Writer launches Palmyra X6, harness upgrade to cut AI costs

Writer, which offers AI tools and agents for marketers, launched a new flagship model on Thursday called Palmyra X6. It is built as a post-training variation on Z.ai's open source model GLM-5.2. Alongside the model, Writer released a significant upgrade to its standard agentic harness, the software layer that manages how a model carries out a task. Both the new model and the harness upgrade became available to Writer's clients starting Thursday.

Writer says the new system should provide deployment-ready capabilities at a much lower price, and the company estimates that combining Palmyra X6 with the harness changes will cut costs for its customers by as much as 50% on basic tasks, a ceiling figure rather than a guaranteed or average result. The new approach targets complex, multi-step tasks in particular, aiming to carry them out faster and with fewer tokens. Writer treats harness optimization, not model choice alone, as the central lever for getting there.

That emphasis is backed by a recent paper from Writer's own researchers, who tested small changes to harness efficiency across multiple different models. They found that in many cases, changing the harness was a more reliable way to cut costs than switching models, with costs falling an average of 40% across their testing, a separate figure from the 50% ceiling Writer cites for Palmyra X6's own launch. The researchers describe harness efficiency as the one gain that compounds across every model an organization runs, present and future.

For Writer's clients, the experience stays model-agnostic: Palmyra X6 sits alongside Writer's other models and outside models imported through Azure or Amazon Bedrock. CEO May Habib told TechCrunch that enterprises are sick of chasing the next benchmark and want flattening costs that nobody has been able to deliver. She connects that pressure to a broader shift: growing distrust toward major AI labs, which the article describes as having a financial incentive to drive up token use. The cost explosion, she said, is unprecedented for customers, and so is the degree to which CIOs are giving up on those labs, adding that they do not deeply understand how to help an enterprise get benefit from AI.

Key facts

Writer launched a new flagship model, Palmyra X6, on Thursday, built as a post-training variation on Z.ai's open source GLM-5.2, together with a major upgrade to its standard agentic harness; both became available to clients the same day.
Writer estimates that combining Palmyra X6 with the harness changes can cut customer costs by as much as 50% on basic tasks, a ceiling estimate rather than a guaranteed or average figure.
A separate paper by Writer's own researchers found that changing the harness, more than switching models, was the more reliable cost lever in many cases, with costs falling an average of 40% across their testing.
Palmyra X6 stays model-agnostic within Writer's platform, sitting alongside Writer's other models and outside models imported through Azure or Amazon Bedrock.
CEO May Habib said enterprises are sick of chasing the next benchmark and want flatter costs, framing the push as part of a broader CIO shift away from major AI labs, which the article says have a financial incentive to drive up token use.

Why it matters

The launch responds to a cost-conscious moment the article frames at the industry level: AI deployments are getting more expensive, and while open source models offer lower per-token costs, matching one to a given job is not straightforward. Writer's answer bundles two things released together on the same day: a new model, Palmyra X6, built as a post-training variation on Z.ai's open source GLM-5.2, and a substantial upgrade to Writer's agentic harness, the layer that manages how a model executes multi-step tasks. Habib frames the underlying shift as enterprises growing sick of benchmark chasing and wanting flatter, more predictable costs instead. Writer treats harness optimization as the more durable lever for getting there: a recent paper from its own researchers tested harness-efficiency changes across multiple models and found they cut costs by an average of 40% in many cases, more reliably than switching models outright. That research finding is separate from, but supports, the up to 50% cost-cut ceiling Writer cites for combining Palmyra X6 with the new harness on basic tasks.

Who it affects

The direct audience is Writer's enterprise customers, mainly the marketing teams the company already serves, who get both the new model and the harness upgrade starting Thursday, with no separate migration step described in the source. Because Palmyra X6 and the harness upgrade sit alongside Writer's other models and outside models imported through Azure or Amazon Bedrock, existing Writer deployments are not locked into switching. Habib frames the audience more broadly as CIOs, who she says are giving up on major AI labs over runaway costs, labs the article describes as having a financial incentive to drive up token use.

How to use it

Both Palmyra X6 and the harness upgrade became available to Writer clients starting Thursday. The source gives no per-token or subscription price for Palmyra X6, and no detail on how it compares to earlier Palmyra models or whether it replaces one. Clients can run it alongside Writer's other models, or alongside outside models brought in through Azure or Amazon Bedrock, so adopting Palmyra X6 does not require dropping models already in use.

How solid is it

Both headline figures come from Writer itself: the up to 50% cost-cut estimate is the company's own projection for combining Palmyra X6 with the harness changes, and the 40% average reduction comes from a paper by Writer's own researchers, not an independent study. The source names no customer case study or outside validation for either number, does not detail which models were tested for the 40% figure or over what tasks or time period, and does not define what counts as a basic task for the 50% figure or say what the cost cut looks like on other work. The harness paper itself is undated and unnamed in the source, described simply as a recent paper, with no link and no named authors, so its methodology cannot be checked from the article alone.

Risks and caveats

Palmyra X6 is a post-training variation of Z.ai's open source GLM-5.2 rather than a model built from scratch, tying Writer's flagship release to a third party's base model; the article carries no response from Z.ai. Both cost figures are self-reported, drawn from Writer's own testing and estimates rather than outside verification, and Writer has a commercial interest in customers believing the cuts are achievable. No technical specifications, such as parameter count, context window or benchmark scores, are given for either Palmyra X6 or GLM-5.2, and the source gives only the day of the week, Thursday, for the launch, with no absolute calendar date.

“The cost explosion here is just unprecedented for customers, and so is the degree to which CIOs are giving up on the labs.”

— May Habib, CEO of Writer