Writer releases Palmyra X5, delivers near GPT-4.1 performance at 75% lower cost

WriterEnterprise generative AI company worth $ 1.9 billion, was released today Palmyra X5, In corporate environments, a new large language model (LLM), which provides an extensive 1 million-token context window that promises to accelerate the admission of autonomous AI agents.

San Francisco-based company counts that Mamur, Froggy, Uberand Avant-garde Hundreds of enterprises have placed the model between customers as an effective alternative to offers from industrial giants Open and AnthropicalIn a million input versions at $ 0.60 and a $ 3 dollar with $ 3.

“This model is really unlocked by the agent of the world,” said Matan-Paul Shetrit, in an interview with WentureBeat, the director of the writer. “GPT-4.1 is faster and cheaper than the equivalent large context window models, and it allows you to start work as a really step-streaming agent when you combine it with the ability to make a large context window and model.”

AI Economy Jump: How the writer taught a power plant model for $ 1 million

Unlike many opponents, the writer trained Palmyra X5 The GPU is part of a part of the requiring about $ 1 million in synthetic information – other leading models. This represents cost efficiency, or a significant separation of millions of millions of millions of millions million million million years of spending industry.

“Our beliefs are generally cheaper and cheaper than tokens and the calculation is cheaper and cheaper,” said Shetrit said. “We are here to solve real problems than to assess nickel and customers.”

The cost of the company’s cost is due to special techniques developed in several years. In 2023, the study was published among the writerTo induce oneself“Tuning Tuning, which offers early stopping criteria for minimum instructions.

“Unlike other basic stores, our landscape must be effective. We must be effective here,” he said. “We must provide the fastest, most cheapest models to our customers, because ROI is really important in these cases.”

Million-Token Marvel: Technical architecture that increases the speed and accuracy of Palmyra X5

Palmyra X5 can quickly operate in full millions of time in about 22 seconds and implement multi-turn functional calls in 300 millisions – the writer claims the performance measurements “agent behavior that banned previously spent or time.”

The architecture of the model includes two main technical updates: Hybrid focus mechanism and a A mixture of experts the approach. “Hybrid focus mechanism … Provides a focusing mechanism that allows the appropriate parts of entrances while creating each exit within the model,” he said. This approach accelerates the offspring of response while maintaining accuracy along the extensive context window.

In assessment tests, Palmyra X5 has achieved noticeable results relative to its price. In Openai MRCR 8-needle test – Models to find eight identified in a mass conversation – Palmyra X5 hit 19.1%, 7.1% compared to GPT-4O, GPT-4O compared to GPT-4.1 and 17.63%. It also places in the eighth coding Bigcodebench benchmark With a score of 48.7.

Demonstrates these criteria

From Chatbots to business automation: How does the enterprise of AI agents change workflows

The release of Palmyra X5 comes shortly after The writer presented AI HQ At the beginning of this month – a centralized platform for enterprises to set up, place and control AI agents. This dual product strategy, the growing company’s growing enterprise requirement is to capitalize the requirement of the requirement that can autonomy the complex business processes

“Models that offer fewer than 1 million verses of agents, will not work quickly for business-critical use,” said writer CTO and co-founder Vehicle Alshikh.

At this point: “It has been a great gap between the AI ​​agents and what they will do in fact. However, we see real-world agents with customers of large enterprises.

Early adopts, financial statements, RFP answers, support documents, including supporting documents and customer feedback, placing Palmyra X5 for work flows.

In case of compulsory use, it covers the most important-step work flows, which automatically pushes off updates and automatically confirming the outdated content of an AI agent, and automatically confirmed updates.

From a simple text generation to automate this sliding process, it represents a fundamental evolution of businesses in the establishment of the EU – the human work is moving to automate all business functions.

Cloud Expansion Strategy: AWS Partnership, writer brings AI to millions of enterprise developers

In addition to the model release, the writer announced both Palmyra X5 and their predecessor, Palmyra X4now available Amazon BedrokAmazon Web Services’ Fully managed service to access basic models. AWS becomes the first cloud provider to present fully managed models from the writer who significantly expanding the company’s potential.

“Writer Palmyra will allow seamless access to seamless access to developers and scalting and scaling and scaling and scaling AWS, the director of Amazon Bedrock in Amazon Bedrok

AWS integration is a critical obstacle to the adoption of AI: Technical complexity of the placement and management of models on a scale. By preparing a Palmyra X5, which is available through the Simplified API of Bedrock, the writer may have millions of developments that do not have specialized experience to work with direct foundation models.

Self-Learning AI: Writer vision for developing models without human intervention

The writer is a bold claim for context windows, announced that all future models released by 1 million verses will be minimal. This obligation reflects the importance of the company that is important for enterprise-grade AI agents that interact with large number of systems and data sources.

Looking upon the opposite, the applications that make the same developing models, the enterprise has set the next major progress in the AI. “The reality is today, agents do not need to be at the level we want and do not.” “I think that the users are coming to AI HQ, they start to map this process … and then, or afterwards, or in it, or models that are developing yourself in your company.”

This self-development opportunities will change how the AI ​​systems improve in time. Models will study continuously, and improve their performance for use for a certain enterprise, rather than the periodic re-preparation or delicate arrangement by AI specialists.

“This idea is not real” not real, “said Shetrit,” Shetrit, “said the various needs of various business groups. “Even two different products teams, many different ways of work, PMS are themselves.”

Enterprise New Math of EU: How to Promote 1.9b of the writer $ 1.9b strategy and anthropic

The writer’s approach involves dramatically Open and AnthropicalThose who have raised billions of financing, but focusing on the development of a general purpose. The writer focused on the establishment of special models of enterprises with expenditures that provide a total distribution.

This strategy has attracted a significant investor interest in the company Increase $ 200 million in financing $ 1.9 billion on the last November. Was carried jointly by the round To invest in rewards, Radical businessesand Iconiq growthincluding the participation of strategic investors Salesforce Ventues, Adobe Venturesand IBM enterprises.

According to Forbes, pointing to the writer 160% net grip rateShows that customers usually expand their contracts after initial adoption. This year, the company will be related to the number of contracts and projects worth more than $ 50 million this year this year.

For enterprises that assess generative AI investments, writer Palmyra X5 A flexible value offers the offer: strong opportunities in one part of competitive solutions. As the EI Agent Ecosystem is adultered, the company’s effective, enterprise-oriented models are betting on better financed opponents in accordance with Business ROI requirements.

“Our goal is to accept a widespread agent as soon as possible in our customer base,” he said. “Economy is true, if our enterprises are very high, it will not be able to compare the value of the EU agent against a human worker and the only way to place a large level in large enterprises.”

Often technical capabilities and theoretical performance ceilings can prove the pragmatic attention of writers, in a sector in a sector focused on cost efficiency, as a result of the other decimal point of benchmark. As developed in measuring the EU’s work effect on the EU, the question “How strong is your model?” “How convenient is your intelligence?” – And the writer bets that the economy will only determine not opportunities, but the EU company’s owners.