Palmyra is the family of large language models developed by Writer, Inc. and used by its WRITER platform. Writer meters most of its platform and API usage in tokens, pages and gigabytes rather than in seats, and publishes the rates on the developer pricing page.[1] For a licence or cost manager the practical questions are which model a workload uses, which rate applies in which surface (API, Agent Builder, no-code or WRITER Agent), when a model is withdrawn, and whether a model is covered by the Enterprise agreement or by a separate licence. This article answers them from Writer’s published developer documentation, retrieved on 2026-10-13.
Palmyra models
The models page compares three current Palmyra models.[2]
| Feature | Palmyra X6 | Palmyra X5 | Palmyra X4 |
|---|---|---|---|
| API model ID | palmyra-x6 | palmyra-x5 | palmyra-x4 |
| Description | Agentic model for WRITER Agent and long-horizon enterprise workflows | General-purpose model with adaptive reasoning | Model for complex tasks that need extended context |
| Context window | 1M tokens | 1M tokens | 128k tokens |
| Max output | 8,192 tokens | 8,192 tokens | 4,096 tokens |
| Input per 1M tokens | 2.00 USD | 0.60 USD | 2.50 USD |
| Output per 1M tokens | 8.00 USD | 6.00 USD | 10.00 USD |
Palmyra X6 is the default model for WRITER Agent.[2] Palmyra X5 and X4 are listed with Amazon Bedrock model identifiers, and a separate integration guide describes invoking them through Bedrock; X6 has no Bedrock identifier in the comparison.[2][7] The models page also states that to analyse images, customers use the vision capabilities of Palmyra X5, and the pricing page marks the earlier Palmyra Vision model as deprecated.[1][2]
The rate depends on the surface
The base table rates apply to API, Agent Builder and no-code usage. WRITER Agent uses Palmyra X6 as its default agent model and bills at the X6 rate unless a different model is chosen. Palmyra X5 is the exception: inside WRITER Agent it is billed at 5.00 USD input and 12.00 USD output per 1M tokens, against 0.60 and 6.00 in the base table.[1] A workload moved from the API to WRITER Agent on the same model can therefore change cost, and the Credit Terms confirm that usage is measured and billed at the rates published on the pricing page.[10]
Deprecation and lifecycle
Writer’s pricing and models pages carry a notice that Palmyra X4 (palmyra-x4) and Palmyra X5 (palmyra-x5) will be deprecated on December 14, 2026, with Palmyra X6 as the migration path; X6 has a 1M token context window against 128k for X4.[1][2] The deprecation policy says that:[2]
- the deprecation of a model is announced at least three months in advance;
- deprecated models continue to be supported for a period afterwards;
- the end of support for a deprecated model is announced at least six months in advance, on a timeline that varies by model.
The deprecated models list on the same page gives these dates: Palmyra X5 and X4 on 2026-12-14; Palmyra X 003 Instruct, Palmyra Vision, Palmyra Med, Palmyra Fin and Palmyra Creative on 2026-07-13; Palmyra Fin 32k on 2025-03-03; and Palmyra X 002 32k, X 32k Instruct and X 002 Instruct on 2024-09-06.[2] The pricing page keeps their last published rates, for example 5.00 USD input and 12.00 USD output for Palmyra Med, Fin and Creative.[1] For a licence position this means that a contract or internal chargeback that names a specific Palmyra model needs a re-baseline at each deprecation date.
WRITER-managed third-party models
WRITER Agent can also run third-party models. The pricing page lists a catalogue that WRITER hosts and bills, marked “Rates are subject to change. Last updated: 9/17/26”. It includes models from Anthropic (for example Claude Opus 4.5, 4.7 and 5 at 5.00 USD input and 25.00 USD output per 1M tokens, and Claude Sonnet 5 at 2.00 and 10.00), Google, DeepSeek, Moonshot, MiniMax, OpenAI and Z.ai, with image and video models priced per token or, for Veo 3.1, per second at 0.40 USD.[1] These models are WRITER-managed and available in WRITER Agent only; the external models guide adds that they are not yet available through the API or SDK.[1][3]
In WRITER Agent, Palmyra and WRITER-managed third-party models are metered per token through WRITER, with 1 credit equal to 1 USD, at the rate shown for each model, and usage appears on the WRITER dashboard.[1] Administrators control access: only WRITER models are enabled by default, with Palmyra X6 as the default model, and third-party models stay disabled until an IT admin enables them. The page also lets an admin enable all models from a provider and set an “Always enable new models” option, which turns on newly added models for all teams automatically.[8][3] A licence manager should note that this option changes cost exposure without a new order.
Bring your own model
Enterprise customers can instead connect external models with their own provider credentials. Writer calls this bring your own model (BYOM) and supports AWS Bedrock and OpenRouter as providers, for text generation and embedding models; video, audio and image generation models are not supported through BYOM.[3] The provider bills its own usage directly, and WRITER adds a 10% pass-through fee on top of the provider rate card, paid via credits.[1][3] BYOM is available on Enterprise plans across WRITER Agent and the API, and usage is visible both on the WRITER dashboard and in the provider’s billing.[1]
Knowledge Graph and Tools API rates
The pricing page lists retrieval and document processing separately from model tokens.[1]
| Capability | Published cost |
|---|---|
| Knowledge Graph hosting | 0.085 USD per GB of storage per day |
| Data extraction | 0.00015 USD per page |
| File parsing (OCR) | 0.055 USD per page |
| Web access | 0.12 USD per page |
| Tools API PDF parser | 0.055 USD per page |
The same page states that data connectors for Knowledge Graph are available for Enterprise plans and that connector usage and Knowledge Graph are priced under their own sections, not within the model token rates.[1] These rates make storage and ingestion volume a cost driver independent of the number of users.
Rate limits and API terms
The API reference states default limits of 400 requests per minute and 25,000 tokens per minute, and offers custom limits through Sales.[4] The pricing page directs customers needing custom pricing to Sales.[1] The best-practice guidance tells clients to prepare for HTTP 429 (too many requests) responses and to retry with exponential backoff.[4]
The API usage policy states that the API is provided for commercial use, that users are responsible for the security of their API keys and any data obtained, and that users may not sell, resell or lease the API or its data to third parties; non-compliance may result in suspension or termination.[5] The policy can be revised at any time without notice and users are responsible for checking it.[5]
Tracking usage
AI Studio provides a Consumption view for organisation admins. The cost view shows monthly usage-based spend, spend by agent type (API, Framework and No-Code), and a breakdown by model and by other services such as Knowledge Graph hosting, web access and OCR. The activity view shows token usage, the top users and agents and usage by model.[9]
Palmyra open models and the Writer Open Model License
Writer also distributes model weights and documentation, which it calls the Writer Materials, under a separate Writer Open Model License, version released May 31, 2024. It grants a non-exclusive, worldwide, non-transferable, royalty-free limited licence to use, reproduce, distribute, copy, modify and create derivative works of the Writer Materials, but only for non-Commercial purposes.[6] The agreement defines “Commercial” as primarily intended for or directed towards commercial advantage or monetary compensation, and says that for any Commercial use a licence must be requested from Writer, which it may grant at its sole discretion.[6] A recipient who gets the materials as part of an integrated end-user product from a direct or indirect licensee is not bound by the non-commercial limit.[6]
Redistribution requires providing a copy of the licence to the recipient, carrying prominent notices on modified files, retaining a NOTICE file with the Writer attribution and complying with the Acceptable Use Policy and trade laws. The licensee owns its modifications, subject to Writer’s ownership of the materials, but the licence terminates if the licensee brings litigation or other proceedings against Writer alleging that the Writer Materials infringe its intellectual property.[6] The licence is governed by California law with exclusive jurisdiction in California courts.[6] Commentary: because the licence is limited to non-Commercial purposes, it does not allow every use that open source licences allow, and use of these weights inside a revenue-generating product needs a separate agreement with Writer.
Out of scope
This article does not list every third-party model rate, which Writer updates on the pricing page. It does not cover the terms of the underlying model providers.