DeepSeek has officially released its highly anticipated next-generation artificial intelligence models, DeepSeek-V4-Pro and DeepSeek-V4-Flash, marking a pivotal moment in China’s push for AI self-sufficiency. The launch, which took place on Friday, April 25, 2026, introduces a massive 1.6 trillion parameter architecture with a 1 million token context window, an eightfold expansion over the 128,000 token limit of the previous flagship. This release positions DeepSeek as a formidable competitor to leading models from Anthropic, OpenAI, and Google, while also deepening its integration with domestic hardware providers such as Huawei and Cambricon.
The new V4-Pro model claims to outperform all existing open-source models in agentic coding and complex reasoning tasks. According to the company’s technical report, covered by the South China Morning Post, V4-Pro performs favorably against Anthropic’s Claude Opus 4.6, OpenAI’s GPT-5.4, and Google’s Gemini 3.1 Pro. While the report acknowledges that V4-Pro “falls marginally short of GPT-5.4 and Gemini 3.1 Pro,” it suggests a developmental trajectory that trails state-of-the-art frontier models by only approximately three to six months. This rapid catch-up is particularly notable given the stringent US export controls on advanced AI chips that Chinese companies face.
The V4-Flash variant, designed for speed and cost efficiency, is priced identically to DeepSeek’s V2 model from June 2024, making it one of the most affordable cutting-edge models available worldwide. This pricing strategy reflects DeepSeek’s broader approach of using aggressive cost structures to capture market share from established US providers and build a global developer base.
(Related: Stanford AI Index 2026: China Has Nearly Erased America’s AI Lead)
Disruptive Pricing and Domestic Hardware Integration
DeepSeek continues its aggressive pricing strategy to capture market share. The V4-Pro model is priced at $3.48 per million output tokens, significantly undercutting OpenAI’s $30 and Anthropic’s $25 for comparable models. The smaller V4-Flash model is even more accessible at just $0.28 per million tokens. For context, Chinese competitor Kimi charges $4 per million tokens, meaning DeepSeek is undercutting even domestic rivals. The company expects to lower V4-Pro prices further later in the year as Huawei scales up production of its Ascend 950 AI processors and achieves greater manufacturing efficiencies.
Crucially, Huawei has announced “full support” for DeepSeek’s new models on its Ascend processors and supernode systems. This partnership is a critical development in China’s effort to build a domestic AI ecosystem independent of Nvidia’s CUDA platform. Cambricon Technologies, another major Chinese AI chip developer, also announced compatibility with the V4 models. Nikkei Asia reported that Huatai Securities analysts noted that the V4 technical report explicitly mentions compatibility with domestic chips, highlighting a strategic shift toward hardware self-reliance that aligns with Beijing’s broader technology independence agenda.
The significance of this Huawei partnership extends beyond mere technical compatibility. It represents a clear signal to the global market that China’s AI software and hardware ecosystems are increasingly capable of functioning as a coherent, self-sustaining unit. As US export controls continue to limit Chinese access to Nvidia’s most advanced chips, the viability of the Huawei-DeepSeek partnership will be a critical test of whether China can maintain its AI development trajectory without American silicon.
(Related: Huawei Ascend 910C Demand Surges 20% as DeepSeek V4 Pushes to Break CUDA Dependence)
Global Implications and the Investment Landscape
The launch of DeepSeek V4 has not gone unnoticed by US industry leaders. Nvidia CEO Jensen Huang recently commented on the Dwarkesh Podcast that “the day that DeepSeek comes out on Huawei first, that is a horrible outcome for [the U.S.].” This sentiment reflects growing concerns in Washington about the efficacy of export controls and the rapid advancement of China’s indigenous AI capabilities. The fact that DeepSeek’s new model was released with Huawei support as a headline feature suggests that Huang’s feared scenario is already materializing.
DeepSeek, owned by the Chinese hedge fund High-Flyer, is also reportedly seeking external investment for the first time. Both The Information and the Financial Times report that the company is raising funds from tech giants Tencent and Alibaba at a $20 billion valuation. This influx of capital could further accelerate DeepSeek’s research and development efforts, enabling it to maintain its aggressive pricing and rapid iteration cycles. The V3 model was famously trained on a budget of $5.6 million worth of processors — a figure that AI researcher Andrej Karpathy described as a “joke of a budget” compared to what US labs spend.
The broader context of the V4 launch is equally significant. OpenAI released GPT-5.5 this week, and Anthropic is reportedly testing a new model codenamed Mythos. The simultaneous release of major models from both US and Chinese labs underscores the accelerating pace of the AI race. As the gap between Chinese and American frontier models continues to narrow, the strategic and commercial stakes for both governments and investors are rising sharply.
The V4 launch also comes amid intensifying US diplomatic pressure. The State Department issued a global cable on April 24 warning allies about alleged Chinese AI distillation campaigns, specifically naming DeepSeek as a key actor. DeepSeek has not publicly responded to those accusations, and the company’s technical report makes no mention of them. For investors tracking the AI race, the key question is whether the V4 model’s performance and pricing can sustain DeepSeek’s momentum in the global developer market, or whether the escalating geopolitical pressure will ultimately constrain its international ambitions. The answer will likely shape the trajectory of the US-China AI competition for years to come.