Abstract: As public concern about the quality of large language models (LLMs) continues to grow, vendors are increasingly adopting open-source strategy as an indirect means to signal quality, aiming to enhance consumers' perception of model performance and alleviate security concerns. This paper constructs a signaling game model in which the open-source degree of LLMs serves as a quality signal, and examines how a vendor's quality disclosure strategy affects the open-source degree, market demand, and profits. The study finds that when the quality difference between the vendor's potential high- and low-type states is small, the vendor conceals its quality type under a non-disclosure strategy (pooling equilibrium). Compared with active disclosure, in this case a high-type vendor chooses a lower open-source degree and experiences lower first-period market demand but higher second-period demand, while the opposite pattern holds for a low-type vendor. When the quality difference is large, the vendor reveals its type under non-disclosure (separating equilibrium), where a high-type vendor adopts a higher open-source degree and achieves higher first-period demand but lower second-period demand compared to active disclosure, while a low-type vendor's open-source degree and market demand remain unaffected. Active quality disclosure always enhances the profit of a high-type vendor, but reduces the profit of a low-type vendor when the quality difference is substantial.

Key words: quality information asymmetry? large language models? signaling game? open-source degree