【Tech 24H】On the evening of September 22, DeepSeek released the latest upgraded version of its online model, named DeepSeek-V3.1-Terminus. This update includes various improvements based on user feedback, such as mitigating language consistency issues like mixed Chinese-English output and occasional abnormal characters. Additionally, the capabilities of the Code Agent and Search Agent have been optimized. The upgraded model supports an output length of 4K to 8K tokens in non-thinking mode and 32K to 64K tokens in thinking mode, aiming to enhance user experience and model performance. These updates are designed to address shortcomings in the previous version.