Photo credit: venturebeat.com
Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More
Writer, an enterprise generative AI company valued at $1.9 billion, has announced the launch of Palmyra X5, a new large language model (LLM) featuring a remarkable 1-million-token context window. This innovation is expected to significantly enhance the integration of autonomous AI agents in business settings.
Based in San Francisco, Writer services a diverse range of enterprise clients, including Accenture, Marriott, Uber, and Vanguard. The company positions its model as a cost-effective alternative to those from major players like OpenAI and Anthropic, offering pricing at $0.60 per million input tokens and $6 per million output tokens.
“This model revolutionizes the agentic landscape,” stated Matan-Paul Shetrit, Writer’s Director of Product, during a discussion with VentureBeat. “It offers speed and affordability that surpasses existing models with large context windows, enabling the execution of complex multi-step functions.”
A comparative analysis indicates that Writer’s Palmyra X5 achieves approximately 20% accuracy on OpenAI’s MRCR benchmark, priced at about $0.60 per million tokens. This positions it advantageously against models like GPT-4.1 and GPT-4o, both costing over $2.00 per million tokens.
AI economics breakthrough: How Writer trained a powerhouse model for just $1 million
In contrast to its competitors, Writer has trained Palmyra X5 using synthetic data with GPU costs amounting to about $1 million, a stark reduction compared to the typical tens of millions spent on model development in the industry.
“We believe that the cost of tokens and computing is continually decreasing,” explained Shetrit. “Our focus is on addressing practical challenges rather than burdening clients with excessive charges.”
Writer’s cost advantage arises from proprietary methodologies honed over several years. In 2023, the company released findings on “becoming self-instruct,” introducing early stopping criteria that lead to reduced instruct tuning costs, which Shetrit claims significantly lowers expenses during training.
“Unlike many foundational AI firms, our proven approach is to prioritize effectiveness and efficiency,” Shetrit mentioned, emphasizing the importance of delivering quick and economical models for clients.
Million-token marvel: The technical architecture powering Palmyra X5’s speed and accuracy
Palmyra X5 processes an entire million-token prompt in about 22 seconds, and it can perform multi-turn function calls in roughly 300 milliseconds. Writer asserts that these performance benchmarks allow for “agent behaviors that were previously not feasible due to cost or time constraints.”
The model is built on two pivotal technical features: a hybrid attention mechanism and a mixture of experts technique. Shetrit elaborated on the hybrid attention mechanism, which enables the model to selectively concentrate on pertinent input segments when generating outputs, thus speeding up response times while maintaining accuracy.
On benchmark evaluations, Palmyra X5 produced commendable results for its cost. In the MRCR 8-needle test from OpenAI—which tests models on their ability to identify eight identical requests within large conversations—Palmyra X5 scored 19.1%. This is nearly in line with GPT-4.1’s score of 20.25% and surpasses GPT-4o’s score of 17.63%. Additionally, it ranked eighth in coding on the BigCodeBench benchmark with a score of 48.7.
These results imply that while Palmyra X5 might not dominate every benchmark metric, it offers near-flagship functionality at a much lower price point—an appealing trade-off for enterprise clients focused on return on investment.
From chatbots to business automation: How AI agents are transforming enterprise workflows
The launch of Palmyra X5 aligns with Writer’s recent introduction of AI HQ, a centralized platform for businesses to develop, deploy, and oversee AI agents. This comprehensive strategy positions Writer to meet the increasing demand for AI that can autonomously manage complex business operations.
Shetrit expanded on this notion, emphasizing that there has historically been a significant gap between the anticipated benefits of AI agents and their actual performance. “At Writer, we are now witnessing genuine implementations with major enterprise clients,” he stated, insisting that these clients address real-world challenges rather than hypothetical scenarios.
Organizations are implementing Palmyra X5 for diverse tasks, encompassing financial reporting, responses to requests for proposals (RFPs), development of support documents, and analysis of customer feedback.
A notable application involves multi-step workflows where an AI agent identifies outdated information, generates recommendations, seeks human approval, and automatically updates a content management system with the approved changes.
This evolution from basic text generation to comprehensive process automation marks a significant transformation in AI’s role within enterprises—shifting the focus from assisting human tasks to fully automating business functions.
Cloud expansion strategy: AWS partnership brings Writer’s AI to millions of enterprise developers
In tandem with the model’s release, Writer announced that both Palmyra X5 and its earlier version, Palmyra X4, can now be accessed through Amazon Bedrock, Amazon Web Services’ fully managed service for foundation models. This partnership marks AWS as the first cloud provider to offer Writer’s managed models, greatly broadening the company’s market reach.
“Easy access to Writer’s Palmyra X5 will empower developers and organizations to create and scale AI agents, revolutionizing their handling of extensive enterprise data—leveraging AWS’s security, scalability, and performance,” stated Atul Deo, Director of Amazon Bedrock at AWS.
This integration addresses a significant hurdle for organizations aiming to adopt AI: the technical challenges of deploying and managing large models. By providing Palmyra X5 through Bedrock’s simplified API, Writer could reach millions of developers who may lack the technical expertise to work directly with foundation models.
Self-learning AI: Writer’s vision for models that improve without human intervention
Writer has boldly announced that a 1-million-token context window will be the standard for all future models. This commitment indicates the company’s belief that large context functionality is critical for AI agents operating across multiple systems and data sources.
Looking to the future, Shetrit highlighted self-evolving models as the next significant leap in enterprise AI. “Currently, agents fall short of the performance levels we aspire to achieve,” he observed. He envisions a scenario where users interact with AI HQ to map processes, layered with self-evolving models that adapt based on corporate workflows.
The envisioned self-evolving features could radically transform how AI systems enhance their performance. Instead of requiring frequent retraining or tuning by specialists, these systems would continuously learn from their engagements, progressively increasing their effectiveness in specific organizational contexts.
“The concept of a singular agent handling all tasks is impractical,” Shetrit noted, acknowledging the diverse operational requirements of various teams within a business.
Enterprise AI’s new math: How Writer’s $1.9B strategy challenges OpenAI and Anthropic
Writer’s strategy diverges markedly from the approaches of OpenAI and Anthropic, which have amassed billions for general-purpose AI innovations. Writer focuses specifically on crafting enterprise-targeted models with cost structures that facilitate broad usage.
This focus on enterprise has piqued substantial investor interest, as evidenced by the company’s raising of $200 million in Series C funding last November, leading to its $1.9 billion valuation. This funding round was co-led by Premji Invest, Radical Ventures, and ICONIQ Growth, with strategic participation from Salesforce Ventures, Adobe Ventures, and IBM Ventures.
Forbes reports that Writer has an impressive 160% net retention rate, suggesting that clients generally increase their contracts by 60% following their initial engagement. The company boasts more than $50 million in contracts secured, with projections indicating this could rise to $100 million in the current year.
For businesses assessing generative AI investments, Writer’s Palmyra X5 offers a compelling argument: robust capabilities at a considerably lower price compared to its competitors. As the ecosystem of AI agents evolves, the company’s focus on economically viable, enterprise-centric models may carve out a strategic edge over competitors with more substantial funding but less alignment to corporate ROI needs.
“Our objective is to promote widespread adoption of AI agents among our customers as rapidly as possible,” Shetrit highlighted. “The economic rationale is clear—if we set our pricing too high, companies will simply weigh the cost of an AI agent against human labor and may not perceive enough value. To foster adoption, we must deliver superior speed alongside significantly lower costs. That’s essential for achieving large-scale deployment of agents in major enterprises.”
In an industry often consumed by discussions of technical prowess and theoretical limits, Writer’s grounded focus on cost efficiency may ultimately forge transformative impacts compared to minor enhancements in benchmarks. As enterprises become more adept at evaluating AI’s influence on business outcomes, the priority may pivot from “How powerful is your model?” to “How affordable is your intelligence?”—with Writer betting its future on the premise that economic factors will dictate the champions of the enterprise AI landscape.
Source
venturebeat.com