18 stories
Show: Summaries / Headlines only
On one cross-platform benchmark, Google's Willow ran about 66,000 times as many operations per second as Quantinuum's Helios-1, and Helios-1 ran circuits about seven times as large. Each vendor headlines the axis it leads on.
Ranked by logical qubits, fault-tolerant applications look like a staircase. Ranked by the gates each one needs, the list within reach of announced machines is shorter. Elliptic-curve attacks come before every molecular ground-state estimate in Google's 2025 tables.
Quantinuum has reported logical-qubit counts of 94, 50, 48, and one on the same machine with 98 physical qubits, from four different experiments. With five questions, a reader can tell what most such announcements demonstrated.
Quantinuum's team cut logical errors about 12-fold with each step up in code distance on the 98-qubit trapped-ion machine, for one stored logical qubit. The preprint has not been peer reviewed.
The four join Microsoft and PsiQuantum in Stage C, where government evaluators test whether a proposed utility-scale machine can be built as designed. Of the four, only Diraq disclosed its initial funding, $51 million for the next year.
Polynomial-time quantum algorithms for factoring and quantum simulation are backed by complexity theory. Grover search gains are at most quadratic. Optimization and machine learning on ordinary data rely mostly on heuristics, and researchers have matched several claimed speedups classically.
The code Infleqtion used is guaranteed to flag any single error but not to correct one, apart from the loss of an atom. A full paper with methods, resource counts, and statistics is still to come.
One block of Infleqtion's code stores 18 logical qubits in 98 physical data qubits. Infleqtion's Victory Omole wrote that those numbers are "not a hardware resource estimate." The company has not run the code on hardware.
Quantinuum's team encoded two logical qubits in 20 physical data qubits on its Helios machine and beat unencoded baselines without discarding runs. The team decoded errors after the runs and ran no non-Clifford gates.
Fujitsu put tin-vacancy centers into photonic chips that run at about 1.55 kelvin and, in testing, can be used through its software platform. It gave no qubit count, gate fidelity, coherence time, or entanglement rate.
