Google Quantum AI reports generative quantum advantage on a 68-qubit superconducting processor
In arXiv:2509.09033 (submitted 2025-09-10), Google Quantum AI researchers (including Hartmut Neven, Ryan Babbush, and Jarrod McClean) introduced families of trainable generative quantum models claimed to be classically hard to simulate and free of barren plateaus, and demonstrated learning of classically intractable distributions and quantum circuits on a 68-qubit superconducting processor.
A lab-affiliated demonstration of generative 'quantum advantage' on real hardware qualifies under the academic filter and is a credible benchmark, but as an unrefereed preprint making an advantage claim it lands at the sub-domain-relevant level pending independent verification.
If the trainability and hardness claims survive scrutiny, the result advances the case for near-term quantum machine-learning applications and sustains Google Quantum AI's experimental leadership narrative.