During March, when domestic large models are competing to iterate, the highly anticipated DeepSeek V4 finally revealed new developments. Although the official hint suggests the release might be in April, its latest job announcement has already given away the key focus of this "performance monster."
Recently, the DeepSeek website has launched multiple core positions, including Deep Learning Algorithm Researcher for Agent, Data Evaluation Experts, and Infrastructure Engineers. These positions not only emphasize traditional algorithmic skills but also explicitly require applicants to be proficient in AI coding tools such as Claude Code, Cursor, and Copilot.
This move directly confirms the industry rumor: AI coding will become the core breakthrough of DeepSeek V4. It is said that version V4 will significantly enhance logical reasoning, mathematical ability, and programming implementation. Some reports even suggest that its practical performance may rival the current industry benchmark Claude. However, to truly challenge Claude Opus4.6, which has a strong computing foundation and multiple iterations, DeepSeek V4 clearly needs to put in more effort in model refinement.
The current industry competition has entered a fierce stage, and the time for DeepSeek to refine its model is not abundant:
MiniMax: Its iteration speed is astonishing, evolving from M2.5 to M2.7 within just one month.
Dark Matter of the Moon: Its Kimi K3 is ready to launch, with the official even hinting at a tenfold performance improvement.
