Google Quantum AI has launched a global open-competition on Kaggle offering a $150,000 prize pool to incentivize competitive programmers, AI researchers, and circuit architects to compress critical subroutines in fault-tolerant quantum algorithms.
Detailed in a research preprint published on arXiv (“Leveraging verifiable structure in quantum circuits to find more efficient compilations with agentic AI“) and accompanying posts by Google software engineers Tanuj Khattar, Ryan Babbush, and Craig Gidney, the competition centers on kickmix, a new open-source C++20 and Python verification library integrated within Google’s Qualtran software framework. The library enables classical CPUs to fuzz-test candidate circuits with tens of millions of gates at rates up to 60 billion instructions per second on a single core.
| [ Google Quantum AI Verifiable Circuit Compilation Competition Benchmarks ] | |||
|---|---|---|---|
| Target Domain / Task | Scientific Application | Core Arithmetic Bottleneck | Optimization Target |
| • Fusion Energy (Benchmark I) | • Plasma stopping power in inertial confinement fusion | • Fixed-point 1/√x phasing for pairwise Coulomb interactions | • Minimizing Toffoli gates & QROM accuracy trade-offs |
| • Optimization (Benchmark II) | • Decoded Quantum Interferometry (DQI) for OPI problems | • Reversible Reed–Solomon syndrome decoder (EEA) | • Minimizing error locator/evaluator circuit depth |
| • Quantum Chemistry (Benchmark III) | • Electronic structure of nitrogenase (FeMoCo cluster) | • Tensor factorizations, Givens trees, & QROM lookups | • Co-optimizing classical tensor rank and QROM footprint |
Overcoming the Classical Verification Bottleneck with Kickmix Circuits
While classical simulation of arbitrary quantum circuits becomes intractable above roughly 40 qubits due to exponential Hilbert space expansion, the subroutines that dominate the cost of fault-tolerant algorithms often do not rely on arbitrary internal quantum interference. Instead, they are constructed from a restricted gate family termed kickmix circuits—comprising reversible classical logic, diagonal phase gates, and measurement-based uncomputation (HMR gates).
Because kickmix instructions map computational basis states directly to computational basis states while tracking a scalar phase, classical computers can execute randomized fuzz testing in linear time. Combining classical SIMD bit-packing with the Fiat–Shamir heuristic (which seeds test inputs with a salted SHA-256 digest of the candidate circuit), the verifier provides an automated, non-interactive reward signal for AI coding agents and heuristic optimizers to discover low-level bit-twiddling hacks without compromising algorithm accuracy.
The competition opened on October 22, 2026, and will run through the final submission deadline on January 20, 2027.
Participate via the Kaggle Competition Portal here, review the technical release on the Google Open Source Blog here, examine the research paper on arXiv here, read Craig Gidney’s commentary on Algassert here, and view Tanuj Khattar’s announcement on LinkedIn here.
October 22, 2026