This month’s Oracle Cloud Infrastructure (OCI) AI updates focus on giving teams more ways to tailor AI systems to their specific workloads, users, and governance needs. New Model Import additions expand the range of options for reasoning, coding, transcription, healthcare, and agentic use cases, while private endpoints and guardrail updates help support security and control requirements. The July edition also highlights new learning and webinar resources designed to help practitioners build skills around agentic AI and emerging infrastructure patterns.
New updates and features:
OCI Enterprise AI adds GLM 5.2 and expands imported model choice
GLM 5.2 is now available on OCI Enterprise AI through Model Import, adding a new option for long-context reasoning, coding, and agentic workflows. OCI also added support for additional imported models including:
- OpenAI Whisper Large V3 Turbo
- Google MedGemma 27B Text IT
- Qwen3-Next-80B-A3B-Instruct
- DeepSeek V4 Flash and V4 Pro
- MiniMax M3, Mistral Medium 3.5
- Moonshot AI Kimi K2.6 and K2.7 Code
By giving customers more model options across reasoning, coding, transcription, healthcare, and agentic workloads, OCI helps teams choose the right model for each use case instead of defaulting to a one-size-fits-all approach, supporting better price-performance and more cost-efficient AI architectures.
Learn more about GLM 5.2 via Model Import and see a full list of compatible Model Import models here
Private Endpoints now available for imported models
OCI Enterprise AI now supports private endpoints for imported models, enabling customers to access imported model endpoints through a private network path without routing traffic over the public internet.
This helps support security and network-control requirements for custom and imported model deployments, enabling production AI workloads that require private connectivity, network isolation, and tighter control over data paths.
Learn more about Private Endpoints for Imported Models
OCI Enterprise AI Guardrails adds image moderation and version pinning
OCI Enterprise AI Guardrails now supports image moderation through the ApplyGuardrails API, helping identify unsafe content in standalone images and multimodal requests that include text and images. Customers can also now pin a specific guardrails version to maintain stable behavior for content moderation, prompt injection, and PII detection in production.
These updates help give customers more consistent controls when deploying AI applications that handle user-uploaded images, generated images, screenshots, or multimodal inputs, while making guardrail behavior more predictable across production releases.
Learn more here: Image moderation, and pinning a guardrail version
New resources for AI at OCI:
New Oracle Agentic AI Foundations training and certification
Oracle University has launched the Oracle Agentic AI Foundations course and certification to help learners understand how AI agents work and how to build them. As enterprises move from chat-based AI to goal-driven agents, this training gives developers, data scientists, and AI/ML engineers a structured path to build practical agentic AI skills and validate their knowledge.
Learn more about the certification
New webinar replay: Power production AI with OCI and NVIDIA RTX PRO
A new on-demand webinar, Power Production AI with OCI and NVIDIA RTX PRO, explores how OCI RTX PRO 6000 bare metal instances and NVIDIA AI Enterprise can support multimodal inference, simulation, and visualization on a unified high-performance platform. Customers can learn how to consolidate AI and visualization workloads, accelerate deployment with NVIDIA AI Enterprise and NIM, help improve GPU utilization, and build scalable production AI systems on OCI.
For more Oracle AI News: What’s New in Oracle AI? June 2026 Edition