Skip to main content

Oracle Cloud Infrastructure (OCI)

LiteLLM supports the following models for OCI on-demand GenAI API.

Check the OCI Models List to see if the model is available for your region.

Supported Models​

For model lifecycle, retirement dates, and recommended replacements, see OCI's on-demand model retirement page. Oracle is the authoritative source.

Chat / Text Generation​

Meta Llama Models​

  • meta.llama-4-maverick-17b-128e-instruct-fp8 (multimodal)
  • meta.llama-4-scout-17b-16e-instruct (multimodal)
  • meta.llama-3.3-70b-instruct
  • meta.llama-3.3-70b-instruct-fp8-dynamic
  • meta.llama-3.2-90b-vision-instruct (multimodal)
  • meta.llama-3.2-11b-vision-instruct (multimodal)

xAI Grok Models​

  • xai.grok-4.3
  • xai.grok-4.20
  • xai.grok-4.20-multi-agent
  • xai.grok-4
  • xai.grok-4-fast
  • xai.grok-4.1-fast
  • xai.grok-3
  • xai.grok-3-fast
  • xai.grok-3-mini
  • xai.grok-3-mini-fast
  • xai.grok-code-fast-1

Cohere Models​

  • cohere.command-latest
  • cohere.command-a-03-2025
  • cohere.command-a-reasoning-08-2025
  • cohere.command-a-vision-07-2025 (multimodal)
  • cohere.command-a-translate-08-2025
  • cohere.command-plus-latest
  • cohere.command-r-plus-08-2024
  • cohere.command-r-08-2024

Google Gemini Models (via OCI)​

  • google.gemini-2.5-pro (multimodal)
  • google.gemini-2.5-flash (multimodal)
  • google.gemini-2.5-flash-lite (multimodal)

OpenAI Open-Source Models (via OCI)​

  • openai.gpt-oss-120b
  • openai.gpt-oss-20b

Embedding Models​

  • cohere.embed-v4.0 (1536 dimensions, multimodal)
  • cohere.embed-english-v3.0 (1024 dimensions)
  • cohere.embed-english-light-v3.0 (384 dimensions)
  • cohere.embed-multilingual-v3.0 (1024 dimensions)
  • cohere.embed-multilingual-light-v3.0 (384 dimensions)
  • cohere.embed-english-image-v3.0 (1024 dimensions, multimodal)
  • cohere.embed-english-light-image-v3.0 (384 dimensions, multimodal)
  • cohere.embed-multilingual-image-v3.0 (1024 dimensions, multimodal)
  • cohere.embed-multilingual-light-image-v3.0 (384 dimensions, multimodal)

Authentication​

LiteLLM supports two authentication methods for OCI:

Method 1: Manual Credentials​

Provide individual OCI credentials directly to LiteLLM. Follow the official Oracle tutorial to create a signing key and obtain the following parameters:

  • user
  • fingerprint
  • tenancy
  • region
  • key_file or key
  • compartment_id

This is the default method for LiteLLM AI Gateway (LLM Proxy) access to OCI GenAI models.

Environment Variables

Instead of passing credentials in code, you can set the following environment variables and LiteLLM will read them automatically:

export OCI_REGION="us-chicago-1"
export OCI_USER="ocid1.user.oc1.."
export OCI_FINGERPRINT="xx:xx:xx:xx:xx:xx:xx:xx:xx:xx:xx:xx:xx:xx:xx:xx"
export OCI_TENANCY="ocid1.tenancy.oc1.."
export OCI_COMPARTMENT_ID="ocid1.compartment.oc1.."
# Provide either the private key content OR the path to the key file:
export OCI_KEY_FILE="/path/to/oci_api_key.pem"
# export OCI_KEY="-----BEGIN PRIVATE KEY-----\n..."

Method 2: OCI SDK Signer​

Use an OCI SDK Signer object for authentication. This method:

  • Uses the official OCI SDK for signing
  • Supports additional authentication methods (instance principals, workload identity, etc.)

To use this method, install the OCI SDK:

uv add oci

This method is an alternative when using the LiteLLM SDK on Oracle Cloud Infrastructure (instances or Oracle Kubernetes Engine).

Regions and Government Realms​

LiteLLM builds the inference endpoint as https://inference.generativeai.{region}.oci.{realm_domain}. The region is oci_region (or the OCI_REGION environment variable, defaulting to us-ashburn-1). The realm domain depends on which OCI realm hosts that region: oraclecloud.com for the commercial realm (OC1), oraclegovcloud.com for the US Government realms (OC2 and OC3), oraclegovcloud.uk for the UK Government realm (OC4), and the matching domain for every other realm the OCI SDK knows. A request in us-luke-1 therefore goes to https://inference.generativeai.us-luke-1.oci.oraclegovcloud.com without any endpoint configuration.

LiteLLM takes the realm domain from the first of these sources that answers:

  1. The realm key inside oci_compartment_id. Every OCID carries it as its third segment, so ocid1.compartment.oc2.. selects the US Government realm and ocid1.compartment.oc1.. the commercial realm.
  2. The region registry of the OCI Python SDK (oci.regions), when the oci package is installed.
  3. The region metadata the OCI SDK also reads: ~/.oci/regions-config.json (a JSON array) first, then the OCI_REGION_METADATA environment variable (one JSON object). The entry whose regionIdentifier matches the region supplies realmDomainComponent.
  4. The commercial realm, oraclecloud.com.

Setting api_base skips this resolution entirely, see Overriding the endpoint with api_base.

Because the compartment OCID already names the realm, a Government deployment needs the same parameters as a commercial one:

from litellm import completion

response = completion(
    model="oci/meta.llama-3.3-70b-instruct",
    messages=[{"role": "user", "content": "Hello"}],
    oci_region="us-luke-1",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_key_file="<path/to/oci_key.pem>",
    oci_compartment_id="ocid1.compartment.oc2..",
)

Sources 2 and 3 only matter when oci_compartment_id is missing or is not an OCID with a realm key LiteLLM recognizes. Metadata entries that fail validation are skipped with a warning instead of failing the request, and a region that no source describes keeps the commercial endpoint, so a Government model never changes the endpoint of the other OCI models on the same proxy.

Overriding the endpoint with api_base​

Set api_base to send requests to an exact host, for example a realm LiteLLM does not know yet, a private endpoint, or a gateway in front of OCI. LiteLLM uses the value as given and appends the action path itself (/20231130/actions/chat for chat, /20231130/actions/embedText for embeddings). If api_base already ends with one of those paths, that suffix is removed first so the URL is not doubled. Requests are still signed with your OCI credentials against the host in api_base, and oci_region no longer affects the URL.

from litellm import completion

response = completion(
    model="oci/meta.llama-3.3-70b-instruct",
    messages=[{"role": "user", "content": "Hello"}],
    api_base="https://inference.generativeai.us-gov-ashburn-1.oci.oraclegovcloud.com",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_key_file="<path/to/oci_key.pem>",
    oci_compartment_id="<oci_compartment_id>",
)

Usage​

Input the parameters obtained from the OCI signing key creation process into the completion function:

from litellm import completion

messages = [{"role": "user", "content": "Hey! how's it going?"}]
response = completion(
    model="oci/xai.grok-4",
    messages=messages,
    oci_region="<your_oci_region>",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_serving_mode="ON_DEMAND",  # Optional, default is "ON_DEMAND". Other option is "DEDICATED"
    # Provide either the private key string OR the path to the key file:
    # Option 1: pass the private key as a string
    oci_key="<string_with_content_of_oci_key>",
    # Option 2: pass the private key file path
    # oci_key_file="<path/to/oci_key.pem>",
    oci_compartment_id="<oci_compartment_id>",
)
print(response)

LiteLLM Proxy Usage​

Here's how to call OCI GenAI through the LiteLLM Proxy Server.

1. Setup config.yaml​

model_list:
  - model_name: oci-grok-4
    litellm_params:
      model: oci/xai.grok-4
      oci_region: os.environ/OCI_REGION
      oci_user: os.environ/OCI_USER
      oci_fingerprint: os.environ/OCI_FINGERPRINT
      oci_tenancy: os.environ/OCI_TENANCY
      oci_key_file: os.environ/OCI_KEY_FILE
      oci_compartment_id: os.environ/OCI_COMPARTMENT_ID

  - model_name: oci-cohere-command
    litellm_params:
      model: oci/cohere.command-latest
      oci_region: os.environ/OCI_REGION
      oci_user: os.environ/OCI_USER
      oci_fingerprint: os.environ/OCI_FINGERPRINT
      oci_tenancy: os.environ/OCI_TENANCY
      oci_key_file: os.environ/OCI_KEY_FILE
      oci_compartment_id: os.environ/OCI_COMPARTMENT_ID

All possible auth params:

oci_region: Optional[str],
oci_user: Optional[str],
oci_fingerprint: Optional[str],
oci_tenancy: Optional[str],
oci_key: Optional[str],          # private key content as string
oci_key_file: Optional[str],     # path to .pem file
oci_compartment_id: Optional[str],
oci_serving_mode: Optional[str], # "ON_DEMAND" (default) or "DEDICATED"
oci_endpoint_id: Optional[str],  # only used with DEDICATED

2. Start the proxy​

litellm --config /path/to/config.yaml

3. Test it​

curl --location 'http://0.0.0.0:4000/chat/completions' \
--header 'Content-Type: application/json' \
--data '{
  "model": "oci-grok-4",
  "messages": [
    {"role": "user", "content": "what llm are you"}
  ]
}'

Usage - Streaming​

Just set stream=True when calling completion.

from litellm import completion

messages = [{"role": "user", "content": "Hey! how's it going?"}]
response = completion(
    model="oci/xai.grok-4",
    messages=messages,
    stream=True,
    oci_region="<your_oci_region>",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_serving_mode="ON_DEMAND",  # Optional, default is "ON_DEMAND". Other option is "DEDICATED"
    # Provide either the private key string OR the path to the key file:
    # Option 1: pass the private key as a string
    oci_key="<string_with_content_of_oci_key>",
    # Option 2: pass the private key file path
    # oci_key_file="<path/to/oci_key.pem>",
    oci_compartment_id="<oci_compartment_id>",
)
for chunk in response:
    print(chunk["choices"][0]["delta"]["content"])  # same as openai format

Usage Examples by Model Type​

Using Cohere Models​

from litellm import completion

messages = [{"role": "user", "content": "Explain quantum computing"}]
response = completion(
    model="oci/cohere.command-latest",
    messages=messages,
    oci_region="us-chicago-1",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_key="<string_with_content_of_oci_key>",
    oci_compartment_id="<oci_compartment_id>",
)
print(response)

Using Dedicated Endpoints​

OCI supports dedicated endpoints for hosting models. Use the oci_serving_mode="DEDICATED" parameter along with oci_endpoint_id to specify the endpoint ID.

from litellm import completion

messages = [{"role": "user", "content": "Hey! how's it going?"}]
response = completion(
    model="oci/xai.grok-4",  # Must match the model type hosted on the endpoint
    messages=messages,
    oci_region="<your_oci_region>",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_serving_mode="DEDICATED",
    oci_endpoint_id="ocid1.generativeaiendpoint.oc1...",  # Your dedicated endpoint OCID
    oci_key="<string_with_content_of_oci_key>",
    oci_compartment_id="<oci_compartment_id>",
)
print(response)

Important: When using oci_serving_mode="DEDICATED":

  • The model parameter must match the type of model hosted on your dedicated endpoint (e.g., use "oci/cohere.command-latest" for Cohere models, "oci/xai.grok-4" for Grok models)
  • The model name determines the API format and vendor-specific handling (Cohere vs Generic)
  • The oci_endpoint_id parameter specifies your dedicated endpoint's OCID
  • If oci_endpoint_id is not provided, the model parameter will be used as the endpoint ID (for backward compatibility)

Example with Cohere Dedicated Endpoint:

# For a dedicated endpoint hosting a Cohere model
response = completion(
    model="oci/cohere.command-latest",  # Use Cohere model name to get Cohere API format
    messages=messages,
    oci_region="us-chicago-1",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_serving_mode="DEDICATED",
    oci_endpoint_id="ocid1.generativeaiendpoint.oc1...",  # Your Cohere endpoint OCID
    oci_key="<string_with_content_of_oci_key>",
    oci_compartment_id="<oci_compartment_id>",
)

Usage - Function Calling / Tool Calling​

OCI GenAI supports OpenAI-compatible function calling. LiteLLM normalizes the request and response shape so the same code that targets OpenAI works with OCI Cohere and Generic (xAI Grok, Meta Llama, Google Gemini) models.

from litellm import completion

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_current_weather",
            "description": "Get the current weather in a given location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {
                        "type": "string",
                        "description": "The city and state, e.g. San Francisco, CA",
                    },
                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
                },
                "required": ["location"],
            },
        },
    }
]

response = completion(
    model="oci/xai.grok-4",
    messages=[{"role": "user", "content": "What's the weather in Boston today?"}],
    tools=tools,
    tool_choice="auto",
    oci_region="us-chicago-1",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_key_file="<path/to/oci_key.pem>",
    oci_compartment_id="<oci_compartment_id>",
)

# Inspect the tool call
print(response.choices[0].message.tool_calls)

Tool calling works with both Cohere (cohere.command-*) and Generic (xai.grok-*, meta.llama-*, google.gemini-*) model families. LiteLLM adapts the OpenAI tool schema to each vendor's native format internally.

Usage - Vision / Multimodal​

OCI GenAI exposes vision-capable models that accept images alongside text. Pass images using the standard OpenAI image_url content block.

from litellm import completion

response = completion(
    model="oci/meta.llama-4-maverick-17b-128e-instruct-fp8",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "What is in this image?"},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
                    },
                },
            ],
        }
    ],
    oci_region="us-chicago-1",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_key_file="<path/to/oci_key.pem>",
    oci_compartment_id="<oci_compartment_id>",
)
print(response.choices[0].message.content)

Vision-capable models on OCI include:

  • meta.llama-4-maverick-17b-128e-instruct-fp8
  • meta.llama-4-scout-17b-16e-instruct
  • meta.llama-3.2-11b-vision-instruct
  • meta.llama-3.2-90b-vision-instruct
  • cohere.command-a-vision-07-2025
  • google.gemini-2.5-pro, google.gemini-2.5-flash, google.gemini-2.5-flash-lite

Both URL and base64-encoded data URIs are supported.

Usage - Reasoning / Thinking​

OCI Generic-vendor models (xAI Grok reasoning variants, Google Gemini, etc.) support a reasoning step. LiteLLM exposes this via the OpenAI-compatible reasoning_effort parameter; accepted values are "low", "medium", "high", and "disable" (mapped to OCI's NONE).

Returned reasoning tokens are surfaced on usage.completion_tokens_details.reasoning_tokens, matching the OpenAI shape.

from litellm import completion

response = completion(
    model="oci/xai.grok-3-mini",
    messages=[{"role": "user", "content": "If 3x + 7 = 22, what is x? Show your reasoning."}],
    reasoning_effort="high",  # "low" | "medium" | "high" | "disable"
    oci_region="us-chicago-1",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_key_file="<path/to/oci_key.pem>",
    oci_compartment_id="<oci_compartment_id>",
)

print(response.choices[0].message.content)
print("Reasoning tokens:", response.usage.completion_tokens_details.reasoning_tokens)
note

reasoning_effort is only honored on Generic-vendor reasoning models (e.g., xai.grok-3-mini, xai.grok-4, google.gemini-2.5-pro). It is silently ignored for OCI Cohere models, which are not reasoning models.

Optional Parameters​

ParameterTypeDefaultEnvironment VariableDescription
oci_regionstringus-ashburn-1OCI_REGIONOCI region where the GenAI service is deployed
oci_serving_modestringON_DEMAND–Service mode: ON_DEMAND for managed models or DEDICATED for dedicated endpoints
oci_endpoint_idstringSame as model–(For DEDICATED mode) The OCID of your dedicated endpoint
oci_compartment_idstringRequiredOCI_COMPARTMENT_IDThe OCID of the OCI compartment containing your resources
oci_userstring–OCI_USER(Manual auth) The OCID of the OCI user
oci_fingerprintstring–OCI_FINGERPRINT(Manual auth) The fingerprint of the API signing key
oci_tenancystring–OCI_TENANCY(Manual auth) The OCID of your OCI tenancy
oci_keystring–OCI_KEY(Manual auth) The private key content as a string
oci_key_filestring–OCI_KEY_FILE(Manual auth) Path to the private key file
oci_signerobject––(SDK auth) OCI SDK Signer object for authentication
reasoning_effortstring––Reasoning level for Generic-vendor reasoning models: low, medium, high, disable

Embeddings​

LiteLLM supports OCI Generative AI embedding models. These models use the same authentication methods described above.

from litellm import embedding

response = embedding(
    model="oci/cohere.embed-english-v3.0",
    input=["Hello world", "Goodbye world"],
    oci_region="us-ashburn-1",
    oci_user="<your_oci_user>",
    oci_fingerprint="<your_oci_fingerprint>",
    oci_tenancy="<your_oci_tenancy>",
    oci_key="<string_with_content_of_oci_key>",
    oci_compartment_id="<oci_compartment_id>",
)
print(response)

Embedding Optional Parameters​

ParameterTypeDefaultDescription
input_typestring-The type of input: search_document, search_query, classification, clustering
truncatestringENDTruncation strategy when input exceeds max tokens: END or START

Using Dedicated Embedding Endpoints​

response = embedding(
    model="oci/cohere.embed-english-v3.0",
    input=["Hello world"],
    oci_serving_mode="DEDICATED",
    oci_endpoint_id="ocid1.generativeaiendpoint.oc1...",
    oci_region="us-ashburn-1",
    oci_compartment_id="<oci_compartment_id>",
    # ... auth params
)
LiteLLM Enterprise
SSO/SAML, audit logs, spend tracking, multi-team management, and guardrails, built for production.
Learn more →