DeepSeek Launches New V4 Open-Source AI Models
DeepSeek on Friday introduced preview versions of its latest flagship open-source artificial intelligence model, V4, highlighting improved reasoning and overall performance.
The company unveiled two variants—DeepSeek-V4-Pro and DeepSeek-V4-Flash—expanding its presence in the rapidly evolving AI landscape.
Model Variants and Capabilities
The V4-Pro model features approximately 1.6 trillion parameters, positioning it among the largest AI models available. In contrast, the V4-Flash version is a more lightweight alternative, with 284 billion parameters, offering efficiency alongside strong performance.
Both models include a context window of up to 1 million tokens, significantly increasing the amount of information they can process in a single interaction.
Focus on Text Processing and Future Multimodal Expansion
At present, the models are designed to handle text-based tasks only. However, DeepSeek confirmed that it is actively working on integrating multimodal capabilities, which would enable the models to process images and video in future updates.
Performance Benchmarks and Industry Position
DeepSeek stated that its most advanced system, V4-Pro-Max, delivers top-tier results in coding benchmarks and significantly narrows the gap with leading closed-source AI models in reasoning and agent-based tasks.
On the widely followed MMLU-Pro benchmark, V4-Pro demonstrated performance comparable to OpenAI’s GPT-5.4, while slightly trailing models from Google and Anthropic.
Hardware and Training Speculation
DeepSeek did not disclose the specific hardware used to train its new models. Earlier reports had raised concerns among U.S. officials regarding potential use of restricted chips from NVIDIA.
More recent information suggests the models may have been trained using advanced AI chips developed by Huawei, whose Ascend 950-powered systems are reportedly compatible with the V4 models.
A Major Release Following R1 Breakthrough
This marks DeepSeek’s first major model launch since its R1 system in early 2025, which drew global attention for delivering performance comparable to leading closed-source AI models.
The R1 release had a notable impact on global technology markets, prompting investors to reassess the need for costly AI infrastructure as more efficient models gained traction.
Investment Interest and Market Position
The latest announcement comes shortly after reports that Tencent and Alibaba were exploring potential investments in DeepSeek at a valuation exceeding $20 billion.
DeepSeek is widely recognized as one of China’s leading “AI Tigers,” a group of fast-growing AI firms driving the country’s ambitions in the global artificial intelligence sector.
