Three questions to ask about quantum progress in 2026

Those three verbs are a useful first filter. A paper reports an experiment under stated conditions. Documentation describes a programming capability, with target-specific limits. A resource estimate explores an application under hardware and correction assumptions.

This October reading guide uses primary sources available at review time. It is not a list of announcements all published this month. In particular, the Willow error-correction paper was published online in December 2024, appears in Nature’s 2025 volume and has a linked author correction from April 2026.

The Willow paper reports below-threshold surface-code memories: increasing code distance improved logical memory behavior within the tested setting. That is a substantial result about error correction. It does not, by itself, demonstrate arbitrary useful fault-tolerant applications.

When reading the paper, identify the encoded task, logical error measure, decoder and operating conditions. Follow the linked correction as part of reading the current record. Avoid using a processor’s physical-qubit total as a substitute for those details.

IBM’s current control-flow guide shows how measurement outcomes can influence later operations in a circuit. The practical question is whether your particular target supports the needed instructions and restrictions. Language expressiveness and hardware acceptance are separate checks.

This matters when importing or converting programs. If the classical feedback cannot survive the conversion, the resulting static circuit is a different program. Keep the source and the limitation instead of treating the missing control flow as a cosmetic detail.

Microsoft’s resource estimator models the physical resources needed by a fault-tolerant computation. Its inputs include the application, hardware architecture, correction model and error budget. A resource figure belongs with those assumptions.

Use several named scenarios to understand sensitivity. A favorable estimate can reveal an interesting direction, but it is not evidence that a current cloud backend can execute the modeled application.

QCalEval evaluates how vision-language models interpret quantum calibration plots. Its benchmark concerns plot understanding, not an independently demonstrated ability to operate every laboratory instrument. That distinction is useful when reading broader claims about AI and quantum control.

Follow the evaluation task and the reference comparison before borrowing a performance claim. A model that answers questions about a diagnostic plot still needs a validated connection to a device and an experimental procedure before it becomes part of a reliable control loop.

For a beginner, the right next step may be a measurement lesson. For a developer, it may be checking control-flow support. For a researcher, it may be reproducing a limited comparison from the paper. None of these requires pretending to have reproduced the full result.

QFlow Hub can keep primary sources and compatible examples close together. Read the status labels, retain the original context and state which small part your experiment actually tests.