Writer Unveils AI Model & Harness to Slash Token Costs

Across the artificial intelligence industry, organizations are increasingly recognizing the substantial expenses associated with their AI deployments, prompting a renewed urgency to implement cost-saving measures. While open-source models offer significantly lower per-token costs, identifying the optimal model for specific operational requirements can often prove challenging.

Addressing this industry-wide concern, Writer, a company specializing in AI tools and agents for marketers, introduced its new flagship model, Palmyra X6, on Thursday. Designed to mitigate these challenges for its users, Palmyra X6 is built as a post-training variant of Z.ai’s open-source GLM-5.2 model. Writer states that this new system is poised to deliver deployment-ready capabilities at a considerably reduced price point. The company estimates that the integration of Palmyra X6, combined with enhancements to its harness infrastructure, will enable customers to achieve cost reductions of up to 50 percent for routine tasks.

In conjunction with the new model, the company also unveiled significant upgrades to its standard agentic harness. Both innovations became available to Writer clients starting Thursday.

“I think the enterprise is absolutely sick of chasing the next benchmark,” CEO May Habib told TechCrunch. “They want flattening cost, and it seems like nobody can deliver that.”

This novel approach places a particular emphasis on optimizing complex, multi-step tasks, enabling their execution with greater speed and reduced token consumption. Writer identifies harness optimization as a crucial enabler for achieving these efficiencies.

A recent research paper published by Writer’s own researchers lends considerable weight to this strategy. The study involved testing minor adjustments in harness efficiency across a diverse range of models. The research revealed that, in numerous instances, modifications to the harness proved to be a more dependable method for reducing costs than the choice of model itself, resulting in an average cost decrease of 40% throughout their testing.

“The harness is the one component whose efficiency multiplies across every model an organization runs—present and future,” the researchers articulated.

For Writer’s clientele, the operational experience remains model-agnostic; Palmyra X6 is designed to function alongside other proprietary Writer models or external models imported via platforms such as Azure or Amazon Bedrock. However, Ms. Habib also perceives the push for cost reduction as fostering a broader distrust towards major AI laboratories, which she believes possess a financial incentive to drive up token usage.

“The cost explosion here is just unprecedented for customers, and so is the degree to which CIOs are giving up on the labs,” Habib conveyed to TechCrunch, further asserting that AI labs “don’t deeply understand right how to help an enterprise get benefit from AI.”