AI companies and their customers are paying increasing attention to the cost of running large-scale deployments. While open-source models can reduce the price per token, businesses still face the challenge of choosing the right model and making it efficient for different workloads.
On Thursday, Writer introduced its new flagship AI model, Palmyra X6, alongside upgrades to its agentic harness, the infrastructure that helps AI agents complete tasks. Writer says the combination could reduce customer costs by as much as 50% for basic workloads.
Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram
A Focus On Lower-Cost AI
Palmyra X6 is built as a post-training version of Z.ai’s open-source GLM-5.2 model. Writer says the system is designed to deliver deployment-ready capabilities while using fewer tokens and completing complex, multi-step tasks more efficiently.
Writer CEO May Habib told TechCrunch that enterprise customers are becoming less interested in chasing benchmark records and more focused on controlling the cost of AI deployments.
The company’s new model and harness upgrades are available to Writer customers starting Thursday.
Why The AI Harness Matters
Writer’s approach goes beyond the model itself. The company has also significantly upgraded its standard agentic harness, arguing that improvements to the way AI agents operate can have a major impact on overall costs.
A recent research paper from Writer researchers supports that argument. After testing harness efficiency across multiple models, the researchers found that optimizing the harness was often a more reliable way to reduce costs than simply changing models. Their tests showed an average cost reduction of about 40%.
The researchers described the harness as a component whose efficiency can multiply across every model an organization uses, making optimization potentially valuable even as companies switch between models.
Keeping The Model Choice Flexible
Palmyra X6 will not replace other models available through Writer. Customers can continue using Writer’s models alongside third-party systems imported through Microsoft Azure or Amazon Bedrock.
Habib also sees the growing focus on cost efficiency as a sign that enterprises are becoming more skeptical of major AI labs. She argued that companies are increasingly concerned about the rising cost of AI deployments and whether model providers are sufficiently focused on helping businesses generate practical value.
For Writer, the strategy is therefore not simply about launching another AI model. By combining a lower-cost system with a more efficient agentic infrastructure, the company is positioning cost control as a central part of enterprise AI adoption.
