UNISOUND has officially launched its next-generation, self-developed general-purpose large language model named U2. The U2 model is constructed using a state-of-the-art Mixture of Experts (MoE) architecture based on fast and slow thinking fusion. It leverages high-precision data purification and high-density semantic representation capabilities to achieve top-tier large model performance with a scale of 266 billion parameters.
The model's core foundational logic centers on "Intelligence Density × Token Value," focusing on enhancing the effective knowledge capacity per token and its business output value, thereby establishing a high-value, high-efficiency foundational model.
U2 is a natively Agent-driven large model, delivering integrated capabilities for full-stack development, intelligent orchestration, and deep reasoning. Utilizing pioneering native reasoning path distillation technology, it can autonomously perform task decomposition, execution, verification, and optimization, forming a closed-loop process, making it an ideal native foundation for next-generation AI Agents.
Simultaneously, the model innovatively incorporates a hybrid thinking mode and employs a model Harness synchronous training mechanism, significantly reducing token consumption for single tasks and markedly improving overall token usage efficiency, resulting in a notable advantage in reasoning costs. This makes it widely adaptable for commercial applications across various domains and scenarios.
In authoritative professional evaluations, the U2 model demonstrates excellent overall performance. In challenging benchmarks such as instruction following, long-context understanding, scientific reasoning, software engineering agents, and specific toolchain adaptation (Claw evaluation), its performance is comparable to the world's leading models.
The official release of the U2 large model signifies a comprehensive upgrade in the company's AI commercialization pathway. On one front, leveraging its vast data accumulation and deep industry expertise in specialized fields like healthcare, insurance, and transportation, the company will use the iteratively upgraded U2 foundational model as its core to rapidly expand business scale and steadily advance the optimization and upgrading of its business model.
On another front, the company will delve deeper into standardized cloud-call services such as large model API calls, exploring entirely new incremental market spaces, continuously broadening the company's development boundaries, and unlocking the commercial value of AI technology.
Comments