DeepSeek, Xiaomi Just Made Frontier AI 99% Cheaper. American Labs Went the Other Way
DeepSeek made a 75% discount on DeepSeek V4-Pro permanent, and Xiaomi slashed MiMo-V2.5 API prices by up to 99% for cached inputs. These aggressive reductions make two of the most capable AI models vastly cheaper, especially compared to U.S. labs that have raised or held steady on prices. API pricing, critical for businesses integrating AI, now allows up to 82 billion tokens for $100 via Xiaomi—over 60 billion words. Xiaomi achieved this by optimizing how prior information is stored and reused, reducing computation and storage costs by about 80%. DeepSeek V4-Pro uses interleaved compression and attention methods to shrink cache size and cut inference cost by 73%. As a result, both models now run at $0.435 (input) and $0.87 (output) per million tokens, with cache hits at an ultra-low $0.0036. These prices are 15–30 times lower than leading American models, with equivalent technical performance. The price gap further collapses for workloads with stable prompts and repeated context, significantly lowering the operational costs for AI-powered products in China compared to the U.S.
