Chinese startup DeepSeek has made a major announcement in a bid to remain competitive in the artificial intelligence (AI) technology market. The company has declared a 75 percent reduction in usage costs for developers for its newly launched model, “DeepSeek-V4-Pro.” This special discount will be available until 5 May.
The information was reported by Reuters on Monday (27 April). At the same time, the company also announced a reduction in its entire API line’s “input cache hit” cost—related to data retrieval—to just one-tenth of its previous level.
Key Features of DeepSeek V4:
The newly unveiled model, officially launched last Friday, has already generated significant discussion in the technology world. Its main features include:
- Use of Huawei chips: Reducing reliance on U.S. chipmaker Nvidia, the model has been specifically designed for China’s own Huawei Ascend chips.
- Two versions: It is available in two variants—“Pro” and “Flash.” The Pro version is suited for complex tasks, while the Flash version is optimized for faster performance at lower cost.
- Large data capacity: Both versions support a context length of 1 million tokens, significantly higher than the previous V3 model.
- AI agent capability: Designed to function as an “AI agent,” the V4 model can perform more complex and autonomous tasks compared to traditional chatbots.
Position in Competition
According to DeepSeek, its “V4 Pro” version ranks among the best open-source models in the world. In terms of general knowledge and mathematical reasoning, it is approaching the performance of Google’s powerful model “Gemini-Pro-3.1.” Although Gemini remains slightly ahead due to its closed-source nature, DeepSeek currently leads in the open-source category.
Why the Price Cut?
Market analysts suggest that DeepSeek has adopted a cost-effective strategy to compete with giants such as OpenAI, Google, and Anthropic. Despite strict chip export policies imposed by Western countries, DeepSeek V4 demonstrates that China can remain competitive by leveraging its own hardware and advanced software.
Currently, the standard cost of DeepSeek’s V4 Pro model is $1.74 per 1 million input tokens and $3.48 per 1 million output tokens. With the 75 percent discount, it becomes highly affordable for developers.
DBTech/BMT/OR
