Chinese startup DeepSeek launched its fourth-generation model, DeepSeek V4. The new model arrives in two versions, such as Pro and Flash, and is set to replace V3.2, the current product version. Pricing, a final launch date and detailed benchmark data have not yet been disclosed.
Earlier this week, we reported that the AI startup was seeking fresh funding at a valuation above $20 billion.
V4 builds on DeepSeek’s breakout momentum
DeepSeek became one of the most talked-about names in artificial intelligence after its R1 reasoning model gained global attention for delivering strong performance at a fraction of the cost of US rivals.
V4 is now being seen as an important test of whether DeepSeek can sustain that momentum.
The Pro version is aimed at enterprises, developers, and advanced users seeking stronger reasoning and broader world-knowledge performance. DeepSeek claims it outperforms other open-source models on those benchmarks, trailing only Google’s closed-source Gemini 3.1-Pro.
The Flash version is positioned as a faster, lower-cost option for broader deployment. Both models support an ultra-long context window of one million tokens, allowing significantly more information to be processed in a single session than earlier versions.
Huawei partnership signals China’s chip ambitions
One of the most closely watched elements of the launch is DeepSeek’s growing relationship with Huawei. Huawei announced that its Ascend supernode systems will fully support the V4 series.
The shift toward Huawei’s domestic chip ecosystem reflects China’s broader push to reduce reliance on American semiconductor technology following US export restrictions. Huawei also said its full Ascend high-performance product line now supports V4, signalling deeper cooperation between Chinese model developers and domestic compute providers.
Strategically, it strengthens China’s efforts to build a self-sufficient AI supply chain spanning chips, infrastructure, and software.
Competition, controversy and rising market stakes
DeepSeek’s launch came just one day after the White House accused China of stealing intellectual property from US AI labs on an industrial scale, adding fresh strain ahead of a planned leaders’ summit next month, according to Reuters.
DeepSeek has also been drawn into that wider dispute. Washington has accused the company of breaching export controls by acquiring advanced Nvidia chips, while Anthropic and OpenAI have previously alleged improper model distillation practices.
For now, V4 remains a preview. But through stronger performance claims, lower-cost options, and closer ties to Huawei, DeepSeek has once again made itself impossible to ignore.
