IBM's Heron r2 processor launched with bold claims: dramatically lower crosstalk, improved two-qubit gate fidelity, and a new heavy-hex variant topology. After six months of cloud access through IBM Quantum Platform, enough real-world circuit execution data has accumulated to evaluate whether Heron r2 delivers on those promises.

Methodology

We submitted a battery of 40,000 circuits spanning four workload types: random clifford circuits (for benchmarking), VQE ansatz circuits for H2 and LiH molecules, QAOA circuits on Max-Cut instances with 10–50 nodes, and randomized benchmarking sequences targeting individual two-qubit gate pairs. All circuits were transpiled using Qiskit's Level 3 optimization with hardware-aware routing.

Key Findings

Two-qubit gate fidelity improved meaningfully versus Eagle r3. Median ECR gate fidelity across the 156-qubit device measured at 99.3%, compared to 98.7% on the best Eagle r3 units we benchmarked in Q1 2025. The tunable coupler architecture suppresses residual ZZ coupling during idle periods — this shows up clearly in multi-qubit idle crosstalk benchmarks.

CLOPS (Circuit Layer Operations Per Second) for Heron r2 consistently measured between 1,500 and 2,100 CLOPS on our standard 100-qubit circuits, reflecting improved classical control electronics and reduced shot overhead in Qiskit Runtime's batch execution mode.

Effective circuit depth for chemistry workloads increased by approximately 40% before noise dominates, compared to Eagle r3 baselines. This is the most practically significant improvement for near-term variational algorithms.

Limitations

Not everything improved. Readout fidelity on the outer qubit ring (qubits 128–155) is noticeably lower than the core — average readout assignment error of 2.1% vs 1.1% on the core — suggesting calibration or connectivity edge effects. Queue wait times during peak hours remain significant, often exceeding 90 minutes for 100+ shot circuits without reservation access.

Verdict

For chemistry simulation and QAOA workloads under 80 qubits, Heron r2 is a meaningful step forward. For users targeting the full 156-qubit device, the fidelity gradient across the chip requires careful qubit selection to avoid degraded performance regions.