IBM announced the release of Qiskit SDK v2.5 on July 14, 2026, introducing improvements aimed at increasing the speed of quantum circuit compilation and making compilation-pipeline construction more flexible, while also expanding the capabilities of the C-language application programming interface. The release is part of the Qiskit v2.x series, which focuses on developing a faster and more scalable software foundation while preserving compatibility with existing workflows as much as possible.
The release adds a new framework for multirepresentation compilers and ready-made pipelines for compiling circuits into representations associated with fault-tolerant quantum computing. The transformation process also benefits from improvements to the LightSabre algorithm, multithreading, and an optimization stage focused on execution accuracy on real hardware.
Expanding the C Interface for Inspecting Dynamic Circuits
Qiskit v2.5 enables inspection of classical control flow in dynamic circuits directly from the C interface, instead of returning to Python when analyzing these instructions. This includes if/else, while, for, switch, box, break, and continue operations.
Developers can query control-flow types, inspect nested circuit blocks, analyze conditions and loop parameters, and access the links between nested blocks and the original circuit. IBM also added a companion interface for inspecting classical expressions, including expression trees, variables, literals, logical and arithmetic operations, comparison operations, and type information.
These additions enable compiler tools and low-level software to fully analyze dynamic circuits and classical expressions from the C interface. However, creating control-flow operations or classical expressions from C is still unsupported, as support in this release is limited to inspection.
A Framework for Multirepresentation Compilation Pipelines
The release introduces the MultiStagePassManager component for building compilation pipelines that operate across multiple intermediate representations. Rather than requiring the entire workflow to use a single representation, each stage can use the representation best suited to the task it performs.
This change targets advanced users, such as researchers and developers who build custom compilation pipelines or work on fault-tolerant quantum-computing workflows. The framework can be adopted gradually alongside existing Qiskit pass managers, but users must specify the transitions between stages, as they are not inferred automatically.
Ready-Made Pipelines for Fault-Tolerant Computing
Qiskit v2.5 adds two ready-made pass managers for compilation pipelines associated with fault-tolerant computing:
- generate_preset_pbc_pass_manager() for compiling circuits into a Pauli-based computing representation, in which operations are expressed using Pauli measurements and Pauli-product rotations.
- generate_preset_clifford_t_pass_manager() for compiling circuits into Clifford+T instruction sets, which are common representations in fault-tolerant compilation pipelines.
These pipelines remain customizable: developers can inspect and modify compilation stages and add other optimization techniques. The Clifford+T pipeline also allows the trade-off between approximation accuracy and the number of T gates to be adjusted.
Compiler Performance Improvements and Upgrade Requirements
Compilation in the new release benefits from algorithmic improvements to LightSabre, which now handles incoming gates as layers rather than using a fixed window for look-ahead. IBM says this improves scalability for large circuits while maintaining or improving circuit quality compared with previous implementations.
Qiskit has also expanded its use of multithreading, including in stages such as TwoQubitPeepholeOptimization and Optimize1qGatesDecomposition, in addition to improvements to existing stages such as ConsolidateBlocks and UnitarySynthesis. The TwoQubitPeepholeOptimization stage focuses on expected execution accuracy on real hardware, using backend error rates and device characteristics, even if this sometimes increases circuit depth or gate count.
IBM explains that the upgrade generally does not require code changes, but the C interface remains unstable between minor releases, and extensions built with Qiskit v2.4 may need to be rebuilt. Qiskit v2.5 also requires NumPy version 2.0 or later and SciPy version 1.14 or later. For users of the Qiskit v1.x series, support ended with the v1.4.6 release on June 12, including bug fixes, security updates, and all other patches.
In parallel with the release, IBM announced the renaming of Qiskit Runtime Service to IBM Quantum Compute Service to clarify the distinction between Qiskit, an open-source and hardware-agnostic development package, and IBM’s managed service for running algorithms on IBM quantum computers. The change will be implemented gradually beginning July 14 and over the following weeks, with no changes to existing APIs, workflows, or integrations.