Chinese AI start-up DeepSeek has introduced an update to its foundation model, V3, enhancing its agentic capabilities and resolving bugs based on user feedback. The new version, DeepSeek-V3.1-Terminus, was released on Monday, just two months after the previous V3.1 iteration, which was praised by Artificial Analysis as the company’s most sophisticated model to date. DeepSeek regards V3.1 as a pivotal step towards the “agent era,” aiming to support software that automates specific tasks for users. The Hangzhou-based company highlighted improvements in coding, search capabilities, and language consistency. Prior to the update, users reported issues with the chatbot producing responses with illegible symbols and switching languages unexpectedly. According to DeepSeek’s self-reported metrics, V3.1-Terminus exhibits slight improvements across several popular benchmarks.
What advancements does DeepSeek’s updated AI model bring?
