Multi-Vendor
AWS's fully managed quantum computing service. Access IonQ, IQM, Rigetti, QuEra, and AQT hardware through a single SDK, alongside free local simulation and managed cloud simulators.
About Amazon Braket
Amazon Braket, launched in general availability in 2020, is AWS's fully managed quantum computing service. Rather than building its own quantum processor, AWS took a multi-vendor approach: Braket provides a single managed environment to access hardware from IonQ (trapped ion), Rigetti and IQM (superconducting), QuEra (neutral atom analog), and AQT (trapped ion, Europe), plus a suite of managed cloud simulators and a free local simulator included with the SDK.
The Braket Python SDK abstracts provider-specific circuit representations behind a common interface. You define a circuit using Braket gate syntax, choose a device, and submit a task. Under the hood, Braket translates the circuit into each provider's native gate set and handles job management, result retrieval, and logging through CloudWatch. OpenQASM 3 is the standard interchange format for gate-based devices.
For hybrid quantum-classical workloads, Braket Hybrid Jobs runs the classical outer loop in a managed container (EC2-backed), co-located with quantum task submission, reducing the network round-trip latency between classical optimizer steps and quantum circuit evaluations. PennyLane-Braket integrates Braket hardware as a PennyLane device, making it straightforward to run QML variational circuits with automatic differentiation on real hardware.
Hardware Specifications
| Specification | Value |
|---|---|
| Hardware providers | IonQ, IQM, Rigetti, QuEra, AQT (multi-vendor) |
| Qubit technologies | Trapped ion (IonQ, AQT), superconducting (Rigetti, IQM), neutral atom (QuEra) |
| Largest system | QuEra Aquila, 256 qubits (analog, neutral atom) |
| Gate-based systems | IonQ Forte (36q), Rigetti Cepheus-1-108Q (108q), Rigetti Ankaa-3 (84q), IQM Emerald (54q), IQM Garnet (20q), AQT IBEX-Q1 (12q) |
| Simulators | Local (free), SV1 (statevector), DM1 (density matrix), TN1 (tensor network) |
| Primary SDK | Amazon Braket Python SDK |
| Supported SDKs | Braket SDK, PennyLane-Braket, Qiskit-Braket provider |
| Gate sets | Vary by device: Rigetti native gates; IonQ GPi/MS; QuEra analog pulses |
| OpenQASM support | OpenQASM 3 across gate-based devices |
| Hybrid workloads | Braket Hybrid Jobs with managed classical compute |
36 qubits
IonQ's Forte trapped-ion systems (Forte-1 and Forte Enterprise-1 on Braket) offer all-to-all connectivity and high gate fidelity. Trapped-ion qubits have longer coherence times than superconducting qubits but slower gate times, and the native all-to-all connectivity avoids most SWAP overhead.
20 / 54 qubits
IQM's superconducting systems, hosted in the AWS Stockholm region. Garnet (20 qubits) uses a square-lattice topology; Emerald scales the architecture to 54 qubits. Among the lowest per-shot prices of any gate-based hardware on Braket.
84 / 108 qubits
Rigetti's superconducting processors with a square-lattice architecture and tunable couplers. Ankaa-3 has 84 qubits; Cepheus-1-108Q extends the family to 108 qubits. Faster gate times than trapped ion with lower connectivity.
256 qubits (analog)
QuEra's neutral-atom analog quantum computer. Aquila does not execute gate-based circuits; instead it runs Rydberg Hamiltonian analog programs. Useful for studying quantum phase transitions and certain combinatorial graph problems through continuous-time evolution.
Use Cases
Braket's unified SDK lets you submit the same circuit to IonQ trapped-ion, Rigetti superconducting, and simulated backends with minimal code changes, making it ideal for hardware-agnostic benchmarking and cross-platform research.
The Quantum Approximate Optimization Algorithm is well supported across Braket's gate-based devices. Braket Hybrid Jobs handles the classical optimization loop with managed compute, reducing orchestration overhead.
The PennyLane-Braket plugin routes QML circuits to any Braket backend, enabling hybrid quantum-classical training pipelines that run gradient computations on IonQ or Rigetti hardware with PyTorch or JAX on the classical side.
QuEra's Aquila device on Braket supports neutral-atom analog simulation for Rydberg Hamiltonian problems. This is suited for studying phase transitions, combinatorial graph problems, and quantum magnetism beyond gate-based approaches.
Braket's DM1 density matrix simulator supports custom noise models, allowing developers to prototype noise-aware circuits and test error mitigation techniques before spending hardware budget on real devices.
TN1 (tensor network simulator) efficiently handles certain structured circuits with low entanglement across many qubits. It is well suited for simulating quantum circuits that approximate real-world optimization or chemistry problems.
Getting Started
pip install amazon-braket-sdkThe local simulator is included in the SDK and runs on your machine with no AWS credentials required. This is the fastest way to start learning Braket syntax and testing circuits before you need real hardware.
from braket.circuits import Circuit
from braket.devices import LocalSimulator
device = LocalSimulator()
circ = Circuit()
circ.h(0)
circ.cnot(0, 1)
circ.probability()
task = device.run(circ, shots=1000)
print(task.result().measurement_probabilities)This runs entirely locally. You should see probabilities close to 0.5 for "00" and 0.5 for "11", confirming the Bell state.
# Install the AWS CLI, then run:
aws configure
# Enter your AWS Access Key ID, Secret Access Key,
# default region (e.g. us-east-1), and output format.To access Braket cloud simulators and real hardware, you need an AWS account with Braket enabled. Create one at aws.amazon.com/braket and enable the service in your chosen region.
from braket.aws import AwsDevice
from braket.circuits import Circuit
device = AwsDevice("arn:aws:braket:us-east-1::device/qpu/ionq/Forte-1")
circ = Circuit()
circ.h(0)
circ.cnot(0, 1)
task = device.run(circ, shots=100)
print(task.result().measurement_counts)
Hardware tasks are asynchronous. Call task.result() to block until the job completes. Results are also accessible via the Braket console. Charges apply per task and per shot.
pip install pennylane-amazon-braket
# Then in Python:
import pennylane as qml
dev = qml.device("braket.local.qubit", wires=2)
@qml.qnode(dev)
def circuit(params):
qml.RY(params[0], wires=0)
qml.CNOT(wires=[0, 1])
return qml.expval(qml.PauliZ(0))
Replace braket.local.qubit with braket.aws.qubit and pass a device ARN to route the same circuit to real hardware. PennyLane handles automatic differentiation for parameter-shift gradients.
Pricing
Braket's local simulator is completely free. Cloud simulators and hardware have per-task and per-shot fees. Most development work can be done locally before committing hardware budget.
Free, no account needed
The Braket SDK includes a local simulator that runs on your own machine. No AWS account required. Suitable for circuit development, testing, and learning the SDK. Runs statevector simulation up to the limits of your local RAM.
View details →1 free simulator hour/month, first 12 months
The AWS Free Tier includes one hour of on-demand quantum circuit simulation time per month for the first 12 months of a new AWS account. This covers light use of the SV1, DM1, or TN1 managed simulators.
View details →$0.30/task + per-shot fee
Braket QPUs charge a $0.30 per-task fee plus a per-shot fee that varies by device, from $0.000425 per shot on Rigetti Cepheus to $0.08 per shot on IonQ Forte. A task is one circuit submission; shots are the number of measurement repetitions. QuEra Aquila is $0.01 per shot.
View details →Reserved capacity agreement
Braket Direct provides direct agreements with hardware providers for reserved access and priority scheduling. Suited for organizations with regular production workloads needing predictable throughput. Contact AWS for pricing.
View details →Learning Resources