IBM has announced the Nighthawk r3, a superconducting quantum processor capable of running 10,000 quantum operations across 520 programmable qubits.
The company plans to link two of these chips in 2027 to create a single 1,040-qubit system. The current Nighthawk r2 already executes workloads up to 19 times faster than earlier IBM processors.
Specs and current performance
The new design scales to 520 programmable qubits supported by 2,034 physical qubits and 10 l-couplers. By comparison, the live Nighthawk r2 features 120 programmable qubits, 218 couplers and 120 reset elements, totalling 458 physical quantum components.
Performance on the r2 reaches 100,000+ circuits per second. This is roughly 25 times the speed of the Heron processor, which manages approximately 4,000 circuits per second. IBM reports an initialization error about 25 times lower than previous models, while neighbouring qubits remain undisturbed.
Partners have run 34 billion circuits on the r2 so far. Reported speedups include:
- E.ON: 15x improvement in quantum machine learning for energy grid optimisation.
- Oak Ridge National Laboratory: 12x speedup on magnetic materials simulations matching neutron scattering data.
- Algorithmiq: Circuits ran 8x faster than the original Heron experiment. IBM estimates tensor-network methods would require 240x more memory to match results.
- Q-CTRL: A 60-qubit fermionic simulation with 4,497 two-qubit gates dropped 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 over 7,500 gates, a milestone on its 2026 roadmap.
Changes in Nighthawk r3
The r3 retains the square lattice and fast operations of r2. It adds 10 chip-to-chip interconnects known as l-couplers to support the planned 1,040-qubit scale.
These links connect processors. IBM previously connected two Nighthawk r1 chips using four l-couplers and 80 cm cables. State transfer fidelity reached up to 99.3%, averaging 99%. The company also linked two cryogenic modules, reaching below 15 millikelvin in under five days.
Platform integration
The IBM Quantum Platform now runs high-performance computing and quantum jobs in one managed environment. Users can attach classical compute from a laptop, a supercomputer or the platform itself.
IBM and RIKEN reran a sample-based quantum diagonalisation workflow roughly 5x faster with 100x more computations. Cleveland Clinic and RIKEN processed 608 million samples 33x faster with workflow automation. Pulse-level controls are available for error correction research.
Comparison with other processors
| Spec | IBM Nighthawk r2 | IBM Nighthawk r3 (planned) | IBM Heron r3 | Google Willow | Quantinuum Helios |
|---|---|---|---|---|---|
| Qubit type | Superconducting | Superconducting | Superconducting | Superconducting | Trapped ion |
| Programmable qubits | 120 | 520 | 156 | 105 | 98 |
| Physical elements | 458 | 2,034 physical qubits | Not disclosed | Not disclosed | Not disclosed |
| Connectivity | Square lattice, 4 neighbors | Square lattice + 10 l-couplers | 2 to 3 neighbors | Avg 3.47 | All-to-all |
| Circuit throughput | 100,000+ circuits/s | Not disclosed | ~4,000 circuits/s | Not disclosed | Not disclosed |
| Accurate circuit size | 7,500+ gates | 10,000 operations (target) | Not disclosed | Not disclosed | Not disclosed |
| 2-qubit gate fidelity | Heron-class (no figure) | Not disclosed | Not disclosed | Not disclosed | 99.921% |
| Availability | IBM Quantum Platform | 2027 (2-chip system) | IBM Quantum Platform | Not disclosed | Not disclosed |
What it means
Developers using the IBM Quantum Platform see immediate gains in throughput. The r2 chip allows faster execution of complex circuits without waiting for qubit resets. Partners report significant reductions in runtime for specific simulations, such as energy grid optimisation and material science.
The r3 represents a design phase rather than shipping hardware. The 1,040-qubit system is a plan for 2027. Chip-to-chip links currently average 99% fidelity, falling short of the 99.9% target required for higher reliability.
