During the last weekend of April 2026, China's AI industry was abuzz with excitement over a series of moves by DeepSeek. On April 24, DeepSeek released the preview version of its V4 series, open-sourcing two models: Pro and Flash, both supporting ultra-long contexts of up to one million tokens. The following days, on the 25th and 26th, DeepSeek successively slashed prices, reducing the input cache hit price per million tokens to 0.02 yuan for V4-Flash and 0.025 yuan for V4-Pro, setting a new global low for large model pricing. The price reductions for the V4 series were made possible by a revolution in underlying architectural efficiency, significantly decreasing the floating-point operations required for single-token inference and the KV cache footprint. Meanwhile, eight domestic AI chip brands, including Huawei Ascend and Cambricon, announced the completion of their adaptation to DeepSeek-V4, enabling a synergy between domestic computing power and DeepSeek that is expected to unlock a vast market for AI applications.
