⚙️ Chinese startup DeepSeek updates language model V3

DeepSeek has released an update for its core language model V3, removing the mention of the reasoning network R1 in the chatbot. As reported by SCMP, the context window has now been expanded to 128,000 tokens, allowing for the processing of significantly larger volumes of data, equivalent to a 300-page book.

In the Aider Polyglot benchmark, which assesses the ability of LLMs to solve complex programming tasks in various languages, DeepSeek V3.1 outperforms Claude 4 Opus.