Writer · 2026-08-13 · major
Palmyra X6 — Writer's flagship model halves the cost of an agent task
Palmyra X6 is Writer's new flagship model, post-trained from Z.ai's open-weight GLM-5.2. Writer says per-task cost and latency are roughly halved against its previous generation, and the model can run unattended for up to eight hours.
Writer's new flagship model, post-trained from GLM-5.2, runs agent jobs unattended for up to eight hours.
Quick facts
| Maker | Writer |
|---|---|
| Model ID | palmyra-x6 |
| Base model | GLM-5.2 (Z.ai, open weights) |
| Context window | 1M tokens |
| Unattended run time | Up to 8 hours on one goal |
| Availability | WRITER Agent and AI Studio |
| What's new | Per-task cost and latency roughly halved vs the previous generation |
Benchmarks
| Palmyra X6 | 0.87 | |
|---|---|---|
| Claude Opus 4.8 | 0.86 | |
| GPT-5.5 | 0.8 | |
| Gemini 3.1 | 0.77 |
Pricing
| Input | $2.00 / 1M tokens |
|---|---|
| Output | $8.00 / 1M tokens |
What is it?
Palmyra X6 targets the high-volume research, personalization and content work that go-to-market teams run every day, and is priced to be deployed across thousands of employees. Writer post-trained it from GLM-5.2, the open-weight mixture-of-experts model from Z.ai, rather than training a base model from scratch. Writer measures per-task cost and latency at roughly half its previous generation with no quality regressions.
How does it work?
The saving comes from two layers stacked together, not from the model alone. Writer rewrote the harness — the orchestration code that decides how an agent plans, batches work, hands jobs to sub-agents and recovers from failed tool calls — and reports that layer by itself makes WRITER Agent 44% faster and 41% cheaper per task across every model it tested, third-party ones included. Adding Palmyra X6 on top brings the combined figures to 52% lower cost, 48% better speed and 10% better quality.
Why does it matter?
Token spend, not model quality, is what stops most companies moving agents from pilot to production. Admins now get a real lever over it: AI Studio reports live consumption by team, billing group or seat, and can raise an alert or block work outright above a set limit. Writer also lets admins enable outside models from Microsoft Azure, AWS Bedrock and NVIDIA NIM in the same governed workflow.
Who is it for?
enterprise go-to-market and platform teams
Frequently asked questions
- How does Palmyra X6 compare to Claude Opus 4.8 and GPT-5.5?
- On Writer's own evaluation of nine agent capabilities — covering grounding and retrieval, tool use, content generation and sub-agent delegation — Palmyra X6 averaged 0.87, ahead of Claude Opus 4.8 at 0.86, GPT-5.5 at 0.80 and Gemini 3.1 at 0.77. These are Writer's internal numbers on Writer's own task mix, not an independent public benchmark.
- Can Palmyra X6 run a job without someone watching it?
- Writer says Palmyra X6 holds coherent reasoning on a single goal for up to eight hours unattended. That is the change Writer points to for background automation: long jobs previously failed on cost and reliability rather than capability. SiliconANGLE reports the model averages 26 seconds per task and produces 82 tokens per second.
- Is Palmyra X6 open source?
- No. Writer sells Palmyra X6 through its own API and platform rather than publishing the weights. The base it was post-trained from — GLM-5.2, the mixture-of-experts model from Beijing-based Z.ai — is open weight, but the Palmyra X6 post-training and the model itself stay proprietary to Writer.
- What changed in WRITER Agent alongside Palmyra X6?
- Writer rebuilt the harness around the model. WRITER Agent now completes tasks 44% faster and at 41% lower cost on average across every model tested, including third-party ones, by batching high-volume work, delegating to sub-agents and recovering from tool errors. AI Studio also gained live token reporting plus alerts and hard blocks above a spend limit.
Try it
Model ID: palmyra-x6 — available in WRITER Agent and AI Studio