Fermi Universe Unveils World's First Full-Stack Quantum-Enhanced LLM FermiQLLM 1.0
Fermi Universe, a Chinese startup focused on Quantum for AI (Q4AI), has unveiled FermiQLLM 1.0, which it describes as the world's first full-stack quantum-enhanced large language model. The model is built on the open-source Qwen base model and incorporates quantum enhancement across five stages: data representation, model structure, training, reinforcement learning, and evaluation. It is compatible with mainstream AI computing infrastructure, including existing GPUs.
Rather than running the entire model on a quantum computer, the company embeds quantum physics and many-body computation methods—such as tensor networks, quantum simulated annealing, and gauge degrees of freedom—into the full pipeline of classic computing. This allows FermiQLLM 1.0 to be trained and deployed without relying on fault-tolerant quantum computers.
In internal tests against traditional models of comparable parameter size, FermiQLLM 1.0 reportedly achieved over 15% improvement in inference performance and more than 25% reduction in training costs for continuous reinforcement learning. On mainstream benchmarks including MATH-500, GPQA-Diamond, and BBH, its comprehensive performance improved by 10%–20% over the same-scale open-source base model.
The company was founded in May 2026, following the project's start in March 2026. It is currently at the seed stage, having raised a total of 100 million yuan in what is said to be the largest seed round in China's Q4AI sector, with a post-investment valuation of around 1 billion yuan. Nearly half of its employees have Tsinghua University backgrounds, and the company collaborates with the quantum research team at Tsinghua's Yang Zhenning Advanced Research Institute. Looking ahead, Fermi Universe plans to abstract its quantum enhancement capabilities into a general framework for different models and gradually open-source part of its technology.
