Red Hat has released version 4.20 of OpenShift. The solution gets new AI tooling, post-quantum encryption, and more extensive virtualization options. It should help organizations modernize their applications and roll them out independently where necessary.
The new version is a prelude to support for post-quantum cryptography (PQC) algorithms. For the time being, these algorithms only secure traffic between control plane components. The measure is intended to protect against threats that will arise once quantum computers are able to break complex encryption. That threat is still theoretical and the algorithms to combat it are therefore unproven, but every IT vendor will have to support PQC algorithms in the coming years. If this is not enforced by law, it will be by the most demanding customers: waiting for the quantum revolution before adopting PQC could lead to a complete data breach. Any encryption algorithm that can be cracked in theory, even if it takes thousands of years on conventional computers, will likely be child’s play for quantum computers.
Streamlining AI workloads
Back to the present day. With OpenShift 4.20, Red Hat has added new capabilities to simplify large-scale AI workloads. The LeaderWorkerSet API automates orchestration and scales distributed AI models on demand. An image volume source feature allows teams to integrate new models without rebuilding containers.
OpenShift Lightspeed, the AI-driven virtual assistant in the console, now supports multiple clusters through integration with Red Hat Advanced Cluster Manager for Kubernetes. Administrators can search documentation, troubleshoot issues, and navigate system configurations in plain language. The assistant remains cloud-based, as before; Red Hat has previously indicated that it is not willing to expose its own IP by releasing the model in open-weight form.
Virtualization gets upgrades
OpenShift Virtualization now supports CPU load-aware rebalancing and Arm architecture. This improves performance and resource utilization for virtualized workloads. Red Hat has extended OpenShift Virtualization to bare-metal deployments on Oracle Cloud Infrastructure. Operators gain more control over workload placement and data sovereignty.
Improved storage offload functionality in the migration toolkit enables up to ten times faster migrations for organizations moving from legacy virtualization platforms.
Red Hat is also introducing a two-node OpenShift configuration with an arbiter node for high-availability edge environments such as retail and manufacturing locations. This compact configuration supports both x86 and Arm architectures. It includes built-in OpenShift Virtualization for virtualized workloads in offline environments.
Intelligent LLM routing
The vLLM Semantic Router in OpenShift 4.20 automatically selects the most appropriate model for each task to reduce costs and computing power. Lightweight classifiers determine the intent behind a query and send it to a fast or complex model depending on the requirement. This is actually what a single LLM sometimes does: models such as GPT-5 and DeepSeek-R1 activate only a portion of their parameters, or components, for an AI prompt, depending on the task. However, to reduce API costs or save computing power, organizations can deploy different AI models. That’s where OpenShift’s functionality comes into its own.
The router was developed in Rust on Hugging Face’s Candle framework, with low latency and high scalability for cloud-native environments. Integration with Kubernetes is done via Envoy, and companies can deploy the solution on OpenShift in hybrid clouds.
In combination with llm-d, the router supports features such as semantic caching and jailbreak detection. Benchmarks show 10 percent higher accuracy and up to 48 percent less latency and token consumption. The open source project now has more than 2,000 stars and 300 forks on GitHub.
Read also: Red Hat launches Sovereign Support for EU organizations