Overview
- Multiple outlets report the Hangzhou startup is developing a next-generation model focused on agent capabilities, targeting release by the end of 2025.
- The system is described as executing multi-step tasks with minimal user input and improving based on its prior actions.
- Founder Liang Wenfeng is reportedly pushing for a fourth-quarter unveiling, though the timeline has not been confirmed.
- DeepSeek has not responded to requests for comment, leaving the project’s technical details and readiness unverified.
- The effort aligns with a broader shift toward autonomous agents by rivals including OpenAI, Anthropic, Microsoft, and Google, and follows the R1 model’s earlier market disruption.