In the realm of artificial intelligence, users are increasingly aware of the high costs associated with their implementations, prompting a renewed focus on cost reduction. While open source models provide reduced costs per token, selecting the appropriate model for specific tasks remains a challenge.
On Thursday, Writer, a company specializing in AI tools for marketers, introduced its latest model, Palmyra X6, designed to address this issue. This model is a post-training adaptation of Z.ai’s open-source GLM-5.2, and Writer claims it will offer capabilities ready for deployment at significantly lower costs. The company predicts that, when combined with infrastructure updates, this new model could decrease client expenses by up to 50% for routine tasks.
In tandem with the model, Writer announced numerous enhancements to its standard agentic harness. Both new features are accessible to Writer customers starting today.
“Enterprises are clearly frustrated with the ongoing pursuit of the next benchmark,” noted CEO May Habib. “They are seeking stable costs, yet no one seems able to provide that.”
The new strategy especially targets complex, multi-step tasks that can be completed more quickly and with fewer tokens. Writer believes that optimizing the harness is vital for achieving this goal.
Recent research from Writer’s team supports this strategy by examining small modifications in harness efficiency across various models. The findings indicated that adjusting the harness often led to more significant cost reductions, averaging 40%, compared to merely changing the model.
“The harness is a unique element whose efficiency impacts every model an organization employs, both now and in the future,” the research concluded.
For Writer’s users, the experience remains model-agnostic: Palmyra X6 will be available alongside other Writer models and outside models sourced from platforms like Azure or Amazon Bedrock. However, Habib also perceives the drive to lower costs as contributing to a growing skepticism towards leading AI laboratories, which have economic incentives to increase token usage.
“The surge in costs is unprecedented for clients, and CIOs are becoming increasingly disillusioned with the labs,” Habib noted, adding that these AI laboratories “lack a deep understanding of how to assist enterprises in deriving value from AI.”
