Skip to main contentIBM Quantum Documentation Preview
This is a preview build of IBM Quantum® documentation. Refer to quantum.cloud.ibm.com/docs for the official documentation.

View backend details

Package versions

The code on this page was developed using the following requirements. We recommend using these versions or newer.

qiskit-ibm-runtime~=0.43.1

This page explains how to find information about your available backends.


List or filter backends

List backends available to you

You can use either Qiskit or IBM Quantum Platform to view a list of backends available to you, or to search for a particular backend.

QPU names

QPUs hosted on IBM Cloud® have names that begin with ibm_*. All QPUs are given a city name - for example, ibm_kingston. This name does not indicate where the actual QPU is hosted. They are named after IBM® locations around the world.

Use the QiskitRuntimeService.backends() method, as shown in the next code block. This method returns a list of IBMBackend instances.

To run the following code, be sure you have already authenticated to the service. See Set up your IBM Cloud account for more details.

# Initialize your account
from qiskit_ibm_runtime import QiskitRuntimeService
 
service = QiskitRuntimeService()
 
service.backends()

Output:

[<IBMBackend('ibm_fez')>,
 <IBMBackend('ibm_torino')>,
 <IBMBackend('ibm_marrakesh')>]

To search for a specific backend, use the QiskitRuntimeService.backend() method (note that this is singular: backend), which takes the name of the backend as the input parameter and returns an IBMBackend instance representing that particular backend:

service.backend("ibm_fez")

Output:

<IBMBackend('ibm_fez')>
Note

If you are logged in to a specific instance or region, or if you initialized the service with a specific instance or region by using QiskitRuntimeService(), only the backends available to you on that instance or region are returned.

Filter backends

You can filter the available backends by their configuration or status. For more general filters, set the filters argument to a function that accepts a backend object and returns True if it meets your criteria. Refer to the API documentation for more details.

The following code returns only backends that fit these criteria and are available to you on your currently selected instance:

  • Are real quantum devices (simulator=False)
  • Are currently operational (operational=True)
  • Have at least a hundred qubits (min_num_qubits=100)
# Optionally pass in an instance, region, or both, to
# further filter the backends.
service = QiskitRuntimeService()
 
service.backends(simulator=False, operational=True, min_num_qubits=100)

Output:

[<IBMBackend('ibm_fez')>,
 <IBMBackend('ibm_torino')>,
 <IBMBackend('ibm_marrakesh')>]

A similar method is QiskitRuntimeService.least_busy(), which takes the same filters as backends() but returns the backend that matches the filters and has the least number of jobs pending in the queue:

service.least_busy(operational=True, min_num_qubits=100)

Output:

<IBMBackend('ibm_fez')>

Static backend information

Some information about a backend does not change regularly, such as its name, version, the number of qubits it has, its processor type (the bird family name, which indicates the topology and approximate qubit count), and the types of features it supports. This information is available as attributes of the backend object. For a full list of attributes, see the IBMBackend API documentation.

A backend's region (the location of the data center where your data and experiments will be hosted and processed) is listed in its detailed information card on the Compute resources page on IBM Quantum Platform.

backend = service.backend("ibm_fez")
 
print(
    f"Name: {backend.name}\n"
    f"Version: {backend.backend_version}\n"
    f"No. of qubits: {backend.num_qubits}\n"
    f"Processor type: {backend.processor_type}\n"
)

Output:

Name: ibm_fez
Version: 2
No. of qubits: 156
Processor type: {'family': 'Heron', 'revision': '2'}

QPU versioning

Each QPU has a version number in the form X.Y.Z (major.minor.revision). A circuit compiled for a given version number is guaranteed to run on that QPU. If the revision number changes, the circuit will continue to run. If the major or minor number changes, the circuit is not guaranteed to run, although it may do so.

The revision version number will increment for fixes that do not break the existing compiled circuit.

The conditions under which a version or revision number can change are listed in the following table.

Major versionMinor versionRevision version
Sample changesWarmup / cool-down cyclesQPU software updates
Major changes to the control electronicsSwapping out some electronics, if the replacement appreciably affects operationManual calibrations to improve fidelities
Moving the QPU to a new location, if significant behavior changes resultDropping a gate for some duration of time due to calibration issues, and corrections cannot readily be done in softwareSmall electronics changes that don’t affect operation
Changing the direction of a CNOT gate

Supported instructions

Each processor family natively supports a limited set of instructions. This set includes single- and two-qubit gates, as well as non-unitary operations such as measure and reset. Every gate in the circuit must be translated (by the transpiler) to the elements of a QPU's supported instruction set before it can run on the QPU.

You can view the supported instruction sets for a QPU with Qiskit. The IBM Quantum Platform Compute resources page lists only the supported unitary gates (basis gates) for a QPU.

from qiskit_ibm_runtime import QiskitRuntimeService
 
service = QiskitRuntimeService()
backend = service.backend("ibm_kingston")
 
print(f"Backend: {backend.name}")
print(f"    Processor type: {backend.processor_type}")
print(f"    Supported instructions: {backend.supported_instructions}")

Output:

Backend: ibm_kingston
    Processor type: {'family': 'Heron', 'revision': '2'}
    Supported instructions: ['cz', 'id', 'delay', 'measure', 'measure_2', 'reset', 'rz', 'sx', 'x', 'if_else']

Table of supported instructions

Operation categoryName
Single-qubit gatesRZ, SX, X, ID, delay
Two-qubit gatesCZ, ECR
Fractional gatesRX (single-qubit), RZZ (two-qubit)
Non-unitary instructionsmeasure, reset
Control flowif_else (classical feedforward)

Dynamic backend information

Backends also have properties that change whenever the backed is calibrated, such as qubit frequency and operation error rates. Backend properties update after the calibration sequence completes. These properties can be used when optimizing quantum circuits or to construct noise models for a classical simulator. See the Calibration jobs guide for more information.

On IBM Quantum Platform, you can download calibration data as a CSV file. Click a QPU on the Compute resources page to view its detailed information card, then click the download icon in the upper right corner of the Calibration data section.

Qubit properties

Jump to a qubit property: T1 (Relaxation time) | T2 (Dephasing time) | frequency | anharmonicity | readout_error

backend.properties().qubit_property() returns information about the qubits' physical attributes. It contains a dictionary of various properties of the qubit, each paired with its value and the timestamp of the last calibration.

The following code examples demonstrate how to list all properties, or retrieve specific properties, of a particular qubit.

# fundamental physical properties of qubit 1
 
backend = service.backend("ibm_fez")
 
backend.qubit_properties(1)

Output:

QubitProperties(t1=0.00020833187791710663, t2=0.00023395229316289175, frequency=None)
# calibration data with detailed properties of qubit 0
backend.properties().qubit_property(0)

Output:

{'T1': (5.345445238259663e-05,
  datetime.datetime(2025, 11, 25, 6, 52, 23, tzinfo=tzlocal())),
 'T2': (4.569656984603511e-05,
  datetime.datetime(2025, 11, 25, 6, 53, 9, tzinfo=tzlocal())),
 'readout_error': (0.01416015625,
  datetime.datetime(2025, 11, 25, 11, 42, 11, tzinfo=tzlocal())),
 'prob_meas0_prep1': (0.02734375,
  datetime.datetime(2025, 11, 25, 11, 42, 11, tzinfo=tzlocal())),
 'prob_meas1_prep0': (0.0009765625,
  datetime.datetime(2025, 11, 25, 11, 42, 11, tzinfo=tzlocal())),
 'readout_length': (1.56e-06,
  datetime.datetime(2025, 11, 25, 11, 42, 11, tzinfo=tzlocal()))}
# Retrieve qubit properties
qubit_index = 126  # Replace with your qubit index
qubit_props = backend.properties().qubit_property(qubit_index)
 
# Access specific properties
t1 = qubit_props.get("T1", (None,))[0]
t2 = qubit_props.get("T2", (None,))[0]
frequency = qubit_props.get("frequency", (None,))[0]
anharmonicity = qubit_props.get("anharmonicity", (None,))[0]
readout_error = qubit_props.get("readout_error", (None,))[0]
prob_meas0_prep1 = qubit_props.get("prob_meas0_prep1", (None,))[0]
prob_meas1_prep0 = qubit_props.get("prob_meas1_prep0", (None,))[0]
readout_length = qubit_props.get("readout_length", (None,))[0]
 
print(f"Qubit {qubit_index} Properties:")
print(f"  T1: {t1} seconds")
print(f"  T2: {t2} seconds")
print(f"  Frequency: {frequency} Hz")
print(f"  Anharmonicity: {anharmonicity} Hz")
print(f"  Readout Error: {readout_error}")
print(f"  P(0 | 1): {prob_meas0_prep1}")
print(f"  P(1 | 0): {prob_meas1_prep0}")
print(f"  Readout Length: {readout_length} seconds")

Output:

Qubit 126 Properties:
  T1: 0.00011004519582612597 seconds
  T2: 6.519695327562977e-05 seconds
  Frequency: None Hz
  Anharmonicity: None Hz
  Readout Error: 0.003662109375
  P(0 | 1): 0.0048828125
  P(1 | 0): 0.00244140625
  Readout Length: 1.56e-06 seconds
Click to view qubit properties

T1 (Relaxation time)

The T1T_1 time represents the average duration a qubit remains in its excited state 1|1\rangle before decaying to its ground state 0|0\rangle due to energy relaxation. This parameter is used to characterize the qubit's energy relaxation behavior, and is expressed in units of seconds (s).

View with Qiskitbackend.properties().t1(<qubit>)
View on IBM Quantum PlatformCalibration data section, Qubit dropdown menu

T2 (Dephasing time)

The T2T_2 time denotes the timescale over which a qubit maintains phase coherence of a superposition between the 0|0\rangle and 1|1\rangle states. It accounts for both energy relaxation and pure dephasing processes, providing insight into the qubit's coherence properties.

View with Qiskitbackend.properties().t2(<qubit>)
View on IBM Quantum PlatformCalibration data section, Qubit dropdown menu

frequency

This parameter specifies the resonant frequency of the qubit, indicating the energy difference between the 0|0\rangle and 1|1\rangle states, expressed in hertz (Hz).

View with Qiskitbackend.properties().frequency(<qubit>)
View on IBM Quantum Platformn/a

anharmonicity

Anharmonicity is the difference in energy between the first and second excited states of the qubit, also expressed in hertz (Hz).

View with Qiskitbackend.properties().qubit_property(<qubit>, 'anharmonicity')
View on IBM Quantum Platformn/a

readout_error

The readout error quantifies the average probability of incorrectly measuring a qubit's state. It is commonly calculated as the mean of prob_meas0_prep1 and prob_meas1_prep0, providing a single metric for measurement fidelity.

View with Qiskitbackend.properties().readout_error(<qubit>)
View on IBM Quantum PlatformCalibration data section, Qubit dropdown menu

Instruction properties

Jump to an instruction property: prob_meas0_prep1 | prob_meas1_prep0 | readout_length | 2Q error | Median 2Q error | 2Q error (layered) | 2Q error (best) | RZZ error | Median SX error | ID error / √x (sx) error / Pauli-X error / RX error | Median readout error | Median T1 (relaxation time) | Median T2 (dephasing time) | Single-qubit gate length (ns) | Z-axis rotation (RZ) error | Gate length (ns)

The backend.target attribute is a qiskit.transpiler.Target object: an object that contains all the information needed to transpile a circuit for that backend. This includes instruction errors and durations. For example, the following cell gets the properties for a cz gate acting between qubits 1 and 0.

backend.target["cz"][(1, 0)]

Output:

InstructionProperties(duration=6.8e-08, error=0.007854516178238763)

The following cell shows the properties for a measurement operation (including the readout error) on qubit 0.

backend.target["measure"][(0,)]

Output:

InstructionProperties(duration=1.56e-06, error=0.01416015625)
Click to view instruction properties

prob_meas0_prep1

This parameter indicates the probability of measuring a qubit in the 0 state when it was intended to be prepared in the 1|1\rangle state, denoted as P(01)P(0 | 1). It reflects errors in state preparation and measurement (SPAM), particularly measurement errors in superconducting qubits.

View with Qiskitbackend.properties().qubit_property(<qubit>, 'prob_meas0_prep1')
View on IBM Quantum PlatformCalibration data section, Qubit dropdown menu

prob_meas1_prep0

Similarly, this parameter represents the probability of measuring a qubit in the 1 state when it was intended to be prepared in the 0|0\rangle state, denoted as P(10)P(1 | 0). Like prob_meas0_prep1, it reflects SPAM errors, with measurement errors being the predominant contributor in superconducting qubits.

View with Qiskitbackend.properties().qubit_property(<qubit>, 'prob_meas0_prep0')
View on IBM Quantum PlatformCalibration data section, Qubit dropdown menu

readout_length

The readout_length specifies the duration of the readout operation for a qubit. It measures the time from the initiation of the measurement pulse to the completion of signal digitization, after which the system is ready for the next operation. Understanding this parameter is crucial for optimizing circuit execution, especially when incorporating mid-circuit measurements.

View with Qiskitbackend.properties().readout_length(<qubit>)
View on IBM Quantum PlatformCalibration data section, Qubit dropdown menu

2Q error (Heron: CZ, Eagle: ECR)

The two-qubit error per edge from the same batch of measurements used to calculate the 2Q median and 2Q best errors.

View with Qiskitbackend.target['<instruction>'][<qubit 1>, <qubit 2>]
View on IBM Quantum PlatformCalibration section: hover over the qubit connection in Map view, or find the value in Table view under the CZ error (Heron) or ECR error (Eagle) column

ID error / √x (sx) error / Pauli-X error / RX error

Error in the finite-duration discrete one-qubit gates, measured from randomized benchmarking. The randomized benchmarking sequence includes SX, ID, and X gates, and it is assumed their errors are the same. The ID gate is a delay of duration equal to the duration of the √X and X gates. The RX gate is also the same duration as the √X and X gates with variable amplitude, and so it is reported as having the same error as these gates.

View with Qiskitbackend.target['<instruction>'][<qubit 1>, ]
View on IBM Quantum PlatformCalibration section: Qubit dropdown menu

2Q error (layered)

Average error per layered gate (EPLG) in a chain of 100 qubits. Average EPLG measures the average gate error in a layered chain of NN qubits (NN=100 here). It is derived from a similar quantity known as the layer fidelity (LF) where EPLG100_{100} = 4/5(1-LF199^{\frac{1}{99}}) and layer fidelity is the process fidelity of the layered chain of NN qubits. For details, see the paper Benchmarking quantum processor performance at scale. Note that in the paper EPLG is defined for process error, but for consistency with the individually reported gate errors here it is quoted for average gate error, thus the factor of 4/5.

On IBM Quantum Platform, the detailed information card for each QPU has a section called Two-qubit gate error (layered), which provides the expanded view of the lowest two-qubit gate error (layered) measured as a function of the number of qubits in the chain. The final value, at chain length 100, is the value presented in the Details section. In practice, six 100-qubit chains (pre-selected based on expected optimal performance) are measured, and the value reported for number of qubits N is the lowest error found in a subset length N chain searching over the six 100-qubit chains.

View with QiskitFollow the example in this Qiskit Community GitHub notebook
View on IBM Quantum PlatformDetails section, and an expanded view in the Two-qubit gate error (layered) section

Median 2Q error (Heron: CZ, Eagle: ECR)

Average gate fidelity of the two-qubit operation from randomized benchmarking. Measured in "isolation": batches with a minimum separation of two qubits between edges. This randomized benchmarking uses alternating layers of single-qubit Cliffords and two-qubit gates, and thus the final 2Q error value includes the error of the layer of single-qubit Cliffords. Find an example notebook in the Qiskit Community GitHub. Find per-edge data in the calibration data section of the QPU information card.

View with QiskitFollow the example in this Qiskit Community GitHub notebook
View on IBM Quantum PlatformDetails section; also find per-edge data in the Calibration data section

2Q error (best)

The lowest two-qubit error on any edge of the device from the same batch of measurements used to calculate the median (see Median 2Q error).

View with Qiskit(((FILL IN)))
View on IBM Quantum PlatformDetails section

Median SX error

Average gate fidelity of the √X (SX) gate from randomized benchmarking, measured simultaneously on all qubits. The randomized benchmarking sequence includes SX, ID, and X gates, and it is assumed their errors are the same.

View with Qiskit(((Do we show an example of how to loop over all sx errors and then use numpy to take the median? Is there a better way?)))
View on IBM Quantum PlatformDetails section

RZZ error (Heron)

Error in the RZZ gate averaged over the RZZ angles using a variant of randomized benchmarking for arbitrary unitaries.

View with QiskitImportant: Be sure you have set use_fractional_gates=True when you load the backend, then you can use backend.target['rzz'][<qubit 1>, <qubit 2>]
View on IBM Quantum PlatformCalibration section: Select RZZ in the Connection dropdown menu and hover over the qubit connection in Map view. You can also select RZZ error in the Graph output dropdown menu in Graph view, or find the value in Table view under the RZZ error column

Median readout error

Fidelity of the readout operation. Readout error is measured by preparing the qubit in the 0 (1) state and measuring the probability of an output in the 1 (0) state. The reported value is the average of these two errors. The median is taken over all qubits.

Median T1 (relaxation time)

The T1T_1 time represents the average duration a qubit remains in its excited state 1|1\rangle before decaying to its ground state 0|0\rangle due to energy relaxation. This parameter is used to characterize the qubit's energy relaxation behavior, and is expressed in units of seconds (s).

Median T2 (dephasing time)

The T2T_2 time denotes the timescale over which a qubit maintains phase coherence of a superposition between the 0|0\rangle and 1|1\rangle states. It accounts for both energy relaxation and pure dephasing processes, providing insight into the qubit's coherence properties. T2T_2 is reported from a Hahn echo sequence.

Single-qubit gate length (ns)

Duration of a single-qubit gate operation.

View with Qiskitbackend.target['<instruction>'][<qubit 1>, ].duration
View on IBM Quantum PlatformCalibration section: Qubit dropdown menu

Z-axis rotation (RZ) error

Error in the virtual RZ gate. Reported as all 0 since these are performed in software.

Gate length (ns)

Duration of the two-qubit gate operation.

View with Qiskitbackend.target['<instruction>'][<qubit 1>, <qubit 2> ].duration
View on IBM Quantum PlatformCalibration section: Connection dropdown menu

Additional properties

Jump to a property: CLOPS (or CLOPS_h) | Status | Topology diagram or coupling map

Click to view additional properties

CLOPS (or CLOPS_h)

Circuit layer operations per second, is a measure of how many layers of a 100x100 circuit (hardware-aware circuit) a QPU (quantum processing unit) can execute per unit of time. Find the CLOPS code in the Qiskit Community GitHub.

Status

With BackendStatus, you can find the QPU status (for example, Active, Paused, Offline) as well as the number of pending jobs.

View with Qiskitprint(backend.status().status_msg), print(backend.status().pending_jobs)
View on IBM Quantum PlatformDetails section

Topology diagram or coupling map

A diagram that indicates the pairs of qubits that support two-qubit gate operations between them. This is also called the coupling map or connectivity. Qubits are represented as circles and the supported two-qubit gate operations are displayed as lines connecting the qubits.

View with Qiskitfrom qiskit.visualization import plot_gate_map then plot_gate_map(backend)
View on IBM Quantum PlatformCalibration data section; Click on Expand for a larger view
What does `error = 1` mean?

If the benchmarking of a qubit or edge does not succeed over the course of several days, whether due to poor data quality or other internal factors, the reported error value is considered stale and will be reported as 1. This is not an indication that the qubit or edge is necessarily non-working or that the error is 1; rather, the error is considered undefined and you should proceed with caution when operating that qubit or gate.


Next steps

Recommendations