On September 29, 2025, DeepSeek made an official announcement, rolling out and open - sourcing the DeepSeek - V3.2 - Exp model. At the same time, it carried out simultaneous updates on its official App, web portal, and mini - program. In the realm of technical innovation, this new model features a sparse attention architecture. This architectural design is a game - changer as it optimizes the efficiency of long - text training and inference. From a business perspective, it brings substantial benefits to developers. By slashing API prices by more than 50%, it significantly cuts down on the costs that developers have to bear, making it a highly attractive option in the market.