Writer, the enterprise agent platform used by large marketing and revenue teams, released Palmyra X6 on August 13. The company did not train a new base model from scratch. X6 is a post-trained version of Z.ai’s open-weight GLM-5.2, tuned on Writer’s own production workflows: staying grounded in company knowledge, following brand rules, calling tools, and staying coherent on long jobs.
On Writer’s nine-skill internal scorecard, X6 averaged 0.87 out of 1.00 at $2 per million input tokens and $8 per million output tokens. That sat just ahead of Claude Opus 4.8 (0.86 at $15 / $75) and Claude Sonnet 4.6 (0.85 at $3 / $15), and further ahead of GPT-5.5 and Gemini 3.1 on the same tests. Writer says the model finishes a typical task in about 26 seconds, writes at 82 tokens per second, and can work toward one goal for up to eight hours without drifting off.
The model is only half the story. Writer also rebuilt the agent harness that sits around it. Across Writer and third-party models, that harness finished jobs 44 percent faster at 41 percent lower cost. Paired with X6, Writer claims 52 percent lower cost, 48 percent more speed, and a 10 percent quality lift versus its prior stack. Admins can still point the same harness at Anthropic, OpenAI, Azure, AWS Bedrock, or NVIDIA NIM models. New spend dashboards let leaders see which playbooks burn tokens and set alerts.
Source: https://writer.com/blog/palmyra-x6-major-harness-release/
