Red Hat Brings Governed Model Services to OpenShift AI 3.4

As companies move AI systems from pilots into production, models are increasingly shared across business units and applications, raising questions over access, inference traffic, spending and compliance reporting. Red Hat is positioning OpenShift AI as a centralized platform for exposing approved models through APIs, giving platform teams a common control plane and reducing the operational burden and fragmented governance that arise when individual groups deploy and manage their own model-serving stacks.

Red Hat unveiled OpenShift AI 3.4 on May 15, 2026, making Models-as-a-Service generally available with a built-in AI inference gateway. The release adds team-level token quotas, rate limits and self-service API keys that can be revoked immediately. External model routing, still in technology preview, uses an OpenAI-compatible endpoint to direct requests to locally hosted models or services including AWS Bedrock, Microsoft Azure OpenAI and Anthropic. Showback dashboards and enterprise authentication are also in preview.