Writer launches Palmyra X6 and promises up to 50% savings

Tests conducted by Writer researchers show that simple adjustments to the harness can reduce costs by an average of 40%, often more effectively than choosing the model itself. In this context, Writer is launching Palmyra X6 and updates to its harness, available starting Thursday, claiming up to 50% savings for basic tasks. The approach targets complex tasks with fewer tokens and remains agnostic to the model, with cohabitation planned via Azure and Amazon Bedrock.

Internal tests measure 40% savings via the harness

A recent article by Writer researchers examined the impact of small adjustments to the harness on efficiency across several models. According to their findings, these modifications often prove to be more reliable for reducing costs than simply choosing the model, with an average reduction of 40% observed during tests. The researchers emphasize that the harness is the only component whose efficiency affects every model used, present or future.

May Habib describes "unprecedented" costs and distrust

May Habib believes that the surge in costs is unprecedented for clients and that the extent to which CIOs are turning away from AI labs is unprecedented. She views the pressure on costs as a factor contributing to increased distrust towards major labs. She adds that companies want stabilized costs, but no one seems to achieve this, and she argues that AI labs do not truly understand how to help businesses leverage AI.

Palmyra X6 and an updated harness are delivered Thursday

Writer, which offers AI tools and agents for marketers, launched Palmyra X6 on Thursday, a post-training version of the open-source model GLM-5.2 from Z.ai. At the same time, the company released significant updates to its standard agent harness, both available starting Thursday. Writer claims that Palmyra X6 provides ready-to-use capabilities at a much lower cost and believes that, combined with the harness updates, it can enable up to 50% savings for basic tasks. The approach emphasizes complex, multi-step tasks, completed more quickly and with fewer tokens, with harness optimization considered a key lever.

Compatibility announced with Azure and Amazon Bedrock

For clients, the experience remains independent of the model used. Palmyra X6 will work alongside other Writer models as well as external models imported via Azure and Amazon Bedrock.