Writer introduces new AI model and upgraded harness to contain token costs
TechCrunch Russell Brandom ● Covered by 2 sources
Writer launched Palmyra X6 and a better agent harness to cut AI token bills. The company says some basic jobs could cost up to 50% less.
Based on reporting by TechCrunch, Russell Brandom — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Writer is trying to sell something a lot of AI buyers want right now: lower bills. On Thursday, the company launched Palmyra X6, a new flagship model built as a post-training version of Z.ai’s open source GLM-5.2, and paired it with upgrades to its standard agentic harness.
The pitch is simple. Writer says the model and the harness changes should give customers deployment-ready capabilities at a much lower price, with costs for basic tasks falling by as much as 50 percent. Both are available to clients starting Thursday.
What makes the move interesting is that Writer isn’t treating the model as the only place to save money. The company has been pushing the idea that harness efficiency matters just as much, especially for complex multi-step work that needs to run faster and with fewer tokens. A recent paper from Writer researchers backs that up, saying small harness changes across several models often produced more reliable savings than switching models entirely.
In that testing, the researchers found average cost reductions of 40 percent across their experiments. They also argued that harness efficiency compounds across every model an organization uses, now and later. That’s a neat line, but it also reflects where the pain is: enterprises are staring straight at token sprawl and looking for anything that flattens it.
Writer is keeping the setup model-agnostic for its customers. Palmyra X6 will sit alongside other Writer models, and outside models brought in through Azure or Amazon Bedrock. CEO May Habib framed the broader market more bluntly, saying enterprise buyers are tired of chasing benchmarks and that CIOs are increasingly giving up on major AI labs because of the cost explosion.
My take — AI-written commentary, not fact-checked reporting
This is the right fight to pick. The AI market has spent far too long treating token burn like an acceptable tax, as if enterprise buyers should applaud every extra billable thought. Writer is betting that cost control, not benchmark theater, is what gets real deployment — a refreshingly unglamorous thesis in a very glamorous circus.
Read more about this at: TechCrunch