Oracle AI World brought a busy October for AI at OCI, with announcements spanning our expanded OpenAI collaboration, Oracle AI Media Models in preview, and new ways to build and scale agentic applications. In November’s roundup, explore these highlights alongside the latest OCI Enterprise AI updates, our recognition as a Leader in the IDC MarketScape for Agentic AI Platforms, and on demand AI World replays for sessions that you may have missed.
New updates and features
Oracle and OpenAI bring frontier AI into everyday business
Oracle announced an expanded collaboration with OpenAI, connecting Oracle’s enterprise technology with ChatGPT, Codex, and advanced AI models through four offerings:
- Oracle AI Workspace brings supported business applications, enterprise data, and cloud services into one experience through ChatGPT. Employees can find information, coordinate workflows, and take approved actions across connected systems while retaining existing permissions and business controls.
- Oracle AI Security Mesh combines OpenAI cyber models with Oracle’s testing tools to discover vulnerabilities, validate findings, prioritize remediation, and verify fixes. Customers control testing scope and approve remediation changes.
- Oracle App Composer helps employees build applications with Codex and publish them on OCI. Oracle manages deployment and operations, while IT oversees access, approved data connections, policies, and costs.
- Oracle AI SMB Bundle helps eligible U.S. small and medium businesses build products and run their operations with AI tools, conditional ChatGPT credits, and free tiers for NetSuite QuickStart and App Composer. Approved businesses can receive US$12,000 in annual OCI credits, renewable for up to three years total, subject to eligibility and program terms.
Together, these offerings help customers connect AI to everyday work, turn ideas into deployed applications, and strengthen security while maintaining enterprise control. Smaller businesses also gain a lower-cost starting point for developing products and growing on OCI; availability and usage limits vary by offering.
Explore the Oracle and OpenAI offerings.
Oracle AI Media Models bring generative media capabilities to OCI in preview
Oracle AI foundation models for media and entertainment bring generative AI capabilities into content creation and production workflows on Oracle Cloud Infrastructure (OCI)
Explore Oracle AI Media Models and request access, and read the announcement.
Oracle Cost Advisor for AI brings visibility to AI spending
Oracle Cost Advisor for AI is available in limited availability, consolidating usage and spending data from sources including OCI Enterprise AI, OpenAI, and Anthropic. Initially delivered through Oracle Fusion Cloud EPM, it connects AI costs to budgets, owners, and cost centers and recommends opportunities to improve efficiency.
Finance, engineering, and AI leaders can identify spending trends, investigate anomalies, and evaluate lower-cost model alternatives. This shared view helps teams make informed decisions about where to invest and where to optimize as AI adoption grows.
Learn more about Oracle Cost Advisor for AI.
Cerebras inference is coming to OCI
Oracle and Cerebras announced plans to bring Cerebras inference to OCI in calendar 2027. The collaboration builds on Oracle’s existing AI inference capabilities, including Dedicated AI Clusters in OCI Enterprise AI.
Faster inference can help developers create more responsive applications and agents that make multiple model calls to complete a task. Cerebras will add another option for workloads where response speed is critical.
Read the Oracle and Cerebras announcement.
Smart Model Router expands AI access across 30 additional OCI regions
OCI Enterprise AI is expanding to 30 additional commercial regions, with Smart Model Router providing access to supported on-demand models through reusable routing profiles. Customers select the model and permitted destination regions, and OCI routes requests to approved regions with available GPU capacity.
By connecting applications to model capacity beyond their local region, Smart Model Router helps teams extend AI experiences globally without building custom routing logic. Customers retain control over where inference is processed while reducing regional infrastructure management.
See how Smart Model Router works and explore expanded regional access.
Agent Sandbox enables controlled execution for AI agents
OCI Enterprise AI Agent Sandbox is now generally available, providing managed, temporary Linux workspaces where agents can test generated code, analyze data, and run diagnostic scripts. Organizations configure compute resources, execution duration, and network access while OCI manages the underlying infrastructure.
These controlled workspaces let teams give agents more capable tasks without building a separate execution platform. Customers can automate multistep workflows while maintaining execution boundaries and reducing operational overhead.
Get started with sandboxes in OCI Enterprise AI.
New NVIDIA Nemotron, Alibaba Qwen, and Xiaomi MiMo models expand choice
OCI Enterprise AI added Model Import support for NVIDIA Nemotron 3 Ultra FP8 and Nemotron 3.5 Super VL NVFP4, Alibaba Qwen3.8-Flash-Next and Qwen3.6-35B-A3B-FP8, and Xiaomi MiMo V2.6 Pro RL. Customers can import these models and create endpoints on supported Dedicated AI Cluster configurations.
These additions expand the options available for text generation and image-and-text understanding. Teams can evaluate models against their application requirements, with supported hardware and regional availability varying by model.
Explore October’s model additions in the release notes.
Persistent file storage helps hosted applications retain state
Hosted applications now support persistent file storage for model artifacts, intermediate results, checkpoints, and application state across restarts and redeployments. Customers can configure up to 50 GB of storage, with this release supporting a single application replica.
Developers can retain working files and application state without rebuilding them each time an application restarts. This simplifies hosted workflows that need continuity between runs.
Learn how to add persistent file storage.
Context compaction supports longer conversations and agent workflows
The OCI Responses API now supports context compaction, which reduces accumulated conversation history to a smaller representation for subsequent requests. Developers can trigger compaction when needed or configure it to run automatically at a token threshold.
As conversations and agent workflows grow, compaction helps applications carry context forward without passing the full history into every request. This makes it easier to manage longer interactions and multistep tasks while controlling the amount of context sent to the model.
Explore context compaction and the implementation tutorial.
Hosted applications gain an option for public content
Developers can now select “No authentication” when creating a hosted application to serve public content or manage authentication within the application itself. Requests to the endpoint do not require platform-managed authentication.
This gives teams more flexibility when deploying public-facing experiences or integrating their own sign-in systems. Applications that require restricted access must implement their own authentication and authorization when using this option.
Learn about the public-content access option.
Expanded hardware support gives imported models more deployment options
OCI Enterprise AI expanded supported hardware configurations for selected imported models, including Qwen, NVIDIA Nemotron, GLM, and OpenAI gpt-oss models. For example, Qwen3.8-Flash-Next now supports an A100 configuration alongside existing H100 and H200 options.
Customers now have more flexibility to match model deployments to available capacity, performance requirements, and budgets. Supported configurations vary by model and region.
Review expanded hardware support for imported models.
Browser session authentication simplifies access to hosted applications
Identity domain browser session authentication is now generally available for hosted applications, adding OAuth 2.1 support. Users sign in through an OCI identity domain, and the hosted application service establishes their browser session.
Developers can provide an interactive sign-in experience using OCI identity services without building the browser authentication flow themselves. This helps teams make hosted AI applications accessible to authorized users through a familiar sign-in process.
Learn how to configure browser session authentication.
New resources for AI at OCI
Oracle named a Leader in the IDC MarketScape for Agentic AI Platforms
Oracle was named a Leader in the IDC MarketScape: Worldwide Agentic AI Platforms 2026 Vendor Assessment. According to the report, Oracle was recognized for the following strengths: enterprise scale, distribution and end-to-end technology focus, openness in AI model selection, and sovereignty and deployment breadth
For customers evaluating where to build and run AI agents, the report offers an independent perspective on Oracle’s capabilities and strategy. It provides useful context for considering OCI Enterprise AI, which combines agent development, orchestration, and governance with flexibility in model selection and deployment.
Read the IDC MarketScape excerpt.
First Principles guide to building AI agents on OCI
Pradeep Vincent, EVP of Compute Platform and AI Infrastructure Services, and Egor Pushkin, Chief Architect for OCI Data and AI, explain how OCI Enterprise AI brings together model inference, agent hosting, and controlled execution in a new First Principles blog and video. The guide explores how agents use tools, preserve context, and complete multistep tasks while keeping permissions and approvals explicit.
By connecting these architectural concepts to practical examples, this resource helps teams plan agents that can act on business goals and verify their results. Developers can use the guidance to clarify responsibilities, manage execution boundaries, and design workflows that recover from interruptions.
Explore First Principles: Building AI Agents on OCI.
Catch up on Oracle AI World sessions and announcements
Oracle AI World 2026 keynotes, product spotlights, and Oracle TV coverage are available on demand. Learning-session access is available through November 25.
Customers can revisit demonstrations, explore implementation approaches, and share relevant sessions with colleagues as they evaluate new AI capabilities. The replays offer a practical way to catch up on sessions missed during the event.
Explore Oracle AI World replays and resources.
For more Oracle AI news: What’s New in Oracle AI? October 2026 Edition