DeepSeek Announces Permanent 75% Price Reduction for V4-Pro Model API, Setting a New Low in Global Large Model Pricing

DeepSeek has recently announced a major pricing update: the DeepSeek-V4-Pro model API will not revert to its original price after the limited-time promotion ends on May 31, 2026. Instead, it will be permanently reduced to just one-fourth of its original list price.

This follows an earlier price adjustment on April 26, when DeepSeek lowered pricing across its API lineup, slashing the input cache-hit price to one-tenth of its launch price and introducing a limited-time 40% discount for V4-Pro. The latest move means this highly competitive pricing will now become the permanent standard.

After the adjustment, DeepSeek-V4-Pro’s API pricing now sets a new global low for large language models:

• Input (cache hit): ¥0.025 per million tokens

• Input (cache miss): ¥3 per million tokens

• Output: ¥6 per million tokens

This dramatic price reduction not only demonstrates DeepSeek’s technical capabilities in model architecture optimization and computational efficiency, but also reshapes the commercial pricing landscape through extreme cost control.

As the large model industry accelerates its shift from “technological emergence” to “commercial adoption,” enterprise clients are becoming increasingly sensitive to inference costs and API efficiency. By turning a short-term promotion into a permanent price cut, DeepSeek is expected to significantly lower the barrier for developers and businesses to build and deploy AI applications, enabling more high-frequency, high-throughput use cases.

This strategy will likely not only strengthen DeepSeek’s market position, but also pressure the broader generative AI industry to reconsider its pricing models, pushing global LLM API pricing toward a more accessible and sustainable range.