DeepSeek and Z.ai Target Claude Code as Chinese Open Models Gain Ground in AI Coding

DeepSeek is assembling a team to build an AI coding agent aimed at services such as Anthropic’s Claude Code, marking a direct move into software tools that plan, execute, and revise code. The company created an official social media account and published hiring notices for the effort, according to Bloomberg.

The project has been described as “DeepSeek Code,” with a Beijing-based “Harness” team responsible for the systems around the model, including tool use, planning, memory, and agent loops. The Decoder reported that DeepSeek is recruiting product and engineering talent with experience in those areas. The hiring activity gives the initiative a more concrete shape than a model announcement, while Bloomberg’s coverage indicates that the company is now publicizing the effort as a strategic challenge to Claude Code.

Claude Code has helped make autonomous software engineering a central battleground for AI companies. DeepSeek’s planned product would compete with Anthropic’s Claude Code, OpenAI’s Codex, and developer tools such as Cursor, placing the contest beyond chatbot quality and into real-world coding workflows.

An effective code agent needs more than a capable language model. It must inspect repositories, call tools, maintain context, plan across multiple steps, test changes, and recover from mistakes. The Decoder’s description of DeepSeek’s Harness team centers on those supporting systems, while Bloomberg’s account frames the effort as part of a wider push toward agents that perform work on a user’s behalf. That focus also connects with Z.ai’s GLM-5.2, which has drawn attention for coding and long-horizon agent tasks.

Z.ai has become a second Chinese challenger in coding AI after its GLM-5.2 model gained traction among developers. Bloomberg described an upcoming upgrade to the company’s flagship model as part of a series of Chinese open-weight offerings competing with Anthropic and OpenAI. The Times of India reported that GLM-5.2 reached developer rankings and delivered advanced coding capabilities at significantly lower cost.

GLM-5.2’s performance has added a commercial edge to the competition. The Next Web reported that the model ranked fourth on Artificial Analysis’s Intelligence Index v4.1, with a score of 51, while costing a fraction of comparable Claude and GPT-5.5 usage. Those figures come from benchmark and pricing comparisons rather than a single product trial, but they explain why developers and industry observers have compared the release with DeepSeek’s earlier market impact.

Lower prices and open-weight distribution are changing how Chinese AI companies challenge established US labs. Z.ai’s model has attracted attention by pairing relatively strong coding and agent performance with economics that appeal to developers operating under budget and access constraints. DeepSeek’s earlier releases established the broader pattern of Chinese models competing through efficiency, openness, and aggressive pricing.

DeepSeek V4 Preview, covered by Mashable in April, was priced about 85% below GPT-5.5. The comparison does not measure the planned DeepSeek Code product directly, and DeepSeek’s agent work remains distinct from Z.ai’s released model. Together, however, the developments point to a widening contest over model cost, developer adoption, and the software layer that turns models into useful agents. Export restrictions on advanced chips add a geopolitical dimension to that competition, according to The Next Web.

Developers are gaining more choices among coding agents, but model quality will be only one part of the decision. Repository privacy, tool permissions, reliability over long tasks, integration with existing editors, latency, and the cost of repeated agent calls will determine whether new entrants win sustained use.

Anthropic and OpenAI remain the main commercial reference points in the coverage, while Cursor represents an established workflow product built around coding assistance. DeepSeek’s hiring campaign shows that the company is investing in the agent layer directly; Z.ai’s GLM-5.2 shows how an open-weight model can pressure prices before it matches every leading system. For enterprise buyers, the result is a faster comparison cycle and greater scrutiny of data handling, hosting, and support.

DeepSeek’s announced team and Z.ai’s model momentum show that Chinese AI companies are targeting the work performed by leading US systems, not just their benchmark scores. The immediate question is whether DeepSeek Code advances from recruiting and public positioning to a reliable product that developers use for complex software tasks.

Anthropic and OpenAI face pressure on both capability and pricing as open-weight competitors improve. DeepSeek must prove that its agent harness can coordinate tools and maintain accuracy across long coding sessions, while Z.ai must convert developer enthusiasm and benchmark results into durable adoption. The next meaningful signals will be a public release, independent testing, pricing, deployment options, and evidence from software teams using the tools in production.