- DeepSeek V4 Pro enters a tougher US-China AI race.
- US and Chinese firms are expanding open-model efforts.
DeepSeek has released the official version of its V4 Pro artificial intelligence model as competition among Chinese and US open-weight model developers expands.
The company said V4-Pro-0813 improves its agent capabilities and is available through its API, app, and web services. DeepSeek is also increasing API prices for V4 Pro and V4 Flash while introducing separate peak and off-peak rates.
The launch follows an April preview of V4 Pro and the later release of the cheaper V4 Flash model. Independent tests found that V4 Flash outperformed the earlier V4 Pro preview in several areas, despite Pro being positioned as the more capable model in DeepSeek’s lineup.
Artificial Analysis found earlier this month that V4 Flash was the cheapest among a group of widely used models it assessed. At the time, its API rates stood at $0.14 per million input tokens and $0.28 per million output tokens.
DeepSeek said those rates will change from August 17, with API prices for V4 Pro and V4 Flash rising by between 50% and 1,100%, depending on the model, token type, and time of use. The company is also introducing separate peak and off-peak rates.
DeepSeek gained international attention in early 2025 after its R1 model went viral. The release also drew attention to the cost of developing and running advanced AI models compared with systems from major US technology companies.
Chinese rivals compete on scale and cost
Competition within China has since increased. Moonshot AI, Zhipu AI, MiniMax, Alibaba, and ByteDance have all released new models as developers compete on model performance, pricing, and access.
Moonshot released Kimi K3 in July, a 2.8-trillion-parameter open-weight model that it described as the world’s largest of its kind at launch. The company later paused new consumer subscriptions after requests approached the limits of its available computing capacity.
Moonshot has access to around 20,000 Nvidia Hopper-generation chips through an agreement with Alibaba, according to Bloomberg. US export controls continue to restrict Chinese access to some of Nvidia’s most advanced processors.
Reuters reported that lower prices and the ability to customise Chinese open-weight models have helped systems from companies including DeepSeek, Moonshot AI, and Z.ai gain users among US developers.
Data from fintech provider Ramp showed that 6.1% of businesses spending on AI used platforms such as OpenRouter that offer open-weight and Chinese-developed models in July 2026, up from 4.5% a year earlier. The data covers only a portion of overall AI spending.
US companies expand open-weight efforts
Meta said this week that it would resume releasing open models, including a version of its most capable system.
Nvidia is also expanding its open-model portfolio. The company has released Nemotron 3.5 Lightning and is developing the larger Nemotron 4 family, with the biggest planned model expected to contain at least one trillion parameters, according to The Information.
Nvidia has also introduced NeMo Switchyard, software that routes AI tasks between different models.
US infrastructure providers are also adding computing capacity for open-weight workloads. IBM and Together AI have agreed to build a $240 million Nvidia-powered inference cluster on IBM Cloud.
Together AI hosts models from developers including DeepSeek, MiniMax, and Moonshot. Together AI chief revenue officer Kai Mak said the company expects the new capacity to be fully committed two to three months before it becomes available.
DeepSeek, meanwhile, has turned to external capital as it expands its operations. The company raised about $7.4 billion in its first outside financing round in June and later planned another round at a valuation of about $74 billion. Bloomberg reported in late July that DeepSeek had paused the new fundraising process, although it could resume it later.
The June round marked a change for DeepSeek, which had previously avoided outside capital.
DeepSeek has said it plans to at least double staffing across several departments, including teams working on data centres and AI agents. The company has also been recruiting chip-design engineers as it develops its own AI processor.
Reuters reported in July that DeepSeek’s chip project is focused on inference and is intended to reduce the company’s dependence on external suppliers, including Nvidia and Huawei.
DeepSeek has also increased its use of Huawei hardware as US restrictions limit Chinese access to Nvidia’s most advanced processors. Its V4 architecture was adapted to run on Huawei Ascend chips, while V4 Pro and V4 Flash support a one-million-token context window.
DeepSeek has also linked future pricing to the availability of domestic computing hardware. In April, the company said V4 Pro pricing could fall once Huawei’s Ascend 950 systems became available at scale, while noting that supply constraints would remain until production increased.
Want to learn more about AI and big data from industry leaders? Check out AI & Big Data Expo taking place in Amsterdam, California, and London. The comprehensive event is part of TechEx and is co-located with other leading technology events, click here for more information.
Tech Wire Asia is powered by TechForge Media. Explore other upcoming enterprise technology events and webinars here.
The post DeepSeek V4 Pro launches as US-China open AI model race intensifies appeared first on TechWire Asia.