DeepSeek, a Hangzhou-based AI startup, has launched an updated version of its foundation model called DeepSeek-V3.1-Terminus, focusing on improved agentic capabilities and bug fixes.

The update comes two months after the previous version, and aims to enhance coding, search, and language consistency.

Earlier users had reported issues with illegible symbols and unexpected language switching in the chatbot’s responses.

DeepSeek reported slight improvements on several AI benchmarks, including Humanity’s Last Exam and coding tests, but saw a drop in performance on the Chinese-language version of the BrowseComp benchmark.

🔗 Source: South China Morning Post

Recent DeepSeek developments