カオスマップに戻る

Overview

Writer is a generative AI platform purpose-built for the enterprise. It is powered by Palmyra, Writer's in-house family of large language models, and is targeted at large organizations where security, governance, and brand consistency are non-negotiable. The flagship today is Palmyra X6, announced August 13, 2026 and now the default model in WRITER Agent for long-horizon enterprise workflows. Writer occupies a unique niche: building AI agents for regulated industries that generic chat AIs cannot serve. Primary audience: Fortune 500 companies, banks, insurers, healthcare, and law firms.

Key Features

1. Palmyra X6 (Flagship Model)

A sparse mixture-of-experts model with 744B total and roughly 40B active parameters per token, post-trained for tool use on the open-weight GLM-5.2 base via Anchored Supervised Fine-Tuning (ASFT). It beats Writer's previous default across six benchmarks, including top marks on BFCL Core for tool calling, and is priced at $2 per 1M input tokens and $8 per 1M output. Average cost per task lands around $0.12 — roughly half the previous generation.

2. Model Line-Up

Three models are on offer: Palmyra X6 as the agentic default, Palmyra X5 when you want a 1M-token context window cheaply ($0.60 in / $6 out), and Palmyra X4 with a 128k window. The old domain-specific variants for finance, medicine, and creative work have been retired and folded into the general Palmyra line.

3. Enterprise Agent Building and the Harness

Build autonomous agents that call external tools and delegate to other Writer agents. Deployed in financial reporting, legal document analysis, medical record synthesis, and customer feedback analysis. Writer also rebuilt the harness — the orchestration layer that decides how an agent gathers context, calls tools, and manages steps. By its own measurements the harness changes alone cut blended cost per task by 41%, and together with X6 the reduction reaches 52%.

4. Full Governance

Customers own their data and nothing is used for training. SOC 2 Type II, HIPAA, and GDPR compliant. Audit logs, access controls, and automatic brand-guideline enforcement are standard.

5. Amazon Bedrock Integration

Palmyra X5 is also available via Amazon Bedrock, making it easy to drop into existing AWS estates while preserving data sovereignty and regional controls.

Plans & Pricing

Writer is enterprise-first, and Starter is the only self-serve tier with a published price.

PlanMonthlyHighlights
Starter$29/user (annual) / $39/user (monthly)Capped at 5 Pro seats, basic integrations. No permanent free plan — trial only
EnterpriseCustomFull Palmyra access, AI Studio, API, SSO, audit, dedicated support

Below Starter sit Lite seats at $0, a view-oriented tier you can hand to additional people on your domain at no extra cost. API pricing per 1M tokens breaks down as follows.

ModelContextInputOutput
Palmyra X61M$2.00$8.00
Palmyra X51M$0.60$6.00
Palmyra X4128k$2.50$10.00

Typical Use Cases

  • Auto-generating annual reports and financial analyses at banks and insurers
  • Medical record summarization and clinical-guideline lookup agents for healthcare
  • Contract review and case research at law firms
  • Brand-consistent multilingual content rollouts for global enterprises
  • Large-scale customer feedback analysis

Strengths

  • One of the few AI platforms that meets enterprise compliance requirements
  • Owns its LLM stack, avoiding dependency on third-party providers
  • Explicitly engineered around agent token cost, landing near $0.12 per task on average
  • 1M-token context available at a low per-token price
  • Also available through Amazon Bedrock

Weaknesses

  • Overkill and expensive for individuals or small teams
  • Custom quotes make it hard to evaluate casually
  • Smaller community and less public material than OpenAI/Anthropic
  • The finance and medical domain models are gone, so industry-specific tuning now has to come from your own prompts and knowledge bases
  • UI/UX skews toward power users

Official Resources

公式サイト

同カテゴリのツール