On May 20th, it was reported that DeepSeek's V4 series large model has been released for almost a month now, with a very strong cost performance ratio. However, there is still a lot of work to be done in AI programming. Now, employees have confirmed that the company will be launching a competitor to Calude Code.
AI programming can be said to be the strongest field of AI productivity currently, and it is also the key for Anthropic, a US company, to catch up from lagging far behind OpenAI. It only took more than a year to achieve a valuation of over $1 trillion with Claude Code, a programming agent.
Unfortunately, A company is too harsh on domestic developers, and many people hope to have a domestic programming agent that can surpass or at least benchmark Claude Code. DeepSeek's recent job postings also show that they want to focus on Harness programming.
Now DeepSeek senior researcher Chen Deli has confirmed on social media that the company is forming a new Harness team to directly benchmark Anthropic's Claude Code and focus on developing DeepSeek Code Harness.
The previous official recruitment page also showed that DeepSeek has opened two key positions, Harness Product Manager and R&D Engineer, in Haidian District, Beijing, clearly stating that the newly hired members will participate in the entire development process of DeepSeek's desktop Agent product.
In addition, the recruitment information also demonstrates DeepSeek's understanding of Harness. They define the core path as Model+Harness=Agent, and classify all engineering actions outside of the model, such as context management, tool calling, file reading and writing, terminal execution, and test feedback, into the realm of Harness.
After the release of V4, third-party community developers online developed the open-source DeepSeek TUI, which is also an Agent product similar to Claude Code and has a good reputation. However, third-party products can never replace the official position.
DeepSeek has finally personally assembled the Harness team and officially benchmarked the industry benchmark Claude Code, but the specific product release date is still uncertain.
Now with AI coding, developing such a product is not difficult, but the key is how DeepSeek can use its large models and tool capabilities to create a programming ecosystem that is recognized and willing to be used by a large number of developers.
In this regard, OpenAI and Anthropic still have a deep moat. Previously, domestic tech giants have launched their own coding agents, but their influence is not very good, and they cannot reach the level of these two companies, let alone surpass them.
According to previous reports, DeepSeek will release an upgraded version of V4 in June, which should be DeepSeek V4.1. One of the key areas of improvement will obviously be AI programming, and there should be hope for the release of DeepSeek's Claude Code version at that time.
