Israeli AI systems developer AI21, founded by Prof. Amnon Shashua, Prof. Yoav Shaham and Ori Goshen, which operates in the crowded field of developing basic language models, has introduced Maestro - an AI Planning and Orchestration System designed to deliver trustworthy AI at scale for organizations.
Maestro is a new product and a new business model, which resembles a software-as-a-service (SaaS) company more than a model maker. Maestro is an AI-based software system that manages all of an organization's AI processes and "orchestrates" them, as the company puts it, in a way that will work harmoniously and reduce the rate of errors by re-analyzing data and processes. In other words, in addition to the Jamba model it is developing, AI21 will try to improve the way organizations work with models created by competing companies.
AI21 says Maestro improves the accuracy of systems like GPT-4o and Claude Sonnet 3.5 by up to 50% on complex tasks. Goshen says, "The problem in organizations is that the existing language and inference models are so far from accurate that only 6% of the projects created by companies in the field eventually see the light of day, the rest are shelved due to hallucinations or inconsistencies." Organizations try to address hallucinations by improving the formulation of queries or dividing the process into small sub-parts that can be inspected at different points, but the result, according to Goshen, is still far from a consistent result.
"In a world where you can't trust the output you get from AI, Maestro offers you a basket of language models, AI tools, data sources and application programming interfaces (APIs), on the basis of which it will bring the data, and the system will bring the various models to produce an answer and say what its level of reliability is."
Instead of using one or two models, companies can manage all their AI application development from one place, a product that reflects the target customers that AI21 is aiming for - not the end consumer or the CFO in the organization, but the AI managers in organizations, known as AI Builders, software developers, data experts, machine learning or information systems experts who are responsible for centralizing all AI processes in the organization - such as developing automated agents, chatbots and training internal models.
Goshen explains, "We do not yet know which business model will dominate the AI industry, but our belief is that the 'AI builders' in the organization will play a significant role there. Instead of each department raising its own automated agent that will access all relevant information with other computing resources, there will be a single information systems authority that will manage this resource, such as 'agent human resource management'. However, today the IT department in an organization accounts for 10% or 15% of the organization's expenses, whereas in the era of agents, this expense can grow to tens of percent of the organization's total expenses."
Last week, AI21 introduced Jamba 1.6, its most advanced open model for private enterprise deployment. The new model outperforms leading open models from Mistral, Meta, and Cohere across multiple benchmarks while maintaining superior speed, accuracy, and security. The model enhances data classification by 26 percentage points over its predecessor, Jamba 1.5, and ensures over 90% consistency in cited responses for long-context question answering. With flexible deployment options, including VPC and on-premise installations (self hosted), Jamba 1.6 offers enterprises full control over their data. By combining the reasoning power of Transformers with the efficiency of State Space Models (SSMs), Jamba 1.6 optimizes long-context processing, delivering advanced enterprise AI performance.
Published by Globes, Israel business news - en.globes.co.il - on March 11, 2025.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2025.