DeepSeek Launches V4.1 Flash-1017 Model with Over 50% Reduction in Token Usage

Odaily Planet Daily reports: DeepSeek has officially launched its new large model version, V4.1 Flash-1017. Amid the current critical phase of iteration for domestic AI large models, this version has率先 completed its upgrade, achieving significant performance improvements. Previously disclosed technical metrics have already been made public; now, DeepSeek has for the first time independently released its Pareto frontier curve, visually illustrating the model’s comprehensive performance across efficiency and cost dimensions. V4.1 Flash-1017 reaches industry-leading levels on this curve: its per-task composite cost is only one percent to one twentieth that of similar models such as Fable and Astra, offering a cost-performance advantage of over 40 to 50 times. Although the newly released 1017 version maintains the existing API pricing structure, it has undergone a deep restructuring of token usage efficiency—through task-aware compression, dynamic context trimming, and adaptive decoding strategies—it significantly reduces actual token consumption across various request types. Real-world tests show that token usage has decreased by over 50% in most scenarios, with corresponding TCO levels also set to undergo a new substantive decline. (Sina Finance)