IBM Research Proves Quantum Circuits Outperform LLMs on Two Problems

IBM Research published a paper on September 15, 2026, showing a theoretical separation between small quantum circuits and large language models for one functional and one sampling task. The 60-page paper, 'Separating quantum circuits from classical LLMs,' was posted on arXiv on August 4, 2026. The authors note the results are purely theoretical and do not imply practical advantage due to noisy quantum computers.

• Functional task iterated index solved by quantum circuit of depth O(log log n) with one classical AND gate
• Any constant-depth transformer requires width n^Ω(1)
• Diffusion language models cannot replicate parity-sampling distribution even with chain-of-thought
• Authors call work theoretical and do not specify exact scale of advantage