DeepSeek Launches Next-Gen AI Model, Huawei Pledges Full Support with New Chips
DeepSeek has launched its next-generation foundational AI model, the open-source V4, which aims to compete with leading closed-source models from companies like OpenAI and Google DeepMind. This release includes two versions: the V4-pro and the V4-flash, each offering substantial advancements in AI capabilities.
Key Features of DeepSeek’s V4 Models
Both versions of the V4 model were introduced recently. The V4-pro model includes:
- Parameter Count: 1.6 trillion parameters, making it DeepSeek’s largest model.
- Context Window: 1 million tokens, significantly improved from the previous model’s 128,000 tokens.
The V4-flash model is also noteworthy, with 284 billion parameters, providing a balance between performance and efficiency. A higher parameter count typically correlates with greater capabilities, although it also raises computational demands.
Partnerships and Support from Huawei
Huawei has announced its commitment to support these new models fully. The Shenzhen-based technology giant will utilize its Ascend chips along with supernode systems to enhance the performance of DeepSeek’s V4 models during inference.
A live stream event is scheduled for Friday afternoon, where more details about this collaboration will be revealed.
Compatibility with Domestic Technologies
In addition to Huawei, AI chipmaker Cambricon Technologies has declared its compatibility with DeepSeek’s new models. Analysts from Huatai Securities noted that the explicit mention of compatibility with domestic chips suggests a potential boost for local graphics card capabilities and their adoption in the AI sector this year.
Market Impact and Availability
Despite the impressive specifications of the V4-pro model, it is too large for local execution on consumer-grade systems. However, the technical report detailing the model’s architecture and training techniques will be advantageous for global AI developers.
The V4-flash model distinguishes itself as a cost-effective option in the market, offering competitive token pricing similar to DeepSeek’s earlier V2 model released in June 2024.
As the AI landscape evolves, DeepSeek’s focus on performance, efficiency, and strategic partnerships positions it as a significant player in the industry, enhancing the capabilities of artificial intelligence technologies.
