The race to the bottom in China’s artificial intelligence sector has reached a new, aggressive milestone. DeepSeek, the domestic startup that has consistently punched above its weight, announced a radical pricing restructure for its API services. By slashing input cache hit costs to a mere fraction of their previous rates, the company is effectively commoditizing the infrastructure required for complex AI workflows.
Specifically, the cost for cached input hits across the DeepSeek series has been reduced to one-tenth of the original price. For its flagship DeepSeek-V4-Pro model, a time-limited 75% discount brings the price down to a staggering 0.025 RMB per million tokens. This pricing strategy targets "input caching," a technical feature that allows the model to "remember" previous prompts, making it significantly cheaper to run repetitive or high-context tasks common in autonomous agents.
The impact was immediate and profound. Following the announcement, API call volumes reportedly surged fourfold as developers rushed to leverage the lowered barriers to entry. By pricing its Pro and Flash models at such extremes, DeepSeek is not just competing with other startups; it is challenging the fundamental price anchors of the entire industry, forcing rivals to reconsider their margins.
This move highlights a broader shift in the AI landscape from pure capability to operational efficiency. While giants like OpenAI and Anthropic focus on scaling parameters, DeepSeek’s strategy suggests that the next phase of the AI war will be won by those who can provide the most affordable "intelligence-per-watt." For the global market, this sets a daunting precedent for the cost of running large-scale AI ecosystems.