Valentina Akpan: Founder, Rellatech. Admin and technical virtual assistant for businesses and startups with teams.
An AI company launched a flagship model this week and barely mentioned how smart it is. The entire announcement is about money.
On Thursday, Writer released Palmyra X6, built on Z.ai's open source GLM-5.2, alongside upgrades to its agent harness. According to Writer's own release notes, the model paired with its agent delivers an average 52 percent lower cost, a 48 percent improvement in speed, and a 10 percent improvement in quality. The model is priced at $2 per million input tokens and $8 per million output tokens. Writer says it completes tasks in 26 seconds on average, generates 82 tokens per second, and scored an average of 0.87 out of 1.00 across nine evaluations. The harness on its own accounts for 41 percent lower cost per task and 44 percent faster completion.
Those are the vendor's numbers, and I am reporting them as such. What I find more interesting is how the company talks about why it built this. Writer CEO May Habib told TechCrunch: "I think the enterprise is absolutely sick of chasing the next benchmark. They want flattening cost." She added that "the cost explosion here is just unprecedented for customers, and so is the degree to which CIOs are giving up on the labs."
The detail I keep coming back to
Buried in the same announcement is a finding from Writer's own researchers. In many cases, changes to harness efficiency were "a more reliable way to reduce costs than model choice."
Read that again in plain terms. How the workflow is wired matters more to your bill than which model you picked. The expensive part is not the brain. It is the plumbing around it, the retries, the redundant calls, the step that re-reads the same document four times because nobody told it not to.
That is not a model problem. That is an operations problem, and it is exactly the kind of thing nobody notices until the invoice arrives.
You are not an enterprise, and that is the risk
The CIOs in that TechCrunch piece have finance teams, procurement, and someone whose actual job is to question a line item. You have a company card on file and an automation somebody set up in March.
Same runaway meter. Nobody watching it. And Writer notes the model can work unattended toward a single goal for up to eight hours, which is a real capability and also eight hours of unsupervised spending.
I see the small version of this constantly. A client is paying for three tools that do the same job because each was added to solve one urgent thing. A workflow fires on every record update instead of once a day. None of it is dramatic. It just quietly adds up, which is why it survives. This is the unglamorous middle of my CRM and automation work, and a standing reason clients keep me on as an operations virtual assistant rather than booking a one-time build.
What to do with this news
Pull your last three months of invoices for every AI, automation, and integration tool you pay for. Put them on one page. For each line, write down what it does and who asked for it. You will find at least one thing you forgot you were buying, and probably one workflow running far more often than it needs to. My monthly retainers give businesses and startups with teams a dedicated person who runs the systems, keeps the spend honest, and reviews what the automations actually produced instead of assuming they are fine. Book a consultation and I will go through your stack with you.
Contact MeNot ready for a call? Take the free operations assessment and I will send you a written breakdown within 24 hours.
Related Services