Generative AI Models and Regions for Enterprise AI Agents
Enterprise AI Agents support a subset of OCI Generative AI foundational and imported models in select regions. This page lists the supported models and regions for runtime agentic inference and project memory.
To Call Models
The following sections list the foundational and imported models supported for OCI-OpenAI Compatible endpoints. Some models also support OpenAI-Compatible Tools. See each section for the supported access path.
Available Models
Agents can call the following foundational and imported models for agentic inference use cases. The supported access path depends on the model type.
Available Foundational Models
The following pretrained foundational models are supported for the OCI OpenAI-Compatible Endpoints and OpenAI-Compatible Tools:
- Google Vertex AI Platform
- OpenAI Open Source
- xAI Platform
External Calls
The Google Gemini and xAI Grok models are hosted externally. See External Calls for model- and region-specific information.
Available Imported Models
The following imported models are supported for the OCI Responses API:
- Alibaba Qwen
- Qwen/Qwen3.6-35B-A3B
- Google Gemma
- google/gemma-4-31B-it
These models are supported through Model Import and aren't available through the on-demand serving mode. To call one of these models through the OCI Responses API, first import the model into OCI Generative AI, and then create an endpoint for the imported model on a hosting dedicated AI cluster.
Select a model link in the preceding list to see the minimum supported dedicated AI cluster unit shapes for that model. Then use the following hardware-shape information to plan the deployment:
- Hardware Unit Shapes by Region lists the hardware shapes available in each region.
- Hardware Unit Shapes for Imported Models explains how to select a compatible shape for an imported model and target region.
- Hardware Unit Shapes, Service Limits, and AI Unit Counts lists the service limit, required hardware units, and AI unit count for each unit shape.
Available Commercial Regions (OC1)
You can access agentic inference models in one or more of the following OC1 regions:
| Region Name | Location | Region Identifier | Region Key |
|---|---|---|---|
| Brazil East (Sao Paulo) | Sao Paulo |
sa-saopaulo-1
|
GRU
|
| Germany Central (Frankfurt) | Frankfurt |
eu-frankfurt-1
|
FRA
|
| India South (Hyderabad) | Hyderabad |
ap-hyderabad-1
|
HYD
|
| Japan Central (Osaka) | Osaka |
ap-osaka-1
|
KIX
|
| Saudi Arabia Central (Riyadh) | Riyadh |
me-riyadh-1
|
RUH
|
| UK South (London) | London |
uk-london-1
|
LHR
|
| US East (Ashburn) | Ashburn |
us-ashburn-1
|
IAD
|
| US Midwest (Chicago) | Chicago |
us-chicago-1
|
ORD
|
| US West (Phoenix) | Phoenix |
us-phoenix-1
|
PHX
|
Learn About Regions and Availability Domains.
- UAE East (Dubai): OCI-OpenAI Compatible endpoints and tools aren't available in this region.
- Availability: Not every listed model is available in the preceding listed regions. For per-model supported regions and deployment details, see the, see the Models by Region page.
- External Calls: For notes about models with external calls, see External Calls.
The following models are supported for the xAI-Compatible tools used with the OCI Responses API.
Supported xAI Models
Supported Regions
You can access the xAI models, by calling the endpoints in the following OC1 regions:
| Region Name | Location | Region Identifier | Region Key |
|---|---|---|---|
| US East (Ashburn) | Ashburn |
us-ashburn-1
|
IAD
|
| US Midwest (Chicago) | Chicago |
us-chicago-1
|
ORD
|
| US West (Phoenix) | Phoenix |
us-phoenix-1
|
PHX
|
External Calls to xAI Grok Models
The xAI Grok models that can be accessed through the OCI Generative AI service in an OCI region are hosted externally by xAI. Therefore, a call to an xAI Grok model through the OCI Generative AI service results in a call to an xAI location.
To Enable Project Memory
When you create a project, you can enable short-term memory compaction for conversations and responses related to that project. See the following table for the available models and regions for memory compaction.
| Region | Region Code | Short-Term Memory Model |
|---|---|---|
| Brazil East (Sao Paulo) | sa-saopaulo-1 |
|
| Germany Central (Frankfurt) | eu-frankfurt-1 |
|
| UK South (London) | uk-london-1 |
|
| India South (Hyderabad) | ap-hyderabad-1 |
|
| US East (Ashburn) (cross region to US Midwest (Chicago) | us-ahsburn-1 (cross region to us-chicago-1) |
|
| Japan Central (Osaka) | ap-osaka-1 |
|
| Saudi Arabia Central (Riyadh) | me-riyadh-1 |
|
| US Midwest (Chicago) | us-chicago-1 |
|
| US West (Phoenix) (cross region to US Midwest (Chicago) | us-phoenix-1(cross region to us-chicago-1) |
For Extracting Key Information For Long-Term Memory
hat
When you create a project, you can enable information extraction from conversations and responses for a long-term memory feature. You select an extraction model that aims to extract key information from conversations. See the following table for regions and model supported for extracting key information to use for long-term memory.
| Region | Available Extraction Model |
|---|---|
| Available Commercial Regions (OC1) cross reference with regions that OpenAI gpt-oss-120b is available | OpenAI gpt-oss-120b |
When you create a project, you can, you can select an embedding model to store extracted memories as searchable vectors. The available embedding model depends on the project region. See the following table for the embedding model available in each region:
| Region | Region Code | Available Embed Model |
|---|---|---|
| Brazil East (Sao Paulo) | sa-saopaulo-1 |
Cohere Embed Multilingual 3 (Deprecated) |
| Germany Central (Frankfurt) | eu-frankfurt-1 |
Cohere Embed Multilingual 3 (Deprecated) |
| UK South (London) | uk-london-1 |
Cohere Embed Multilingual 3 (Deprecated) |
| India South (Hyderabad) | ap-hyderabad-1 |
Cohere Embed Multilingual Image 3 (Deprecated) |
| US East (Ashburn) (cross region to US Midwest (Chicago) | us-ahsburn-1 |
Cohere Embed 4 |
| Japan Central (Osaka) | ap-osaka-1 |
Cohere Embed 4 |
| Saudi Arabia Central (Riyadh) | me-riyadh-1 |
Cohere Embed 4 |
| US Midwest (Chicago) | us-chicago-1 |
Cohere Embed 4 |
| US West (Phoenix) (cross region to US Midwest (Chicago) | us-phoenix-1
|
Cohere Embed 4 |
External Calls
External Calls to xAI Grok Models
External Calls to xAI Grok Models
The xAI Grok models that can be accessed through the OCI Generative AI service in an OCI region are hosted externally by xAI. Therefore, a call to an xAI Grok model through the OCI Generative AI service results in a call to an xAI location.
External Calls to Google Gemini Models
External Calls to Google Gemini 2.5 Pro for US Regions
The Google Gemini 2.5 Pro model that can be accessed through the OCI Generative AI service in US regions, are hosted externally by Google. Therefore, a call to a Google Gemini 2.5 Pro model (through the OCI Generative AI service) results in a call to a Google location. For Google Gemini 2.5 Pro, a Google Americas regional location is used, which routes the request to only a Google Americas location. Machine Learning Processing takes place within a Google Americas location.
External Calls to Google Gemini 2.5 Pro for EU Regions
The Google Gemini 2.5 Pro model that can be accessed through the OCI Generative AI service in the Frankfurt region, are hosted externally by Google. Therefore, a call to a Google Gemini 2.5 Pro model (through the OCI Generative AI service) results in a call to a Google location. For Gemini 2.5 Pro, a Google European Union (EU) regional location is used, which routes the request to only a Google EU location. Machine Learning Processing takes place within a Google EU location.
External Calls to Google Gemini 2.5 Pro for AP Regions
The Google Gemini 2.5 Pro model that can be accessed through the OCI Generative AI service in the Osaka region, are hosted externally by Google. Therefore, a call to a Google Gemini 2.5 Pro model (through the OCI Generative AI service) results in a call to a Google location. For Gemini 2.5 Pro, a Google Asia Pacific regional location is used, which routes the request to only a Google Asia Pacific location. Machine Learning Processing can take place within any Google location globally.
External Calls to Gemini 2.5 Flash for US Regions
The Gemini 2.5 Flash model that can be accessed through the OCI Generative AI service in US regions, are hosted externally by Google. Therefore, a call to a Gemini 2.5 Flash model (through the OCI Generative AI service) results in a call to a Google location. For Gemini 2.5 Flash, a Google Americas regional location is used, which routes the request to only a Google Americas location. Machine Learning Processing takes place within a Google Americas location.
External Calls to Gemini 2.5 Flash in the Brazil Region
The Gemini 2.5 Flash model that can be accessed through the OCI Generative AI service in the Brazil location is hosted externally by Google. Therefore, a call to a Gemini 2.5 Flash model through the OCI Generative AI service results in a call to a Google location in Brazil. Machine Learning Processing takes place only within Google's Brazil location.
External Calls to Gemini 2.5 Flash for EU Regions
The Gemini 2.5 Flash model that can be accessed through the OCI Generative AI service in the Frankfurt region, are hosted externally by Google. Therefore, a call to a Gemini 2.5 Flash model (through the OCI Generative AI service) results in a call to a Google location. For Gemini 2.5 Flash, a Google European Union (EU) regional location is used, which routes the request to only a Google EU location. Machine Learning Processing takes place within a Google EU location.
External Calls to Gemini 2.5 Flash for AP Regions
The Gemini 2.5 Flash model that can be accessed through the OCI Generative AI service in the Osaka region and the Hyderabad region, are hosted externally by Google. Therefore, a call to a Gemini 2.5 Flash model (through the OCI Generative AI service) results in a call to a Google location. For Gemini 2.5 Flash, a Google Asia Pacific regional location is used, which routes the request to only a Google Asia Pacific location. Machine Learning Processing can take place within any Google location globally.
External Calls to Gemini 2.5 Flash-Lite for US Regions
The Gemini 2.5 Flash-Lite model that can be accessed through the OCI Generative AI service in US regions, are hosted externally by Google. Therefore, a call to a Gemini 2.5 Flash-Lite model (through the OCI Generative AI service) results in a call to a Google location. For Gemini 2.5 Flash-Lite, a Google Americas regional location is used, which routes the request to only a Google Americas location. Machine Learning Processing takes place within a Google Americas location.
External Calls to Gemini 2.5 Flash-Lite for EU Regions
The Gemini 2.5 Flash-Lite model that can be accessed through the OCI Generative AI service in the Frankfurt region, are hosted externally by Google. Therefore, a call to a Gemini 2.5 Flash-Lite model (through the OCI Generative AI service) results in a call to a Google location. For Gemini 2.5 Flash-Lite, a Google European Union (EU) regional location is used, which routes the request to only a Google EU location. Machine Learning Processing takes place within a Google EU location.