Fermi Universe raises $14M seed for quantum-enhanced LLM FermiQLLM 1.0 built on Qwen

量子位 · wechat · 2026-09-27

Fermi Universe, China's first startup focused on Quantum for AI, has raised an aggregate 100M RMB seed round ($14M) at a reported 1B RMB valuation, with nearly half its staff from Tsinghua and research ties to the university's Yang Zhenning Institute.

Its first release, FermiQLLM 1.0, claims to be the first fully quantum-enhanced LLM. Rather than running on quantum hardware, it embeds quantum-physics methods (tensor networks, quantum simulated annealing) into five pipeline stages—data representation, architecture, training, RL and evaluation—on top of a Qwen open-source base, staying compatible with existing GPU infrastructure.

Internal benchmarks claim 15%+ inference gains vs. same-size traditional models, 25%+ lower RL training cost, and 10-20% improvements on MATH-500, GPQA-Diamond and BBH. The team plans to generalize the approach into a framework for other base models and open-source parts within 6-12 months.

Original post →

More from Venture

Venture channel →