A 12,635-Atom Quantum Calculation Didn’t Work Alone
The protein study now has a Gordon Bell finalist nod. Its most useful lesson is how quantum and classical machines divide the work.
A quantum computer helped model a system containing 12,635 atoms. The important word is helped.
Researchers at Cleveland Clinic, RIKEN and IBM combined quantum processors with classical supercomputers to study protein-ligand complexes in water. The largest system’s atom count includes the protein, a binding molecule and surrounding water. This was an electronic-structure calculation, not a quantum computer folding an entire protein.[1]
The project began attracting attention in May. In September 2026, IBM reported that the work had become an ACM Gordon Bell Prize finalist. An updated research preprint also describes a more automated workflow and expanded results.[2][3][1]
Think of the calculation as a division of labor. Classical methods break a large molecular problem into manageable fragments. Quantum processors sample electronic configurations for selected fragments. Classical supercomputers then use that information to solve smaller problems and assemble the larger result.[1]
The September version reports 21,006 quantum circuits, more than 239 hours of quantum execution and 3.0 billion measurement outcomes, using up to 94 . Those totals span the study, not a single instantaneous simulation. Fugaku, Miyabi-G and ROQUO supplied classical computing resources.[1]
The caveat is central: this is not a demonstrated win over the best classical methods. IBM explicitly acknowledged that limitation in its May account. The larger calculation and finalist status should not be repackaged as proof of quantum advantage.[2][3]
What makes the work interesting is the complete pipeline. A method that depends on researchers manually moving data between machines is harder to repeat and scale. Automating those handoffs is less dramatic than announcing another qubit record, but it addresses a real scientific-computing problem.[3]
For researchers studying molecular interactions, this is a more concrete possibility than replacing supercomputers outright: give each machine the parts it handles best. The next test is whether the combined workflow can deliver a useful accuracy, cost or time advantage against a strong classical baseline.