Rayter launches Palmera X 6, a new model and structural upgrade that cuts token costs by 50%

Awareness of the high cost of operating and deploying intelligent models is growing among institutions and companies across the AI sector, creating a urgent desire and increasing pressure to reduce these operational expenses. Although open-source models offer a lower processing cost per token compared with closed models, institutions are often puzzled by the difficulty of identifying and coordinating the appropriate model for each specific task within work environments.

In this context, Rayter, which specializes in providing AI tools and agents for marketing sectors, launched its new flagship model called Palmera X 6, in a direct attempt to address the cost crisis and simplify model selection for its customers. The new system was built as a post-training fine-tuned version based on the open-source GLLM 5.2 model from Z AI, and Rayter states that the system offers capabilities ready for immediate deployment and practical use at a much lower cost.

The company expects that combining the new Palmera X 6 model with the comprehensive updates it has made to its infrastructure for the runtime architecture will reduce institutions’ operational costs by up to 50 % in core tasks.Rayter made both the new model and the structural updates to the agent system available to customers starting Thursday, to boost data-processing efficiency without sacrificing result quality.

In remarks to TechCrunch, Mi Habib, Rayter’s chief executive officer, said that large institutions and companies are now refusing the endless chase of academic benchmark and performance tests, and are looking for genuine solutions to stabilize costs and flatten the spending curve, something many have been unable to provide so far. Rayter’s new vision focuses on executing complex multi-step tasks faster and with lower token consumption, confirming that improving the runtime architecture is the primary and most important lever to achieve this operational equation.

The vision is based on a recent research paper published by Rayter researchers, in which they tested the effect of making small adjustments to runtime-architecture efficiency across a variety of models.The results showed that targeted improvements to the runtime architecture were more reliable for reducing costs than merely changing the type of model used, with costs dropping by an average of 40 % across different test environments.The researchers noted in their papers that the runtime architecture is the only component whose efficiency multiplies across every model an institution runs today and in the future.

At the user-experience level, Rayter’s system remains model-agnostic, as Palmera X 6 operates alongside Rayter’s other models or external models imported through platforms such as Microsoft Azure or Amazon Bedrock. This flexibility allows institutions to deploy different capabilities according to the nature of their operational flows and to choose the most cost- and performance-efficient paths.

Mi Habib believes that the current drive to cut expenses is fostering growing mistrust of the major AI labs, which have financial incentives and a direct interest in raising token consumption rates and increasing bills.Habib affirmed that the present surge in operational costs is unprecedented for customers, prompting senior technology executives to back away from total reliance on the major labs that do not understand the depth of companies’ needs to achieve real benefit from AI.This step shows how runtime architecture and agent structures are becoming the real arena of competition for controlling computing costs and managing AI budgets within institutions.