IBM has announced IBM Quantum Nighthawk r3, its next superconducting quantum processor. It is designed to run 10,000 quantum operations accurately across 520 programmable qubits. In 2027, IBM plans to link 2 r3 chips into a single 1,040-qubit system. For developers, the announcement is speed. The current Nighthawk r2 already runs workloads up to 19x faster than IBM’s earlier processors.
TL;DR
- Size: Nighthawk r3: 520 programmable qubits, 2,034 physical qubits, 10 l-couplers. Nighthawk r2 (live today): 120 programmable qubits, 218 couplers.
- Runs on: IBM Quantum Platform cloud (r2 as
ibm_phoenix). Superconducting chips in dilution refrigerators, paired with laptop-to-supercomputer classical compute. - Performance: r2 executes 100,000+ circuits per second, about 25x Heron, with Heron-class gate fidelity.
- Best: Up to 19x speedup over earlier IBM processors on partner workloads; E.ON reports 15x.
- Worst: r3 is a design, not shipping hardware. Chip-to-chip links average 99% state transfer fidelity, short of the 99.9% target.
- Bottom line: Best: real throughput gains on a cloud chip you can use now. Worst: the 1,040-qubit system is a 2027 plan, not a result.
What is IBM Quantum Nighthawk?
Nighthawk is IBM’s square-lattice processor family built for quantum advantage workloads. Most qubits connect to 4 nearest neighbors, versus 2 or 3 on earlier designs. That connectivity lets circuits express more entangling structure with fewer swap operations.
The current production chip is Nighthawk r2. It has 120 programmable qubits, 218 couplers and 120 reset elements. That is 458 physical quantum elements in total.
How does Nighthawk r2 get 25x more throughput?
The bottleneck was qubit reset. Heron uses conditional reset: measure, then flip any qubit found in the excited state. That method cannot fix leaked qubits, so the chip idles for hundreds of microseconds between shots.
Nighthawk r2 replaces this with a dissipative reset gadget per qubit. A tunable coupler links each qubit to a cold environment on demand. This drops effective T1 from about 200 microseconds to roughly 25 nanoseconds. Idle time between runs falls to as little as 1 microsecond.
The result is 100,000+ circuits per second. Heron manages roughly 4,000. IBM also reports about 25x lower initialization error, while neighbors stay undisturbed.
What speedups are partners reporting?
According to IBM, clients have run 34 billion circuits on Nighthawk r2 so far. The figures below are self-reported by IBM and its partners.
- E.ON: 15x speedup in quantum machine learning for energy grid optimization, versus Heron.
- Oak Ridge National Laboratory: 12x speedup on a magnetic materials simulation matching neutron scattering data.
- Algorithmiq: circuits ran 8x faster than the original Heron experiment. IBM estimates tensor-network methods need 240x more memory to match.
- Q-CTRL: a 60-qubit fermionic simulation with 4,497 two-qubit gates. Runtime fell from 4 minutes 25 seconds to 23 seconds. Deviation from the Heron r3 baseline stayed under 1%.
IBM also reports accurate observable estimation on circuits with 7,500+ gates, a 2026 roadmap milestone.
