.	Classical computation
173 h → minutes	Training cost of the challenge method, on one GPU Source Published paper + our computation · 173 GPU-hours is the published training cost of the challenge method; minutes on one GPU is our own computation.	Classical computation
4 of 4	Formulations ranked in the right order, every gap within 0.3%, protocol frozen before the run Source Device result · IBM Heron, protocol frozen before the run. Detail in the EUV photoresist case.	Measured on quantum hardware
0.0029%	Processor clock read from our own readout; loss traced to one qubit Source Device result · Read from the processor's own output with our device-quality readout.	Measured on quantum hardware
Tighter	Energies than the standard estimator, same chip, same run Source Device result · Same chip, same run, compared with the standard estimator.	Measured on quantum hardware
126 qubits	A single run, constant-depth circuits (three entangling layers at every size) Source Device result · A 126-qubit run on a superconducting processor, three entangling layers.	Measured on quantum hardware
95.6% vs 39.5%	Safe episodes with a hardware-measured disturbance model vs white noise Source Device result + simulation · Disturbance model measured on quantum hardware; the safety evaluation runs in simulation.	Measured on quantum hardware
2^20 plans	Grid-expansion plan, globally optimal Source Our classical computation · Exhaustive classical search over all 2^20 build plans.	Classical computation
72% · 1.4%	Pairwise model error found; protein site model within 1.4% of exact Source Device result + simulation · Measured on quantum hardware and in simulation, validated against exact references.	Measured on quantum hardware
~93 logical qubits	Flow-solver crossover, projected from a measured cost law Source Projection from a measured law · A forward projection from a cost law we measured. Not a measurement.	Projection

Featured evidence

The materials result, in two charts

Four resist formulations ranked before synthesis by predicted electronic gap (lower activates more easily under EUV light). The ranking held on quantum hardware; every difference within 0.3%; protocol frozen before the run.

Accuracy that improves with scale: 74.0% of the target at 12 qubits, 90.4% at 40 qubits, with circuits kept to three entangling layers. GPU computation plus hardware sampling.

Platform network results

Delivered through Merlin’s platform network

Clients anonymised to their sector; quoted with their published conditions; not presented as our own measurements.

Chemicals / manufacturing