Joerg Hiller
Jan 23, 2026 18:06
IBM releases Qiskit SDK v2.3 with custom C transpiler passes, faster circuit optimization, and early fault-tolerant computing tools for quantum developers.
IBM dropped Qiskit SDK v2.3 on January 22, delivering what amounts to a significant infrastructure upgrade for developers building hybrid quantum-classical workflows. The headline feature: custom transpiler passes now work natively in C, letting high-performance computing teams optimize quantum circuits without leaving their existing development environment.
For the uninitiated, Qiskit powers over 550,000 users who’ve executed more than three trillion quantum circuits on IBM hardware since the SDK launched in 2017. This isn’t a niche tool—it’s become the dominant software stack for quantum algorithm research.
What Actually Changed
The v2.3 release introduces QkDag, a circuit representation exposed to C and backed by the same DAGCircuit object Python developers already use. Combined with an expanded QkTarget model, developers can now inspect, modify, and extend compilation processes step by step without rebuilding entire pipelines.
New C functions handle individual transpiler stages: qk_transpile_stage_init(), qk_transpile_stage_layout(), qk_transpile_stage_routing(), qk_transpile_stage_translation(), and qk_transpile_stage_optimization(). This granular control matters for teams integrating quantum workflows with custom hardware or research tools.
VF2Layout and VF2PostLayout—the passes handling circuit-to-hardware mapping—got Rust-driven performance upgrades. Translation: faster compilation and better fidelity when running on actual quantum processors.
Fault-Tolerant Computing Gets Closer
IBM is clearly positioning for the post-NISQ era. Qiskit v2.3 adds the Ross-Selinger (gridsynth) algorithm for Clifford+T basis compilation, exposed both as a standalone function and within the UnitarySynthesis transpiler pass. The OptimizeCliffordT pass from v2.1 also received quality improvements.
A new PauliProductMeasurement instruction enables joint projective measurement across multiple qubits in a single operation. The LitinskiTransformation pass now extends to measurements, creating end-to-end transpilation pipelines for Pauli-based computations—a representation commonly used in error-corrected quantum circuits.
The CommutativeOptimization pass unifies previous cancellation passes, exploiting gate commutativity for more aggressive circuit simplification. It now handles Pauli-based computation circuits too.
Breaking Changes Worth Noting
ControlFlowOp moved to Rust—the last major piece of Qiskit’s internal data model to make that transition. Short-term, this creates overhead: transpiler performance involving ControlFlowOp and BoxOp instructions may temporarily slow down. IBM expects gains in future releases once the refactoring settles.
Python 3.9 support is gone. You need 3.10 or higher. macOS x86-64 (Intel) dropped from tier 1 to tier 2 support—wheels still available, but testing only happens at release time rather than continuously.
The C API function qk_transpiler_pass_standalone_vf2_layout() is deprecated in favor of qk_transpiler_pass_standalone_vf2_layout_average(). This marks the first formal deprecation in the C API.
What’s Next
Qiskit v2.4 arrives later in 2026, with additional v2.x releases possible before v3.0. The trajectory is clear: IBM wants Qiskit positioned as the bridge between today’s noisy intermediate-scale quantum devices and tomorrow’s fault-tolerant systems. For developers already deep in quantum-HPC integration, v2.3 removes friction. For everyone else watching quantum computing mature, it’s another signal that the infrastructure is getting serious.
Image source: Shutterstock
Credit: Source link
