On December 31, 2025, Yuanxiang made a groundbreaking move by officially open-sourcing XVERSE-Ent. This marks the debut of the industry's first foundational large-scale model tailored exclusively for the general entertainment sector, available in both Chinese and English iterations. XVERSE-Ent has undergone extensive fine-tuning to seamlessly integrate into key entertainment scenarios, including social interactions, game storytelling, and cultural content creation. Its design facilitates lightweight deployment and in-depth integration within specialized verticals, effectively bridging the void left by the absence of dedicated open-source models in the general entertainment landscape.

Leveraging cutting-edge MoE (Mixture of Experts) warm-start technology and a meticulously crafted multi-stage training approach, XVERSE-Ent achieves a remarkable balance. It not only significantly boosts performance in general entertainment tasks but also preserves over 98% of its foundational capabilities. The model's architecture supports high-concurrency, single-GPU deployment, dramatically lowering the barriers to entry for developers and enterprises alike. Presently, XVERSE-Ent has been successfully integrated into the AI-driven social platform Saylo, where it has demonstrated outstanding performance across multiple international markets.