DeepSeek has put its next-generation AI model V4 into preview, sharpening attention on China’s place in the global AI race. The Chinese AI company says the open-source model can compete with leading closed-source systems from US rivals including Anthropic, Google, and OpenAI.
The preview arrives a year after DeepSeek drew major attention with R1. That earlier model rattled the US AI industry because DeepSeek claimed it was trained at a fraction of the cost of leading US systems.
What DeepSeek Is Claiming With V4
The central claim around V4 is direct and ambitious: DeepSeek says the model can stand against major American AI systems while remaining open-source. That matters because the comparison is not being made against smaller tools or niche products, but against systems from Anthropic, Google, and OpenAI.
DeepSeek also says V4 represents a major improvement over its previous models. The company is emphasizing coding as one of the areas where the new model has advanced most clearly.
That focus is not incidental. Coding has become a core capability for AI agents, and the source article links this trend to the success of tools such as ChatGPT Codex and Claude Code. In that context, stronger coding performance is not just a technical upgrade; it is part of how AI systems are becoming more useful in practical workflows.
Why Coding Is At The Center
AI coding tools are increasingly important because they connect model ability with direct action. A model that can reason about code, generate code, revise code, and help operate software tasks becomes more than a chatbot. It starts to function as part of a broader agentic system.
DeepSeek’s decision to highlight coding suggests that V4 is being positioned for this part of the market. The company is not only presenting the model as a general AI system. It is pointing to a capability that has helped define high-profile products from US competitors.
The source article does not provide benchmark scores, pricing, deployment details, or a full technical breakdown of V4. What it does make clear is that DeepSeek wants the model judged in the same competitive frame as products associated with OpenAI, Anthropic, and Google.
- Model: V4
- Company: DeepSeek
- Release status: preview
- Positioning: open-source model competing with closed-source US systems
- Highlighted strength: coding
Huawei Compatibility Adds A Second Signal
DeepSeek is also explicitly highlighting compatibility with domestic Huawei technology. The source article describes this as a milestone for China’s chip industry.
That detail gives the V4 preview a broader significance than model performance alone. It ties the release to the hardware side of AI development, where the chips used to train and run models are a central part of the competition.
DeepSeek has not disclosed V4’s training costs. It also has not said what hardware the model was trained on. Those missing details matter because the company’s earlier R1 release drew attention partly because of DeepSeek’s claim that it had achieved strong results at a much lower cost than leading US systems.
Without those disclosures, V4’s preview leaves an important gap. DeepSeek is making a strong performance claim, but the source article does not provide the underlying cost or training-hardware information needed to evaluate how the company reached that point.
The Unanswered Questions Around DeepSeek
The V4 preview also arrives with scrutiny still surrounding DeepSeek. US officials have accused the company of using banned Nvidia chips. Anthropic claims DeepSeek misused Claude to improve its own products.
Those allegations are part of the backdrop to the new model. The source article does not resolve them, and DeepSeek’s V4 announcement does not answer every question about how the model was built. But they shape how the release will be read by rivals, regulators, and AI industry observers.
For now, the main facts are clear. DeepSeek has released a preview of V4. It says the model is open-source and able to compete with closed-source systems from Anthropic, Google, and OpenAI. It says V4 improves on prior models, particularly in coding. It is emphasizing Huawei compatibility. It has not disclosed training costs or the hardware used to train the model.
That combination makes V4 a closely watched release. It is a product preview, a competitive statement, and a signal about China’s AI ambitions, all at once.
What To Watch Next
The next question is whether V4 can match the expectations DeepSeek is setting. The company has framed the model as a serious competitor to leading US systems, but the source article provides no independent performance results or detailed technical evidence.
The open-source positioning is also important. If V4 is broadly usable and meaningfully strong in coding, it could attract attention from developers and teams interested in alternatives to closed-source models.
At the same time, the missing training-cost and hardware details remain central. DeepSeek’s earlier impact with R1 was tied to cost claims. With V4, the company is again making a large competitive claim, but key information about how the model was trained is not yet public.
DeepSeek’s V4 preview is therefore both a launch and a test. The company has placed itself directly in comparison with Anthropic, Google, and OpenAI. The coming scrutiny will focus on whether V4 can support that comparison in real use, especially in coding and AI agent workflows.