DeepSeek V4 Models Challenge US AI Giants

Chinese AI startup DeepSeek has unveiled preview versions of its V4-Pro and V4-Flash models, positioning them as direct competitors to OpenAI's GPT-5.4 and Google's Gemini 3.1-Pro while maintaining open-source accessibility—a stark contrast to Silicon Valley's proprietary approach.

The DeepSeek announcement details V4-Pro's ability to "beat all rival open models for maths and coding" and trail only Gemini 3.1-Pro in world knowledge, with performance "marginally short" of top closed models, according to their official documentation.

Users interacting with the V4-Flash model experience response times under 0.8 seconds on mid-tier hardware—a noticeable improvement over previous open models that often required 3-second waits. The interface loads cleanly without the usual lag of resource-heavy AI chatbots, making it feel more like a native application than a cloud service.

Unlike US competitors, DeepSeek's open-source strategy allows developers to download, modify, and deploy the code freely, a critical advantage as Washington's export controls restrict access to advanced Nvidia chips. The company explicitly credits Huawei's "Supernode" technology for enabling V4's operation on domestic Ascend 950 processors, a shift from the R1 model's reliance on Nvidia hardware.

Analysts note the market has already priced in China's AI competitiveness, making V4's release less disruptive than the 2025 R1 launch that sent Silicon Valley stocks reeling. "R1 shocked US markets because no one expected a Chinese model to compete at that level," explained Morningstar analyst Ivan Su. "V4 is simply a follow-through on that same trend."

DeepSeek's earlier R1 model drew skepticism over its $6 million training cost—far below Silicon Valley's billion-dollar budgets—though some analysts questioned whether the startup had access to undisclosed resources. The company's latest release, however, faces less scrutiny as it operates within China's domestic chip ecosystem, with Huawei confirming its Ascend 950 clusters now support V4's computational demands.

Despite claims of "world-class reasoning," DeepSeek acknowledges V4-Pro still trails Gemini 3.1-Pro in knowledge benchmarks, a gap the company attributes to "a developmental trajectory that trails state-of-the-art frontier models by approximately 3 to 6 months." The open-source community has already begun modifying the code, with early GitHub forks optimizing the model for specific coding tasks—a process that feels more like collaborative tinkering than corporate product development.

Industry observers point to the broader implications: China's AI sector now dominates open-source models while US firms lead in closed systems. "Chinese firms are undeniably dominating open systems," noted Counterpoint Research's Wei Sun, highlighting V4's potential to accelerate global AI development by reducing reliance on Nvidia chips.

Yet the path forward remains fraught. The White House has accused Chinese entities of "industrial-scale" model distillation, and US states, Australia, and European nations have maintained bans on DeepSeek-R1 over privacy concerns. The V4 release arrives amid these tensions, with DeepSeek's blog warning users to "rely only on our official accounts" amid "statements from other channels" that "do not reflect our views."

Whether users will adopt V4 for production workloads remains uncertain, especially as Chinese firms like Alibaba and ByteDance intensify competition. The real test lies not in benchmark scores but in whether developers will trust a model trained on domestic chips to power critical applications—a question that will determine if DeepSeek's open-source strategy can sustain its momentum beyond the initial hype (and the $6 million cost savings that made R1 famous).

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn