Transmon or Trapped Ion? What the Price List Tells You

Everything here is runnable on your own circuit. Try it in the console

Two hardware modalities dominate commercially available quantum computing: superconducting transmon qubits and trapped ions. Comparisons of the two usually proceed through physics. It is more informative to proceed through the price list, because the commercial terms encode the engineering trade-off more honestly than any specification sheet.

Here are two devices available through the same cloud aggregator, at the same per-task fee:

                              Rigetti Cepheus-1-108Q   IonQ Forte Enterprise 1
Modality                      Superconducting          Trapped ion
Qubits                                    108                   36
Shots per task                    10 - 50,000          100 - 5,000
Per task                                $0.30                $0.30
Per shot                            $0.000425                $0.08
1,000-shot experiment                  $0.725               $80.30

Why the superconducting shot is cheap

A transmon is a lithographically fabricated circuit: a Josephson junction shunted by a capacitor, cooled to around 15 millikelvin, addressed with microwave pulses. Because it is fabricated rather than assembled, adding qubits is a layout problem rather than a physics problem, which is why superconducting devices reached three-digit qubit counts first.

Gate times are the decisive economic property. Single-qubit gates take tens of nanoseconds and two-qubit gates a few hundred, so a circuit executes in microseconds and a 10,000-shot experiment completes in well under a second of device time. Shots are cheap because they are fast.

The costs sit elsewhere. Coherence times are on the order of a hundred microseconds, which bounds useful circuit depth to a few thousand gates before the state has decayed.

Connectivity is fixed by the chip layout, typically nearest-neighbour on a lattice, so a two-qubit gate between distant qubits compiles into a chain of SWAPs and the effective depth grows during routing. And every qubit needs its own control lines into a dilution refrigerator, which is where the scaling difficulty ultimately concentrates.

Why the trapped-ion shot is expensive

A trapped-ion qubit is a single atomic ion held in an electromagnetic trap and manipulated with lasers. The qubits are identical because they are atoms, which removes an entire category of calibration work that superconducting devices require.

The physical advantages are substantial. Coherence times run to seconds rather than microseconds, a factor of roughly ten thousand. All-to-all connectivity is available because ions in a shared trap couple through collective motional modes, so any pair can interact directly and no SWAP chains are needed. Two-qubit gate fidelity is the best available on any platform.

The cost is speed. Gates are driven by laser pulses on microsecond to millisecond timescales, three to four orders of magnitude slower than microwave gates on a transmon.

A circuit that takes a superconducting device microseconds takes a trapped-ion device milliseconds, and device time is what the per-shot price is buying. The 188x price ratio is, to a first approximation, the gate-speed ratio.

Scaling is the other constraint. Ions in a single trap interact through shared motional modes, and adding ions makes those modes progressively harder to control, which is why commercial trapped-ion devices sit in the tens of qubits while superconducting devices are in the hundreds.

What the error rates say

Representative depolarizing error rates by device class, the figures our noisy simulator uses to preview hardware behaviour:

                    1-qubit gate   2-qubit gate
Superconducting            0.001           0.01
Trapped ion               0.0005          0.005

Trapped ions are roughly twice as good on both, and the two-qubit figure is the one that matters. It is an order of magnitude worse than the single-qubit figure on both platforms, so two-qubit gate count is the quantity that predicts whether a circuit returns signal.

A rough estimate makes the consequence concrete. Circuit fidelity falls off approximately as (1 - e)^g for g two-qubit gates at error rate e, so a 100-gate circuit retains about 37% fidelity at e = 0.01 and about 61% at e = 0.005. At 300 gates the same estimate gives 5% and 22%. The gap widens with depth, which is the practical meaning of a factor-of-two fidelity advantage.

This is also where the connectivity difference stops being an abstraction. If a circuit's logical two-qubit gates are between arbitrary pairs, a nearest-neighbour device inserts SWAPs, each of which is three CNOTs.

A circuit with 100 logical two-qubit gates can become several hundred physical ones after routing, on the platform that already has the higher error rate per gate. All-to-all connectivity removes that multiplier entirely.

Both modalities on one molecule

Everything above is device specifications. This is the two modalities doing the same job: the H2 molecule at equilibrium on IQM Garnet, a superconducting transmon device, and on AQT IBEX Q1, a trapped-ion device, alongside the classical engines that solve the same two qubits exactly.

Run on our engines

The H2 molecule at its equilibrium bond length, whose exact electronic ground state is -1.857275 Ha. Two qubits, so it fits every device we offer. Submitted to each kind of compute we offer, on 16 September 2026. Every figure below is a real job on the service, priced as any customer would be priced.

Device	Engine	Kind	Qubits	Result	Cost
	exact.cpu	CPU	2	ZZ = -1.0000, the ideal value certificate	$0.0001
	exact.gpu	GPU	2	ZZ = -1.0000, the ideal value certificate	$0.0001
	qpu.iqm.garnet	QPU	2	ZZ = -0.9326, superconducting, 4,096 shots certificate	$6.239
	qpu.rigetti	QPU	2	ZZ = -0.5420, superconducting, 4,096 shots * certificate	$2.041
	qpu.rigetti	QPU	2	ZZ = -0.5107, the same circuit re-run * certificate	$2.041
	qpu.aqt.ibex	QPU	2	ZZ = -0.9200, trapped ion, 100 shots certificate	$2.650
	neural.cpu	CPU	2	-1.116981 Ha total, 0.0203 Ha above exact certificate	$0.0001
	neural.tpu	TPU	2	-1.116981 Ha total, 0.0203 Ha above exact certificate	$0.074

* The two Rigetti rows are one circuit run twice, an internal reproduction of the published benchmark notebook. A depolarizing noise model puts both versions at about -0.99, so the shortfall is not the circuit shape, but the identical program has not yet run on both devices. The steps are in the docs.

A note on the hardware certificates: they state Hellinger fidelity against the exact distribution. For an optimisation circuit that distribution is spread across many outcomes rather than concentrated on one, so the figure is low by construction and is not a measure of whether the device found a good answer. The result column above is.

The same problem is yours to run: every instance here is seeded, so it rebuilds exactly. Open the console and a cost estimate is free before anything executes.

The hardware rows are not a ranking of the modalities and are not offered as one. The two superconducting devices ran at 4,096 shots each and differ by 0.39, which is real and is a gap between two machines of the same modality rather than between modalities. IBEX ran at 100 shots, so the trapped-ion figure carries far more sampling error than either and cannot be ranked against them. What the table does show is the price argument above running in the direction people do not expect. The cheaper-per-shot superconducting device cost more on the day, because it was given roughly forty times the shots. The full write-up is on the chemistry benchmark.

The same two modalities, on a question shot counts cannot spoil

The table above has to decline a comparison, because an expectation value measured at 100 shots and one measured at 4,096 carry different amounts of sampling error and the difference between them is not all device. A second problem avoids that entirely.

A quantum circuit Born machine was trained to put half its probability on 00 and half on 11, which forbids the other two outcomes. Shots landing on 01 or 10 are therefore errors you can count rather than a distance you have to interpret, and a count becomes a rate, and a rate is comparable across different shot budgets in a way an expectation value is not.

Run on our engines

A quantum circuit Born machine (QCBM) learning a two-qubit target distribution from a random start, then sampled. Submitted to each kind of compute we offer, on 18 September 2026. Every figure below is a real job on the service, priced as any customer would be priced.

Device	Engine	Kind	Qubits	Result	Cost
	exact.cpu	CPU	2	TVD 0.459 untrained, 0.0117 after 60 Adam steps	$0.0806
	exact.gpu	GPU	2	TVD 0.041, the same converged circuit at 1,000 shots certificate	$0.0001
	qpu.rigetti	QPU	2	TVD 0.102, the converged circuit re-run at 1,000 shots	$0.7250
	qpu.iqm.garnet	QPU	2	TVD 0.104, the same circuit and shot count on a second superconducting device certificate	$1.7500
	qpu.aqt.ibex	QPU	2	3 of 100 shots outside the target, against 102 of 1,000 on Rigetti and 38 of 1,000 on Garnet certificate	$2.6500
	neural.tpu	TPU	—	a Born machine is a circuit, not a Hamiltonian, so it does not reach the neural tier	—

A note on the hardware certificates: they state Hellinger fidelity against the exact distribution. For an optimisation circuit that distribution is spread across many outcomes rather than concentrated on one, so the figure is low by construction and is not a measure of whether the device found a good answer. The result column above is.

The same problem is yours to run: every instance here is seeded, so it rebuilds exactly. Open the console and a cost estimate is free before anything executes.

Out of every 100 shots, roughly 3 went astray on the trapped-ion device, 4 on IQM Garnet and 10 on Rigetti. The first two sit together and the third does not. The trapped-ion figure comes from 100 shots and carries about half its own width in uncertainty, which is enough to place it alongside Garnet and below Rigetti and not enough to separate it from Garnet: that would take more shots, and on this device a shot is $0.0235 against Rigetti's $0.000425. The modality argument and the price argument meet exactly here, and which one wins depends on whether you are buying shots or buying the quality of each one.

Both machines ran the identical converged circuit, so nothing in the comparison depends on the training. The instance is on quantum generative models.

Choosing between them

The decision is usually settled by three questions, in this order.

• How many qubits does the circuit need? Above 36, the trapped-ion device is not an option and the question is closed. This is the most common deciding factor and the least discussed
• How deep is the circuit in two-qubit gates, after routing? Shallow circuits run acceptably on either. Deep circuits, or circuits with non-local connectivity, favour trapped ions by enough to justify the price difference. Compare transpiled depth against submitted depth before deciding; the growth is invisible otherwise
• How many shots does the result need? At 188x per shot, a statistics-hungry experiment on trapped-ion hardware becomes expensive quickly. A 5,000-shot task, the IonQ maximum, costs $400.30. The same task on the superconducting device costs $2.425

Note also the lower shot bound. IonQ rejects any task under 100 shots, so the single-shot smoke test that works against every simulator and against the Rigetti device fails against Forte Enterprise 1. Discovering this after a queue wait is avoidable by reading the device's shot range first.

The chemistry table above puts the two modalities on a molecule. This is the same pair of modalities on an optimisation problem instead, which is where routing depth and connectivity start to matter rather than gate error alone. It is satellite observation tasking at 14 requests, on three superconducting devices and the classical engines.

Run on our engines

Satellite observation tasking at 14 requests, seed 20260902, whose exact optimum is value 32.5128. On 25 September the same instance ran on exact.tpu, a Google TPU, which returned 31.2164 and the smallest gap on the table. Submitted to each kind of compute we offer, on 16 and 25 September 2026 at 500 shots. Every figure below is a real job on the service, priced as any customer would be priced.

Device	Engine	Kind	Qubits	Result	Cost
	mps.quimb.cpu	CPU	14	25.1546, gap 7.36 certificate	$0.0001
	exact.cpu	CPU	14	20.0569, gap 12.46 certificate	$0.0001
	exact.gpu	GPU	14	25.1546, gap 7.36 certificate	$0.0001
	qpu.iqm.garnet	QPU	14	26.8778, gap 5.64 certificate	$1.025
	qpu.rigetti	QPU	14	27.0176, gap 5.49 * certificate	$0.5125
	qpu.iqm.emerald	QPU	14	26.9421, gap 5.57 certificate	$1.100
	exact.tpu	TPU	14	31.2164, gap 1.30 certificate•best outcome	$0.0776
	neural.tpu	TPU	—	the tasking QUBO is diagonal, which is not the shape a neural ansatz is for	—

* The Rigetti row is a separate sample of ours on this same instance, with the QAOA angles re-optimised for it. The steps are in the docs.

A note on the hardware certificates: they state Hellinger fidelity against the exact distribution. For an optimisation circuit that distribution is spread across many outcomes rather than concentrated on one, so the figure is low by construction and is not a measure of whether the device found a good answer. The result column above is.

The same problem is yours to run: every instance here is seeded, so it rebuilds exactly. Open the console and a cost estimate is free before anything executes.

No trapped-ion row appears here, and the reason is the first criterion above rather than a preference. This instance was run at a width and a shot count where the per-shot price made a trapped-ion submission a different kind of purchase. The method and the classical baseline are on space and satellites, and the whole fleet is described in the machine comparison.

The step most people skip

Both devices cost real money and both have queues measured in minutes to hours. Neither is the right place to discover that an ansatz was misconstructed or that an observable had the wrong Pauli string length.

Simulation answers those questions for a fraction of a cent and returns in seconds. A noisy simulator using the error rates above will additionally show whether the circuit survives realistic gate noise, and if it does not, no amount of hardware time will change that. The remaining question, which platform, is then answered by the three criteria above rather than by trial.

The full arithmetic of hardware versus simulation, including what variational workloads cost once the iteration count is included, is in What does it cost to rent a quantum computer?. The constraints each platform imposes before a job is accepted are in Why your circuit was rejected.

Common questions

How much does a superconducting quantum computer cost to use?

Renting time on one is the cheapest hardware access available. Rigetti Cepheus-1 is $0.000425 a shot and IQM Garnet and Emerald are $0.00145 and $0.0016, all on top of a flat $0.30 task fee. A complete 1,000-shot experiment therefore costs between $0.73 and $1.90 depending on which superconducting machine you choose.

Why is a trapped-ion machine more expensive than a superconducting one?

Because a shot takes longer to produce. A superconducting chip is driven with microwave pulses and reset thousands of times a second, so shots are cheap to make.

A trapped-ion machine physically manipulates atoms with lasers and repeats far more slowly, so each shot consumes more machine time. That physical difference, not a pricing strategy, is why the same 1,000-shot circuit is $0.725 on Rigetti and $80.30 on IonQ.

Pricing every figure below against your own workload: open the per shot cost calculator.