MIMIQ reaches qBraid with a lesson in quantum scale
A new platform integration makes an emulator easier to reach. Circuit structure determines what its scale means.
.jpg){kind=link}
QPerfect's MIMIQ quantum-circuit emulator became available through qBraid's managed API in an October 29 announcement. The integration gives developers another route to run simulated circuits within the platform. Its attention-grabbing promise reaches thousands of , but the company's disclosure attaches a condition: that scale applies to circuits whose structure suits the matrix product state method. The new development is access through qBraid, rather than a new universal simulation result.[1][2]
The provider documentation makes the distinction concrete. qBraid offers a MIMIQ state-vector option capped at 32 qubits and a matrix product state, or MPS, option whose practical reach depends on the circuit. These are two numerical approaches to representing quantum states on classical computers. A large qubit count alone does not tell a reader how demanding a simulation was.[3]
Why ten more qubits can matter
A full state-vector calculation stores an amplitude for every computational basis state. For n qubits, that means 2 to the power n complex numbers. Each additional qubit doubles that storage requirement. As an illustrative calculation, moving from 30 to 40 qubits multiplies the number of stored amplitudes by 1,024. This arithmetic explains why adding a modest number of qubits can create a severe memory problem for that particular representation.[4]
Ten more qubits. 1,024 times the storage.
For a full state vector: 16 × 2n bytes.
| Qubits | Amplitudes | Storage |
|---|---|---|
| 30 | 230 | 16 GiB |
| 40 | 240 | 16 TiB |
| 50 | 250 | 16 PiB |
MPS uses a compressed description whose required size depends on correlations across partitions of an ordered sequence of qubits. Its bond dimension controls how much of that structure can be retained. When the representation stays compact, much wider circuits become manageable. When it grows too demanding, a chosen resource limit can force approximation. The circuit's structure and the permitted error therefore belong beside its width in a meaningful performance claim.[4]
Structure changes the comparison
Even the word needs care. qBraid's documentation uses a GHZ state as an example: it can involve many qubits while remaining compact across each cut in the MPS description. Calling a circuit entangled does not settle its classical difficulty. The relevant question is how the structure fits the representation used by the simulator.[3]
When returned, MIMIQ's fidelity value is a lower bound on state fidelity—the squared overlap with the ideal simulated state; the documentation says the field can be absent. It describes numerical approximation, rather than a physical processor's gate quality. The simulator settings and any available error information should travel with the result.[3][4]
The benchmark linked to the announcement is historical evidence. Submitted to arXiv in April 2025, it compares seven emulators on selected circuit families reaching 1,024 qubits. All five authors list QPerfect affiliations, with two also affiliated with Strasbourg and CNRS. That makes it a company-authored benchmark worth examining on its methods, rather than independent certification. Its chosen problems do not establish performance for every circuit of the same width.[5]
What a useful benchmark would show
A useful comparison would begin with a question the developer actually needs answered. Is the goal to reproduce measurement probabilities, estimate an observable, test a circuit transformation or examine the effect of a modeled error? Those goals need different outputs. A result can be helpful for one purpose while leaving another unresolved. The evaluator should state the target before choosing a large qubit count as the headline measure.
Reproducibility would then require the actual circuit, its ordering and depth, the simulator settings, the available computing resources and the achieved error criterion. Reporting only the successful job hides the boundary where the method becomes expensive or inaccurate. A more informative demonstration would include a nearby harder case and explain why it changes the result. A comparison is useful when another team can reproduce its conditions and see where its limits begin.
The strongest follow-up would show a specific workflow improved by this access, with the original problem and the resource tradeoffs visible. That would turn an impressive scale claim into something another team could assess. For now, the integration offers an occasion to use more precise language: circuit width describes one dimension of a simulation, while structure, accuracy and resources determine what the result demonstrates.