Writer Launches Palmyra X6 to Cut Enterprise AI Token Costs

Enterprise adoption of generative AI is shifting from benchmark scores to the cost and reliability of completing real work. Multi-step agents repeatedly reason, call tools and replay context, driving token consumption far above that of a single chatbot exchange. Goldman Sachs expects global token usage to increase 24-fold between 2026 and 2030, making inference efficiency, orchestration and spending controls central to whether companies can scale AI without unpredictable bills.

San Francisco-based Writer on Aug. 13, 2026, introduced Palmyra X6, a flagship model post-trained from Z.ai’s open-source GLM-5.2, saying it can lower costs for basic tasks by as much as 50%. Writer also upgraded its Agent harness. In a test covering 22 enterprise tasks, the orchestration layer cut average cost per task by 41% to $0.12 from $0.21, reduced token use 38%, and lowered median completion time 44% to 27 seconds from 48 seconds, while task quality remained broadly steady.