Palmyra X6
Writer · Released Aug 2026
Palmyra X6 is Writer's flagship agentic model for marketing and revenue workflows — post-trained on Z.ai's open-weight GLM-5.2 at $2/$8 per 1M tokens, a fraction of the frontier rate. Every headline number is Writer-run, and the GLM-5.2 provenance is a real procurement question. Adopt it inside Writer for cost-sensitive GTM agent work; look elsewhere for independently benchmarked general capability.
Is it right for you?
Good for
- Cost-sensitive enterprise GTM agents — $2/$8 per 1M tokens undercuts Claude Opus 4.8, GPT-5.5, and Gemini 3.1 on Writer's own launch comparison
- Marketing and revenue workflows — top internal scores in brand voice (0.92), grounding/retrieval (0.91), and content generation (0.88)
- Long-horizon unattended execution — sustains a single goal up to 8 hours; 82 tokens/sec, 26-second median finished task
- 1M-token context for connector-driven work across CRM, Slack, documents, and Knowledge Graphs
- Bias-averse regulated enterprises — pre-registered model-risk eval (19,674 judged responses); best ModelSlant both-sides rate (80%) and FORTRESS +8.6 over the base
Not good for
- Independently benchmarked general capability — X6 publishes no public benchmarks; all scores are Writer's internal eval (n=22, baseline frozen 2026-06-07)
- Deployments outside Writer's platform — no Bedrock model ID at launch (X5/X4 have one); tuned to co-evolve with the Writer Agent harness
- Buyers with a hard block on Chinese open-weight foundations — post-trained on Z.ai's GLM-5.2 (MIT license), a supply-chain question for some US enterprises
How it performs by task
Grounding and retrieval
0.91 internal capability score — Knowledge Graph/document grounding, retrieval and citation
Brand voice
0.92 — strongest single capability; applies brand voice and custom instructions to output
Content generation
0.88 — drafts and edits long- and short-form content to a brief
Sub-agent delegation
0.86 — delegating work to spawned sub-agents and merging results (Writer-run internal eval)
MCP tool use
0.88 — connector-driven actions across CRM, Slack, and documents
Long-horizon unattended execution
Up to 8 hours on a single goal; 82 tokens/sec, 26s median finished task
Pricing
Input
$2 / 1M
Output
$8 / 1M
Context
1M context, 8K max output
Benchmarks
No verdict changes yet
The clock starts day one. Changes land here as our verdict evolves.
Sources
Verification log
No verification checks yet
We haven't logged a verification check for this entry. Once a check runs, its history shows here.