Chinese AI company DeepSeek has recently updated its API documentation, announcing that its flagship large model, DeepSeek-V4-Pro, will implement a permanent price reduction. The previously time-limited 2.5-fold discount, which was expected to end on May 31, will now become a permanent pricing structure, effectively reducing costs by 75% compared to the original price.

Absolute Advantage Over U.S. Flagship Models

New benchmark test data from the third-party authoritative evaluation institution Artificial Analysis shows that this price cut has placed DeepSeek-V4-Pro at the top of the global AI model cost-performance ranking. When performing the same AI intelligence index tests, the cost of running this model is only $268.

This figure demonstrates an astonishing level of cost control compared to overseas competitors, with its operational expenses being just one twelfth of those of OpenAI's GPT-5.5. Compared to Anthropic's Claude Opus4.7, the cost advantage is even more staggering—only one nineteenth of that model's cost.

A New Benchmark for Open Source with No Compromise in Specifications

At the same time as the significant price drop, DeepSeek-V4-Pro maintains world-leading performance levels in technical capabilities. Its technical report indicates that the model has achieved notable improvements in agent (Agent) capabilities.