Quantum Annealing
Quantum annealers with thousands of qubits for real-world optimization. Free trial access for developers via Leap, including real QPU time, not just a simulator.
About D-Wave
D-Wave Systems was founded in 1999 in Burnaby, Canada, and delivered the world's first commercial quantum computer in 2011. While competitors spent years in research labs, D-Wave was already running customer workloads. The company invented quantum annealing as a commercial product and has continued to scale the technology, reaching over 5000 qubits in the Advantage system. In May 2025 its next-generation Advantage2 system reached general availability with 4400+ qubits on the denser Zephyr topology.
D-Wave's qubits are superconducting flux qubits. On Advantage they are arranged in a Pegasus graph topology, where each qubit connects to up to 15 neighbors; Advantage2 uses the Zephyr topology with 20-way connectivity. This dense connectivity is critical for solving optimization problems, because it determines how well real-world problem graphs can be embedded into the hardware. The annealing process starts all qubits in a superposition of 0 and 1, then slowly lowers a transverse magnetic field while raising the problem Hamiltonian. Quantum tunneling allows the system to escape local energy minima during the anneal, potentially finding better solutions than classical greedy algorithms.
The key practical advantage of D-Wave over gate-based quantum computers is scale and hybrid capability today. The LeapHybridSampler combines the 5000-qubit QPU with classical heuristic solvers, allowing it to handle optimization problems with millions of variables. No gate-based quantum computer is remotely close to this scale for practical problem-solving. For combinatorial optimization, D-Wave is the most production-ready quantum platform available.
Hardware Specifications
| Specification | Value |
|---|---|
| Current systems | Advantage (5000+ qubits), Advantage2 (4400+ qubits, GA since May 2025) |
| Qubit technology | Superconducting flux qubits (quantum annealing, not gate-based) |
| Problem type | Optimization: QUBO (Quadratic Unconstrained Binary Optimization) and Ising |
| Topology | Advantage: Pegasus graph, 15-way connectivity; Advantage2: Zephyr graph, 20-way connectivity |
| Anneal time | 1-2000 microseconds per shot (adjustable) |
| Operating temperature | 15 millikelvin |
| Hybrid solver scale | Millions of variables (LeapHybridSampler, QPU + classical) |
| Cloud access | D-Wave Leap (free trial for new developers); Leap also purchasable via AWS Marketplace |
| Primary SDK | Ocean SDK (Python, open source) |
5000+ qubits • Production
D-Wave's long-running commercial system. 5000+ flux qubits in a Pegasus topology with 15-way connectivity. Available via the Leap cloud service, including the free trial. Suitable for optimization problems with up to roughly 5000 binary variables directly, or millions via hybrid solvers.
4400+ qubits • Production
D-Wave's next-generation system, generally available since May 2025. Advantage2 uses the Zephyr topology with 20-way connectivity and, per D-Wave, delivers a higher energy scale, substantially lower noise, and roughly doubled coherence compared with Advantage, targeting more complex optimization landscapes.
Use Cases
D-Wave's native problem type is QUBO, which maps directly to combinatorial optimization problems like job scheduling, bin packing, maximum clique, and graph coloring. No gate-based circuit needed.
Vehicle routing problems (VRP), traffic signal optimization, and supply chain scheduling map naturally to QUBO formulations. D-Wave and partners have published results for real-world logistics applications.
Financial portfolio construction under constraints (budget, risk, sector limits) is a natural QUBO problem. Hybrid solvers can handle realistic portfolio sizes with thousands of assets.
Molecular conformation problems and protein-ligand docking can be formulated as energy minimization tasks suitable for quantum annealing. D-Wave has active partnerships with pharmaceutical researchers.
Ising models arise naturally in condensed matter physics. D-Wave hardware can explore Ising ground states for frustrated magnets and disordered systems at scales difficult for classical solvers.
Boolean satisfiability, graph partitioning, and scheduling under constraints all admit QUBO encodings. The Leap hybrid solvers extend this to large industry-scale instances with millions of variables.
Getting Started
Go to cloud.dwavesys.com/leap and create a free account. The trial includes 1 minute of QPU time or 20 minutes of hybrid solver time (or a combination), valid for one month after signup. No payment information required to start.
pip install dwave-ocean-sdkOcean is D-Wave's open-source Python SDK. It provides tools for formulating QUBO and Ising problems, embedding them onto hardware, running samplers, and analyzing results. See the Ocean SDK reference for full API details.
dwave config createRun this command and enter your API token from the Leap dashboard. The Ocean SDK will save your credentials locally. All subsequent QPU calls will use this token automatically.
import dimod
from dwave.system import DWaveSampler, EmbeddingComposite
# A simple 2-variable QUBO: minimize x0*x1 - x0 - x1
Q = {(0, 0): -1, (1, 1): -1, (0, 1): 2}
sampler = EmbeddingComposite(DWaveSampler())
response = sampler.sample_qubo(Q, num_reads=100)
print(response.first.sample) # lowest energy solution
print(response.first.energy)
This submits a simple 2-variable QUBO to the Advantage QPU. The EmbeddingComposite automatically finds an embedding of your problem graph onto the Pegasus hardware topology. The result is the lowest-energy (best optimization) solution found across 100 annealing runs.
from dwave.system import LeapHybridSampler
# Works with problems too large for direct QPU embedding
sampler = LeapHybridSampler()
response = sampler.sample_qubo(large_Q, time_limit=5)
The LeapHybridSampler combines the QPU with classical heuristic solvers, enabling problems with millions of variables. This is D-Wave's most production-ready offering for real-world industry problems. Hybrid solver time is billed separately from QPU time in Leap.
import dimod
# Use a BinaryQuadraticModel for more readable formulation
bqm = dimod.BinaryQuadraticModel('BINARY')
bqm.add_variable('x0', -1.0)
bqm.add_variable('x1', -1.0)
bqm.add_interaction('x0', 'x1', 2.0)
# Convert to QUBO and sample
sampler = EmbeddingComposite(DWaveSampler())
response = sampler.sample(bqm, num_reads=100)
The dimod library provides BinaryQuadraticModel (BQM) as a high-level abstraction. You can add variables and interactions symbolically, then convert to QUBO or Ising automatically. This is much cleaner than writing raw Q matrices for larger problems.
Pricing
D-Wave Leap's free trial makes it one of the most accessible quantum hardware platforms for developers. The trial minute of QPU time is enough to run hundreds of small optimization problems and learn the full workflow without spending anything.
1 min QPU or 20 min hybrid, 1 month
New Leap accounts get a free trial: 1 minute of QPU time or 20 minutes of hybrid solver time (or a combination), valid for one month after signup. No credit card required to start. Ideal for learning and small experiments. The Leap Quantum LaunchPad program offers qualified applicants a longer three-month trial.
View details →Subscription + QPU time
Paid Leap plans provide ongoing QPU time and access to the hybrid solvers for large-scale problems. Pricing varies by plan tier and organization type. D-Wave also runs academic and developer programs.
View details →Leap subscription via AWS billing
D-Wave systems are no longer offered through Amazon Braket; direct Braket access was retired in late 2022. Instead, Leap subscriptions can be purchased through the AWS Marketplace and billed to your AWS account, while access still runs through the Leap cloud service.
View details →Understanding D-Wave
D-Wave's quantum annealer and gate-based systems like IBM Quantum or IonQ are fundamentally different technologies that solve different problem types. They are not interchangeable.
Gate-based quantum computers run quantum circuits, support universal quantum computation, and are the target for algorithms like Shor's factoring and Grover's search. They currently have low qubit counts (tens to a few hundred usable qubits) and require error correction for deep circuits. D-Wave cannot run these algorithms at all.
D-Wave's quantum annealer runs a physical optimization process and is purpose-built for QUBO and Ising problems. It has 5000+ qubits today, runs at 15 millikelvin, and can sample from complex energy landscapes faster than classical annealing for certain problem structures. The hybrid solvers extend this to industry-scale problems with millions of variables. D-Wave cannot run gate-based circuits or universal quantum algorithms.
For practical combinatorial optimization today, D-Wave's hybrid solvers are among the most competitive options in any computing paradigm, classical or quantum. If your goal is optimization, D-Wave is worth evaluating seriously alongside classical solvers like Gurobi.
Learning Resources