Palmyra X5

Palmyra X5 Writer’s newest enterprise‑grade language model, Palmyra X5 is purpose‑built for large‑scale, multi‑agent systems. Its hybrid transformer architecture delivers a 1  million‑token context window, sub‑second tool‑calling, and top-ranked price‑performance for a commercial long‑context model.

Details

Text, image input; text, structured output
1M context length

Availability

Agent Builder (beta)
No-code
Writer Framework
API
Amazon Bedrock

Price

Input: $0.60 / 1M tokens
Output: $6.00 / 1M tokens

Use cases & capabilities

Agents & actions
Palmyra X5 excels at orchestrating multi‑step workflows by invoking any number of external tools, databases, and Writer domain models in a single call.

Retrieval-augmented generation (RAG)
Built‑in RAG and Knowledge Graph connectors ground every response in current, authoritative data.

Code generation 
Generates production‑ready code snippets or full scripts to automate business logic.

1 M‑token context
Load entire playbooks, regulatory filings, or product catalogs at once—no brittle chunking required.

Benchmarking

Palmyra X5 can process a full million-token prompt in ~22 seconds and fire off multi-turn function-calls in ~300 milliseconds, while costing 3–4× less per token than GPT-4.1. Those two numbers—speed and price—unlock agent behaviors that were previously cost- or time-prohibitive.

On OpenAI’s MRCR 8-needle test—a long-context benchmark that hides eight identical requests in a massive conversation and challenges the model to find the correct one—Palmyra X5 scores 19.1%, compared to 20.25% for GPT-4.1 and 17.63% for GPT-4o. With near top-tier retrieval performance at a dramatically lower cost, Palmyra X5 gives enterprises the ideal balance for scaling RAG pipelines and production agents without breaking budgets.

Palmyra X5 is also one of the top ranked models on the BigCodeBench (Full, Instruct) evaluation with a score of 48.7, showcasing it’s ability tackle practical and challenging programming tasks.

Benchmark highlights:

BBH (Big-Bench Hard): Evaluates complex reasoning and compositional logic. Palmyra X5 achieves a competitive score of 70.99%, aligning closely with top-tier models.
GPQA (Graduate-Level Google-Proof Q&A): Assesses the model’s ability to answer challenging, graduate-level questions in biology, physics, and chemistry that are resistant to simple lookup strategies. X5’s score of 47.20% indicates strong performance in scientific reasoning tasks.
MMLU_PRO: Focuses on professional-level knowledge across various domains such as law, medicine, and finance. Palmyra X5 scores 65.02%, demonstrating its suitability for enterprise applications in regulated sectors.
MATH_HARD: Tests symbolic reasoning and multi-step problem-solving abilities. X5’s score of 71.57% showcases its proficiency in handling complex analytical tasks.

Use cases

For enterprise engineers building AI agents in Agent Builder (Beta), X5 unlocks a variety of long-context use cases like:

Revenue & reporting: Ingest full RFPs, pull from Salesforce, and draft first responses automatically. Or generate fund reports by joining third-party market data with internal research—all in a single agent flow.
Support & knowledge management: Classify tickets, stage CMS updates, and publish content with review workflows baked in. Agents also keep knowledge bases fresh by flagging outdated content and suggesting revisions.
Regulatory & compliance intelligence: Analyze lengthy contracts, 10-Ks, or EHRs in one pass. Extract key clauses, identify risks, and summarize with citations—ideal for finance, healthcare, and legal teams.
Customer & research insights: Summarize thousands of survey responses or research papers. Surface themes, extract insights, and accelerate product or R&D decisions with minimal human input.

Our Agent Library also includes a growing number of X5-powered pre-built agents — large file summary, regulatory document analysis, healthcare thought leadership deliverables, medical record summary, and more. Each agent inherits Palmyra X5’s 1M-token context and sub-second function-calling, so they can digest hundreds of pages and take advantage of X5’s multi-modal inputs to deliver results without any building required.