Enterprise AI spending is quietly becoming a crisis. Companies that rushed to deploy AI agents and workflows are now staring at token bills that nobody budgeted for, and the pressure to bring costs down is real. Writer, which builds AI tools and agents for marketing teams and enterprise clients, is making a direct play for that frustration with two releases at once: a new flagship model and a significantly upgraded agentic infrastructure layer.
As TechCrunch reported, the company launched Palmyra X6 on Thursday alongside updates to its standard agentic infrastructure, both available to clients immediately. Palmyra X6 is built as a post-training variation on Z.ai’s open source model GLM-5.2, and Writer says the combination of the new model and infrastructure changes could reduce costs by as much as 50% for basic tasks. That’s not a trivial number for companies running multi-step AI workflows at scale.
The infrastructure angle is actually the more interesting story here. Writer recently published research testing small efficiency changes across multiple models and found that, in many cases, optimizing the infrastructure layer was a more reliable cost-reduction strategy than switching models entirely. Costs fell an average of 40% across their testing. That’s a meaningful finding because it challenges the default assumption that model selection is the primary cost lever. “The harness is the one component whose efficiency multiplies across every model an organization runs, present and future,” the researchers wrote.
CEO May Habib has a pointed view on why this matters beyond the technical specs. “I think the enterprise is absolutely sick of chasing the next benchmark,” she told TechCrunch. “They want flattening cost, and it seems like nobody can deliver that.” She also pointed to a broader dynamic: the major AI labs have a financial incentive to drive up token consumption, which puts them structurally at odds with their enterprise customers. “The cost explosion here is just unprecedented for customers, and so is the degree to which CIOs are giving up on the labs,” she said.
For Writer’s clients, the platform stays model-agnostic. Palmyra X6 sits alongside other Writer models and outside models brought in through Azure or Amazon Bedrock. So customers aren’t locked in, they’re just getting a cheaper, faster default option for common tasks. The new model puts specific emphasis on complex, multi-step reasoning executed with fewer tokens, which is exactly where enterprise costs tend to pile up.
Where does Writer fit in the competitive picture? The company competes with platforms like Cohere, which also focuses on enterprise deployments and cost efficiency, and broader app-layer plays from Salesforce and ServiceNow that are embedding AI into existing workflows. But Writer’s positioning, especially the infrastructure-level cost argument, is fairly distinct. Most competitors lead with model capability. Writer is leading with total cost of ownership, which is increasingly what procurement conversations actually look like in 2025.
The open source angle also matters. Building on GLM-5.2 rather than a proprietary foundation model gives Writer more pricing flexibility and avoids the royalty and rate structures that come with licensing commercial models. That’s a real structural advantage if the cost story continues to dominate enterprise AI buying decisions, which, right now, it does.
- Palmyra X6 is built on Z.ai’s open source GLM-5.2 with post-training customization
- Combined model and infrastructure changes target up to 50% cost reduction for basic tasks
- Internal research showed infrastructure optimization alone reduced costs by an average of 40%
- The platform remains model-agnostic, supporting outside models via Azure and Amazon Bedrock
- Both Palmyra X6 and the updated infrastructure are available to Writer clients now
Whether the 50% cost reduction holds at real enterprise scale is still an open question. But the framing, that controlling costs at the infrastructure level is as important as model selection, is the right argument to be making right now. Enterprises aren’t looking for the smartest model. They’re looking for the most predictable bill.
