Artificial Analysis Rates DeepSeek V4.1 Flash 1017 at 45, Up From 39, at $0.26 a Task

Artificial Analysis has rated DeepSeek V4.1 Flash 1017 at 45, an increase from the previous score of 39, with a cost of $0.26 per task. DeepSeek's release notes indicate the model provides shorter outputs than its predecessor. Meanwhile, the Max variant of DeepSeek V4.1 Flash 1017 focuses on high-efficiency engineering. Its technical architecture specifically targets inference stack bottlenecks, including memory bandwidth utilization, compute-to-parameter efficiency, and cross-node latency through the use of sparse activations.