Overview
The BYOM (Bring Your Own Model) catalog now reflects the current generation from three providers: OpenAI, Google Vertex AI, and OCI Generative AI. The new models appear automatically in the Connect AI model selector, ready to use in your AI Agents and Behaviors. You pick the model that best fits each task, with the flexibility to always work with the latest available version.
What's new?
New models available in the Connect AI selector:
- OpenAI: gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.5, gpt-5.5-pro, gpt-5.6-luna, gpt-5.6-terra, and gpt-5.6-sol.
- Google Vertex AI: Gemini 3.5 Flash.
- OCI Generative AI: Grok 4.20 (reasoning and non-reasoning) and Grok 4.3.
Other catalog changes:
- OCI tool-calling fix: the Llama 4 Maverick and Scout models now accept tool-calling requests again. Behaviors and agents using these models run as expected.
Removed models: Gemini 1.5, Gemini 2.0, and xai.grok-4 are no longer selectable. Orgs still configured with these models need to pick a replacement in the selector.
How does this help?
- Access to the current model generation with no setup: the latest releases from OpenAI, Vertex AI, and OCI reach the selector automatically, so you always work at the frontier of what each provider offers.
- The right model for each task: with a broader catalog, you tune cost, speed, and reasoning capability per use case, from a lighter model for triage to a more robust one for complex analysis.
- Reliable tool-calling on OCI: Llama 4 Maverick and Scout run tool calls again, so your agents complete the full flow without interruption.
Use cases
1. Ticket triage agent in a Service Desk. The agent reads the opening text of the card and classifies the ticket by type (access, hardware, software, network). With the refreshed catalog, you point this agent to a lighter, more economical model like gpt-5.4-nano, since the task is simple, high-volume classification and doesn't need the most expensive model in the line.
2. Invoice data extraction Behavior in Accounts Payable. The Behavior receives the attachment and extracts vendor, amount, due date, and cost center. You select a model with stronger reasoning like gpt-5.5-pro for documents with complex tables, raising extraction accuracy without changing the process design.
3. Vendor response agent in Procurement. The agent drafts replies to quote requests from the card data. By switching to Gemini 3.5 Flash on Vertex AI, you gain response speed in a high-volume flow while keeping the same configured behavior.
4. Tool-calling agent on OCI in Employee Onboarding. The agent triggers tools to look up databases and create connected cards in other pipes. With the fix to the Llama 4 Maverick and Scout models, those tool calls run again, so the agent completes the flow end to end without failing at the integration step.
5. Migrating an agent configured with a retired model. A company still pointed its agents to xai.grok-4. With the model removed from the catalog, the Org Admin selects a current replacement in Connect AI, and the agents return to operating on a supported model, no longer relying on a discontinued version.
Most relevant for
- Org Admins
- Pipe Admins
- AI Agent builders
- Operations teams
