The economics of agent workloads are a major focus of the release. Unlike conventional chat interactions, agents can reason through multiple stages, use external tools and continue working over extended periods, increasing the number of tokens required to complete a task.
"The enterprise wants token consumption to explode — it means adoption is happening — but they need costs to flatten," WRITER CTO and co-founder Waseem AlShikh said. "For the past five years, WRITER has focused on delivering not just frontier-level model performance, but the efficiency, reliability, and governance that highly regulated enterprises need to confidently put agents into production."
WRITER developed its own evaluations for Palmyra X6 using workflows based on work performed by its customers. The tests cover nine areas, including retrieval and grounding, tool use, content creation, delegation to sub-agents and brand voice. They are also designed to measure how well a model follows company context and guardrails during extended tasks.
Palmyra X6 was post-trained on GLM 5.2. In WRITER's evaluations, X6 produced an average score of 0.87 out of 1.00 across the nine categories. The company priced the model at $2 per million input tokens and $8 per million output tokens.
WRITER reported lower scores in the same evaluation for several competing models. Claude Opus 4.8 scored 0.86, Claude Sonnet 4.6 reached 0.85, GPT-5.5 received 0.80 and Gemini 3.1 scored 0.77. The listed input and output prices were $15 and $75 per million tokens for Opus 4.8, $3 and $15 for Sonnet 4.6, $5 and $15 for GPT-5.5, and $2.50 and $10 for Gemini 3.1. Because those results come from WRITER's own internal evaluation rather than an independent benchmark, they specifically reflect the enterprise capabilities and workflows the company chose to measure.
WRITER also says X6 completes tasks in an average of 26 seconds and generates 82 tokens per second. The model can continue working toward one objective for as long as eight hours, according to the company.
Alongside the model, WRITER has rebuilt its agent harness to adjust the amount of reasoning used depending on the complexity of a request. Straightforward tasks can be handled directly, while more involved jobs can trigger structured planning. For larger workloads, the system can process tasks in batches or distribute portions of the work to sub-agents.
WRITER's research found that the harness improvements benefited both its own and third-party models. Across the models it tested, the updated system reduced average task costs by 41% and cut completion time by 44% while maintaining quality, according to the company.
The update also extends WRITER's multi-model approach to WRITER Agent. Administrators can make models from providers including OpenAI and Anthropic available to their teams, with users able to select a model when beginning a session. Different models can also be assigned to specialized functions such as generating images.
Companies are not limited to WRITER's model catalog. Administrators can connect models supplied through Microsoft Azure, AWS Bedrock and NVIDIA NIM. WRITER is pairing those model and agent changes with new controls for tracking how AI is being used inside an organization. Administrators now have a centralized view of WRITER Agent adoption, performance and spending, along with analytics for Playbooks and Skills, which WRITER uses for repeatable workflows and agent instructions.
The governance system can show usage and costs for individual Playbooks and Skills while allowing administrators to manage their ownership and status. Organizations can also establish consumption alerts and limits to control spending across teams. "Governance gave us the confidence to scale AI past a few pilots," said Robyn Murphy, Chief of Staff & Director, Marketing Operations at H&R Block. "Now, with Palmyra X6 bringing the cost per task down, we will have more confidence to take on high-volume work that was never worth automating before."
WRITER says Palmyra X6, the upgraded WRITER Agent capabilities and the new governance features are available to its customers now. Together, the releases position cost per task, orchestration and administrative control as core considerations for enterprises moving agent-based workflows into broader production use.
This analysis is based on reporting from Yahoo Finance courtesy of businesswire.
Image courtesy of WRITER.
This article was generated with AI assistance and reviewed for accuracy and quality.
