Table of Contents
Several leading Chinese AI models took centre stage on Friday as DeepSeek, MiniMax, and ByteDance announced new products and platform updates. The simultaneous launches highlight the rapid pace of AI development in China, with companies introducing new language and video generation capabilities while continuing to compete on pricing, performance, and open-source innovation.
The announcements also reflect the growing competition between Chinese and American AI companies as businesses race to expand their AI ecosystems.
DeepSeek Expands V4-Flash With Public Beta API
DeepSeek introduced the public beta of the official API for its V4-Flash model, making the platform more accessible to developers. The update adds native support for the Responses API format and compatibility with Codex, while also enhancing the model’s agent capabilities.
According to DeepSeek’s changelog, the upgraded V4-Flash achieves benchmark results that “far surpass” its earlier V4-Pro-Preview model. This performance comparison is based on the company’s own benchmark evaluations. DeepSeek also indicated that the official release of DeepSeek V4-Pro is expected to follow soon.
The V4-Flash model is priced at $0.14 per million input tokens and $0.28 per million output tokens and supports a one-million-token context window, making it suitable for handling large-scale AI workloads.
MiniMax and ByteDance Introduce New AI Models
Alongside DeepSeek’s announcement, other Chinese AI models also received major updates.
MiniMax unveiled its H3 multimodal generation model, which can create videos of up to 15 seconds in 2K resolution with native stereo sound in a single generation process. The company stated that H3 costs less than one-third of comparable mainstream models at 2K resolution and announced plans to release the model’s weights as open-source in the coming days.
ByteDance introduced Seedance 2.5, a video generation model that doubles the maximum video length from 15 seconds to 30 seconds. The company said the model is designed for applications including robotics training and autonomous driving simulation.
Pricing Competition Continues
The latest releases further strengthen the position of Chinese AI models in the global AI market, where pricing has become an important competitive factor.
According to CNBC, Chinese AI companies continue offering significantly lower-cost alternatives to several American AI services while also expanding the international availability of their technologies.
Industry data cited in the report indicates that open-source models developed in China account for approximately 41% of global downloads. In addition, CCTV reported that the five most-used models on OpenRouter by call volume during July all originated from Chinese developers.
Growing Competition Beyond Technology
The rapid expansion of Chinese AI models has also drawn increased international attention.
Some senior U.S. officials have said they intend to investigate Chinese AI companies over allegations that certain models may have used “distillation” techniques involving leading American frontier AI systems. China’s Ministry of Commerce rejected those allegations, stating that they “lack factual basis and legal justification.”
These developments reflect the broader competition surrounding AI innovation, where technological progress, commercial strategy, and regulatory discussions continue to evolve simultaneously.
Open-Source AI Continues to Expand
MiniMax’s decision to release the H3 model with open weights represents another step in the growing adoption of open-source AI development.
Hu Qimu, a professor at the Maritime Silk Road Institute of Huaqiao University, described the recent announcements as the result of years of sustained investment, calling them “a period of concentrated breakthroughs.”
He also suggested that AI competition is increasingly shaped by ecosystem development rather than individual technological advances, highlighting the importance of software platforms, developer tools, and collaborative innovation.
Key Highlights
| Feature | Details |
| DeepSeek Update | Public beta API for V4-Flash |
| New Capabilities | Responses API support and Codex compatibility |
| DeepSeek Pricing | $0.14 input / $0.28 output per million tokens |
| Context Window | One million tokens |
| MiniMax H3 | 15-second 2K video generation with stereo audio |
| ByteDance Seedance 2.5 | 30-second AI video generation |
| Open-Source Trend | MiniMax plans to release H3 model weights |
| Global Adoption | Chinese open-source AI models account for 41% of global downloads |
Why This Matters
The latest launches demonstrate how Chinese AI models are expanding across multiple areas of artificial intelligence, including language models, video generation, and developer tools. Companies are not only improving technical capabilities but are also competing through lower pricing and increased support for open-source technologies.
As AI adoption continues to grow worldwide, these developments highlight how competition is extending beyond model performance to include accessibility, ecosystem growth, and enterprise adoption.
Conclusion
The latest announcements from DeepSeek, MiniMax, and ByteDance highlight the continued advancement of Chinese AI models across language processing, video generation, and developer-focused AI services. While DeepSeek expanded access to its V4-Flash model with enhanced API capabilities, MiniMax and ByteDance introduced new video generation technologies aimed at broader commercial applications. Alongside competitive pricing and increasing support for open-source development, these releases reflect China’s growing role in the global AI industry while illustrating the broader competition shaping the future of artificial intelligence.