According to DeepSeek (@deepseek_ai), the company has released the V3.1 Base model, which features continued pretraining on 840 billion tokens for improved long context extension. The update also includes an overhauled tokenizer and chat template, aiming to enhance language model performance for extended conversations. Both the V3.1 Base and full V3.1 model weights have been open-sourced, offering developers and AI businesses access to advanced large language model capabilities. This release marks a significant step in open-source AI development, enabling enterprises to deploy long-context chatbots and advanced NLP applications with greater efficiency and scalability (Source: DeepSeek Twitter, August 21, 2025).