Industry benchmarks
Nine industries. For each: the application, the published reference (sourced and dated), the Merlin figure with its status, and the baseline it was checked against. Published figures belong to their authors and are quoted for comparison; they are not our measurements. Partners are not named.
Published reference
Published device runs on the same photoresist chemistry: 14 and 22 qubits.
Sources: arXiv:2607.23988 (2026)
Merlin
4/4 candidates ranked correctly on hardware · 40-qubit chemistry model · training 173 GPU-hours → minutes
Measured on quantum hardwareClassical computationBaseline
Exact reference energies; the published method's training cost
Published reference
Unit commitment at 25,755 variables and 48,939 constraints ran feasibly on a photonic QPU, with a commercial MILP solver ahead at short runtimes; generation scheduling at 26 generators × 24 periods on 36 trapped-ion qubits, no optimality proof.
Sources: arXiv:2505.00145 · platform-network results, 2025–26
Merlin
Optimum proven over 1,048,576 plans · reduced-order model 9% worse on the same QPU · network response to 8 time steps on 156 qubits
Classical computationMeasured on quantum hardwareBaseline
Exhaustive classical search; the full model on the same machine
Published reference
No published production-scheduling result on quantum hardware at industrial scale; visible results are challenge prototypes.
Sources: arXiv:2411.10303
Merlin
6 h → 1 min across 100+ grades, in production (classical optimisation, platform network)
Platform networkBaseline
The plant's prior manual and spreadsheet planning
Published reference
Nonlinear flow on superconducting hardware reaches 15 time steps; vehicle routing on device reaches 3 nodes.
Sources: IBM Heron R3, 2026 · arXiv:2504.01567 · arXiv:2509.11469
Merlin
95.6% vs 39.5% safe episodes · 20% less waiting at signals (proof of concept)
Measured on quantum hardwareClassical computationPlatform networkBaseline
White-noise disturbance model; conventional signal timing
Published reference
Published telecom figures in the field state a percentage with no baseline and no date.
Sources: public vendor and operator announcements, undated
Merlin
+10% average downlink on a live network, no added hardware (quantum-inspired, platform network)
Platform networkBaseline
The network operator's existing carrier-aggregation settings
Published reference
Peptides of 14–16 residues folded on 46–61 trapped-ion qubits reach the classical reference energy on some instances.
Sources: arXiv:2604.26861
Merlin
AUC 0.89 on an independent biological benchmark · site energetics within 1.4% of reference
Measured on quantum hardwareClassical computationBaseline
Exact references; an independent biological benchmark
Published reference
Portfolio optimisation benchmark instances are solved to proven optimality by MIP solvers in seconds; the one notable hardware result (corporate-debt fills, 2025) is unreproduced.
Sources: arXiv:2509.17876
Merlin
72% of a 14-entity ring's correlated exposure invisible to pairwise models, measured on 14 qubits
Measured on quantum hardwareClassical computationBaseline
A pairwise correlation model of the same ring
Published reference
Lattice attacks on hardware run at dimension 7 against classical sieving beyond 180; the 2024 polynomial-time claim was withdrawn by its author.
Sources: arXiv:2006.14057 · arXiv:2604.27153
Merlin
A dated risk window per cipher, central 2030–31, range 2028–34
Our assessmentBaseline
Published resource estimates and error-rate requirements
Published reference
Most application methods need about 100 physical qubits before returning anything; national programmes own 5–25-qubit machines.
Sources: published resource requirements across the application layer
Merlin
Fixed-depth circuits on the installed machine · crossover projected at ~93 logical qubits
ProjectionMeasured on quantum hardwareBaseline
The measured cost law; the installed machine's size
How to read a row
Published reference is what has been published, with its arXiv id or lab and year. Merlin is our figure with its status tag. Baseline is what the figure was measured against.
Your industry, next
A resist, a grid plan, a controller, a cipher: you get the answer, the method it was checked against, and its status. The free check runs in your browser.
