IBM, RIKEN Simulate 12,635-Atom System With Quantum Computing

A team from Cleveland Clinic, RIKEN, and IBM has simulated a biologically significant molecular system containing 12,635 atoms. This marks the largest such simulation ever reported on quantum computers. The achievement, which involved running quantum processors alongside some of the world’s most powerful supercomputers, earned the team a finalist spot for the prestigious 2026 ACM Gordon Bell Prize, a top honor in high-performance computing. The prize winner will be announced in November at SC26 in Chicago.

It’s important to understand the nuance of this accomplishment. Though impressive, this peer-reviewed and preprint study highlights both the modest and ambitious aspects of current quantum capabilities.

Here’s a breakdown of what truly happened:

Quantum Computers Handled a Complex Part, Not the Entirety

Drug discovery faces a persistent challenge: determining how strongly a candidate molecule binds to a protein target. This requires calculating the binding energy, which in turn necessitates understanding how electrons arrange themselves within the system. Since electrons are quantum objects, simulating them accurately on classical computers quickly becomes computationally expensive as the number of atoms increases.

Given that quantum computers operate on the same principles as electrons, the team leveraged IBM Quantum Heron processors for the electronic-structure calculations, which is the inherently quantum-mechanical portion of the problem. These quantum chips utilized up to 94 qubits and performed nearly 6,000 quantum operations on the parts of the problem where high precision was crucial.

The remaining computational tasks were handled by classical supercomputers: Fugaku at RIKEN, Miyabi-G (run by the University of Tokyo and University of Tsukuba), and later ROQUO, RIKEN’s newest GPU system. These classical machines reassembled the individual pieces into a complete picture of each molecule. IBM refers to this integrated approach as quantum-centric supercomputing, illustrating how quantum and classical processors collaborate.

Sample-Based Quantum Diagonalization: Quantum Sketches, Supercomputer Paints

The method, known as sample-based quantum diagonalization, was developed by IBM and RIKEN and featured on the cover of Science Advances. The core idea is intuitive:

The quantum computer isn’t tasked with solving the entire electronic-structure problem. Instead, it samples the system, generating a set of probable, critical electron configurations. The classical supercomputer then takes these samples and performs the computationally intensive diagonalization to extract the energy. To apply this to real protein complexes, Cleveland Clinic integrated embedded wavefunction methods.

Think of the quantum processor as an expert at identifying the crucial pieces of a complex puzzle. It provides the right selection of pieces. The supercomputer, with its vast and patient processing power, then meticulously assembles the complete picture.

This hybrid approach allowed the team to simulate a 12,635-atom system without requiring a million-qubit machine, which currently doesn’t exist. The quantum component remained focused and manageable, while the classical component handled the bulk of the computational load. This division of labor is the key innovation, reflecting an honest assessment of what current hardware can and cannot achieve independently.

The Startling Speed of Scaling

Initially, in May 2026, the team reported the first known simulation of a 303-atom protein using quantum computers. This served as their baseline.

Less than a year later, they successfully scaled that method by approximately 40 times, simultaneously achieving a 210-fold improvement in accuracy. These figures are the authors’ own claims based on their experiments, and though not independently verified, they indicate a rapid advancement. In updated results published in September, they further refined the binding-energy calculations, which predict how molecules interact and stick together.

Crucially, they also validated the entire workflow on ROQUO, significantly reducing the need for manual data transfer. By coordinating CPUs, GPUs, and QPUs within a single automated pipeline, the team minimized human error often associated with manually moving results between different machines. This leads to a faster time-to-solution and fewer opportunities for mistakes. Though seemingly mundane, this “plumbing” work transforms a laboratory demonstration into a practical tool for working scientists.

Understanding the Scope: What It Is and What It Isn’t

No claim of quantum advantage has been made. Reproducing and enhancing classical chemistry with quantum assistance is distinct from demonstrating that a quantum machine unequivocally outperforms all classical methods. This paper is a proof-of-principle at an impressive scale, with classical supercomputers performing a substantial amount of the work. Though the 94-qubit processors were essential for accuracy, according to the team, they were not operating in isolation.

Furthermore, this work does not represent a new drug. Binding-energy predictions are part of fundamental research, far removed from clinical applications. The research points toward improved modeling of how medicines might interact with protein targets someday.

Finally, being a Gordon Bell finalist is just that, a finalist status. The winner will not be announced until SC26 in November 2026.

In essence, this achievement is a functioning quantum-classical pipeline that successfully modeled a 12,635-atom system, improved binding energy calculations, and automated complex data transfer steps. The aspirations, from clinical applications to a fully autonomous quantum machine independent of classical supercomputers like Fugaku, remain future goals.