A new System One endpoint serves six decision-model sizes for typed text, JSON, image and video questions, but the code is still nightly-only.

The Q3 Security Signals report treats vulnerabilities, configurations, credentials and sensitive resources as one attack path—shifting priority from isolated scores to reachable outcomes.

A working Muse Spark Open stack shows what operators can use now; broader day-zero model coverage and performance publication remain commitments.

The new recipe preserves the September model’s deployment shape across AMD and NVIDIA hardware, but its published validation still belongs to the older weights.

Three new recipes turn model support into reproducible deployment contracts across AMD text inference, single-GPU speech and NVIDIA image generation.

A Red Hat engineering walkthrough connects policy gates, an SRE agent and human approval into a repeatable failed-build response pattern.

A day-zero image packages text-to-image and multi-reference editing behind OpenAI-compatible endpoints, but the checkpoint is gated and the benchmark is one GB200 run.

The verified recipe separates model-weight fit from KV-cache capacity, with distinct tensor-, expert- and data-parallel choices for B300 and H200 systems.

The new LTS line adds preview semantic routing, fail-closed policy components and a broader MCP and observability surface for agentic integrations.

The result is a software-stack comparison, not a clean GPU shootout, but the engineering behind it offers a useful optimization map.

Red Hat’s architecture guide explains why higher VM density depends on coordinated reclaim, tiering, pressure signals and cluster rebalancing—not one overcommit ratio.

New Red Hat AI checkpoints split the performance problem between lower-precision execution and parallel speculative decoding on NVIDIA Hopper GPUs.

A directional five-year model argues that tiering enterprise AI traffic across open-weight and frontier models can cut TCO, but the result depends heavily on workload mix and utilization.

The Q3 report combines Lightwell package observations with an explicitly modeled Java scenario, estimating how shared remediation could avoid duplicated work.

The design separates the delivery of signed fixes for pinned dependencies from the approvals, testing, rollback and audit work needed to deploy them.

The quarterly engineering update emphasizes safer CI, more predictable device-state handling and fixes for large configurations and connection timeouts.

A new engineering comparison says benchmark parity makes traffic shape, hardware support and ecosystem fit the deciding factors.

A 52× HLS speedup for about $26 shows why cache-aware serving may matter as much as model choice in automated design exploration.

A two-shard tech-preview configuration sharply reduced slow applies during event storms, but added memory cost and regressed under one quota-pressure workload.

A new worked example uses Camel’s AI Tool component and built-in MCP server to expose three narrowly defined CRM operations without handing the model arbitrary SQL access.

A new IBM Research explainer connects llm-d’s router, cache tiers and split worker pools to the uneven traffic generated by AI agents.

The field report connects OpenShift failure injection to release gates, virtualization migrations, database recovery and payment SLOs.

Decision 3.0 ships six Apache-2.0 models that answer choice, yes/no and scoring questions with probabilities instead of generated prose.

A tested OpenShift walkthrough shows how Maximo Manage can expose an existing workflow to an AI agent, while surfacing sharp support and response-design caveats.

The upstream project has daily CUDA 13.4 images, Rubin-tuned kernels and preliminary results showing as much as 7.84× the per-GPU AgentX throughput of GB200.

X41’s proof of concept is not a universal Envoy compromise, but it gives platform teams a concrete test for whether co-located containers can cross an assumed trust boundary.

A principal developer’s workflow analysis finds the clearest gains in bounded, reviewable work—and warns that unverified debugging can erase them.

The new technical report moves beyond text decoding with a shared control plane for speech, image, video, world-model and robot workloads.

The engineering guide connects device constraints to Podman, MicroShift and OpenShift tiers, then follows the operational consequences.

SWA bounded replay, CUDA graphs and model-specific kernel fusion reshape the latency-throughput trade-off for long-horizon agent serving.

Host-side authentication keeps the API key out of model context, while automatic recall still sends selected memories into that context and introduces deployment assumptions operators must own.

The internal OpenShift application separates sessions, credentials and durable state while keeping people in control of research and proposal decisions.

The production system combines classifiers, frontier models, retrieval and human review—but Red Hat’s account does not publish performance or accuracy numbers.

The pattern joins A2A discovery, Rossoctl’s Kubernetes resources, SPIRE-based identity and MLflow traces without pretending every agent deployment needs the full stack.

The new Hub operator establishes a multicluster deployment path, but Red Hat has not yet published the topology, version-skew matrix or installation details in the release erratum.

The engineering follow-up names Nemotron 3 Ultra and Poolside Laguna S 2.1, and says prompt, sampling and harness changes shaped the final selection.

Red Hat’s monthly digest highlights failed 4.4 release candidates alongside proposals for quota visibility, TLS groups and safer MirrorMaker migrations.

The Red Hat Ecosystem Catalog additions cover SQL Server availability for RAG, phased public-sector modernization and application-scoped data access.

A Red Hat–PyTorch engineering project reports double-digit throughput gains in some Hopper workloads, but the implementation remains in a fork while maintainers weigh upstream tradeoffs.

The Red Hat AI Americas field build combines semantic recall with shipped authorization controls, while its own documentation leaves audit, policy, operator and observability work incomplete or internally inconsistent.

The new reports surface affected connected systems and package-roadmap changes, but they remain an opt-in summary layer rather than an upgrade planner or policy engine.

A new reference implementation makes specialist selection, concurrent calls, validation and bounded retries visible as ordinary integration routes.

The first-party guidance turns component lists into a lifecycle checklist for maintenance, vulnerability response, remediation and replacement planning.

Nine guardrail configurations show why accuracy, latency and infrastructure cost still favor task-specific classifiers for common production checks.

A first-party migration toolkit uses Ansible and changed-block tracking, while certified partner Hystax targets agentless moves at larger scale.

The Office of the CTO says agents now help rank research bets, accelerate prototypes and package context for centralized software-delivery pipelines.

The release changes the application boundary: clients can address durable conversations while the gateway records and executes nested agent work as Responses items.

A new Red Hat engineering guide separates malicious model artifacts from learned behavior and lays out a staged intake and runtime defense pipeline.

The chaos-engineering project now measures weighted SLO failures, endpoint downtime and KubeVirt VM connectivity instead of relying on a binary pass/fail result.

Pipeline Failure Analyzer uses deterministic code for log cleaning, reports and tickets, reserving model reasoning for cross-platform root-cause grouping.