Red Hat, the world’s leading provider of open source solutions, has announced significant advancements to its Red Hat AI portfolio with the launch of Red Hat AI 3.4, a comprehensive platform designed to unify AI builders and infrastructure operators on the agentic future.

By delivering a unified, metal-to-agent platform, Red Hat AI 3.4 aims to simplify the development and deployment of agentic workflows, helping organizations scale autonomous systems across the hybrid cloud while maintaining the operational control, security, and hardware efficiency required by the modern enterprise.

The new release addresses a primary hurdle to AI adoption: the friction between developers seeking innovation and infrastructure administrators requiring governance. Without a unified approach, infrastructure access barriers slow progress, and “shadow AI” shortcuts introduce ungoverned risks and unpredictable costs. Red Hat AI 3.4 resolves this tension by providing a foundation for scalable inference and autonomous agent deployments, ensuring the transparency and control needed to meet rigorous risk standards.

Central to Red Hat AI 3.4 is the introduction of Model-as-a-Service (MaaS), which provides a single, governed interface for developers to access curated models. MaaS enables administrators to enforce policies and track consumption, building on a foundation of high-performance distributed inference powered by vLLM and llm-d. This advanced inference capability ensures optimized and efficient model serving and includes request prioritization, allowing latency-sensitive requests to be processed first. Furthermore, newly generally available speculative decoding support improves response speeds by two to three times with minimal quality impact, helping to lower the cost per interaction.

AgentOps capabilities

To manage the exponential demand driven by autonomous systems, Red Hat AI 3.4 introduces comprehensive AgentOps capabilities. These tools manage the lifecycle of agents from development to production with integrated tracing, observability, and cryptographic identity. Because agents operate with independence, visibility into their decision-making is critical for security. The platform addresses this by integrating cryptographic identity management (using SPIFFE/SPIRE) to tie agent actions to a verified identity, creating an auditable trail of reasoning steps and tool calls.

Connecting enterprise data to these models and agents is made easier through new features like prompt management—which treats prompts as first-class data assets—and the evaluation hub. The evaluation hub is a framework-agnostic control plane that replaces fragmented testing methods with a unified approach for benchmarking quality, accuracy, safety, and risk. Automated safety testing and red-teaming are integrated directly into the development lifecycle, leveraging technology from Chatterbox Labs and the Garak project to provide a security-forward path from pilots to production-ready utility.

“The agentic era represents an evolution of our platform from running traditional applications to powering intelligent, autonomous systems,” said Joe Fernandes, vice president and general manager of the AI Business Unit at Red Hat. “We are defining the open standard for how the enterprise executes AI. By providing a hardened, metal-to-agent foundation for AI inference, MaaS and AgentOps, Red Hat provides the operational assurance organizations need to innovate at scale while maintaining rigorous control.”

John Fanelli, vice president of Enterprise Software at NVIDIA, echoed this sentiment, stating that autonomous agents “demand a new level of infrastructure control and security to ensure trustworthy operations at scale.” He noted that the Red Hat AI Factory with NVIDIA provides a unified, open-source foundation that gives developers and operators the necessary governance and confidence.

Red Hat AI 3.4 is expected to be available later this month.