China's huawei-powered ai sets sights on openai, google and anthropic

DeepSeek's latest AI models, DeepSeek V4, have sent shockwaves through Silicon Valley by leveraging Huawei's Ascend chips to reduce America's grip on the technology. This marks a significant shift in China's quest to build an AI ecosystem less reliant on US-made hardware.

V4 pro and flash: competing on price and performance

DeepSeek has unveiled two new models within the V4 family: V-4 Pro and V-4 Flash. While both boast open architecture and a context window of one million tokens, they differ significantly in terms of scale and cost.

V-4 Pro, DeepSeek's most ambitious offering, packs 1.6 billion parameters and can only activate a fraction of them for each query to conserve resources. In contrast, V-4 Flash scales back to 284 million parameters, aiming for rapid and cost-effective responses.

According to DeepSeek, Flash matches Pro's performance on simple agent tasks, while Pro excels in closed-field applications like reasoning, programming, and general knowledge. The price point for V4-Pro is €1.49 per million tokens of input and €2.97 per million tokens of output, while Flash drops to €0.12 and €0.24, respectively.

Despite the intriguing specs, some experts caution against premature judgment, urging independent testing before fully embracing the new models. Nevertheless, the writing appears clear: China is vying to compete with OpenAI, Google, and Anthropic, while also asserting its autonomy through Huawei's chips.