DeepSeek V4 Pro Performance and Leaked Funding Document

A leaked DeepSeek funding document on August 6, 2026 revealed details of the V4 Pro model. - Performance placed in the global top tier, coding ability only 0.3% weaker than Claude's flagship. - Million-token long-context window supported. - API pricing stated as 1/10 to 1/100 of overseas competitors, delivering a leading cost-performance ratio. - Native adaptation to Huawei Ascend 950PR, forming a core of China's domestic 'model + chip' route and achieving a closed-loop domestic computing power. - One user noted the minimal performance gap to Claude justifies a price increase, while hoping for only modest rises to extend usage quotas on AI platforms.
Key insights
V4 Pro Delivers Near-Claude Coding Performance at a Fraction of the Cost
The leaked document states DeepSeek V4 Pro's coding capability is only 0.3% behind Claude's flagship, putting it at the forefront of global models. Combined with API pricing that is 1/10 to 1/100 of foreign alternatives, the model offers a cost-performance gap that is described as insurmountable by competitors. The million-token context further enhances its utility for complex tasks.
Native Huawei Ascend 950PR Integration Closes the Domestic AI Hardware Loop
V4 Pro is natively adapted to Huawei Ascend 950PR, positioning it as a core component of one of China's three major independent 'model + chip' routes. This integration achieves a closed-loop for domestic AI computing, reducing reliance on foreign GPU supply chains and aligning with national self-sufficiency goals.