DeepSeek's New Model Cuts Token Consumption by Over 50%

Odaily Report: DeepSeek has officially launched a new version of its large model, V4.1 Flash-1017. At a time when domestic AI large models are generally in a critical iteration phase, this version has taken the lead in completing its upgrade, achieving a significant leap in overall performance. While public data on its technical metrics had been previously available, this time DeepSeek has for the first time independently published a Pareto frontier curve, visually presenting the model's comprehensive performance across efficiency and cost dimensions. V4.1 Flash-1017 reaches industry-leading levels on this curve: its comprehensive cost per unit task is only one-hundredth to one-twentieth that of comparable models such as Fable and Astra, delivering a cost-effectiveness advantage of more than forty to fifty times. Although the newly released 1017 version maintains the existing API pricing structure, it has completed a deep restructuring of Token usage efficiency—through task-aware compression, dynamic context trimming, and adaptive decoding strategies, significantly reducing the actual Token consumption of various types of requests. Empirical testing shows that in most scenarios, Token usage has dropped by over 50% compared with the previous version, and the corresponding TCO level will likewise see a new round of substantial reduction in tandem. (Sina Finance)