Snowflake Cortex | liteLLM (original) (raw)

LiteLLM supports all models on the Snowflake Cortex REST API, including models from Anthropic (Claude), OpenAI (GPT), Meta (Llama), Mistral, DeepSeek, and Snowflake.

Tip : We support ALL Snowflake Cortex models. Use model=snowflake/<model-name> as a prefix when sending LiteLLM requests.

Authentication​

Snowflake Cortex REST API supports three authentication methods.

Programmatic Access Token (PAT) — Recommended​

The simplest approach. Generate a PAT in Snowsight under User Menu → My Profile → Programmatic Access Tokens.

import os
from litellm import completion

os.environ["SNOWFLAKE_API_KEY"] = "pat/<your-programmatic-access-token>"
os.environ["SNOWFLAKE_API_BASE"] = "https://<account>.snowflakecomputing.com/api/v2/cortex/v1"

response = completion(
    model="snowflake/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "Hello!"}],
)

JWT (Key-Pair Authentication)​

Generate a JWT from a Snowflake key pair. See Key-pair authentication.

import os
from litellm import completion

os.environ["SNOWFLAKE_JWT"] = "<your-jwt-token>"
os.environ["SNOWFLAKE_ACCOUNT_ID"] = "<orgname>-<account_name>"

response = completion(
    model="snowflake/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "Hello!"}],
)

Pass credentials as parameters​

from litellm import completion

# Using PAT
response = completion(
    model="snowflake/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "Hello!"}],
    api_key="pat/<your-pat-token>",
    api_base="https://<account>.snowflakecomputing.com/api/v2/cortex/v1",
)

# Using JWT
response = completion(
    model="snowflake/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "Hello!"}],
    api_key="<your-jwt-token>",
    account_id="<orgname>-<account_name>",
)

For all authentication options, see Authenticating to Cortex REST API.

Usage​

• SDK
• PROXY

from litellm import completion
import os

os.environ["SNOWFLAKE_API_KEY"] = "pat/<your-pat>"
os.environ["SNOWFLAKE_API_BASE"] = "https://<account>.snowflakecomputing.com/api/v2/cortex/v1"

response = completion(
    model="snowflake/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "What is Snowflake Cortex?"}],
)
print(response.choices[0].message.content)

Supported OpenAI Parameters​

temperature, max_tokens, top_p, stream, response_format,
tools, tool_choice

Streaming​

• SDK
• PROXY

from litellm import completion
import os

os.environ["SNOWFLAKE_API_KEY"] = "pat/<your-pat>"
os.environ["SNOWFLAKE_API_BASE"] = "https://<account>.snowflakecomputing.com/api/v2/cortex/v1"

response = completion(
    model="snowflake/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "Write a haiku about data."}],
    stream=True,
)

for chunk in response:
    if chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Supported on Claude and select models. LiteLLM automatically transforms OpenAI tool format to Snowflake's tool_spec format.

• SDK
• PROXY

from litellm import completion
import os, json

os.environ["SNOWFLAKE_API_KEY"] = "pat/<your-pat>"
os.environ["SNOWFLAKE_API_BASE"] = "https://<account>.snowflakecomputing.com/api/v2/cortex/v1"

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get the current weather for a location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {"type": "string", "description": "City name"}
                },
                "required": ["location"],
            },
        },
    }
]

response = completion(
    model="snowflake/claude-sonnet-4-5",
    messages=[{"role": "user", "content": "What's the weather in San Francisco?"}],
    tools=tools,
    tool_choice="auto",
)

print(response.choices[0].message.tool_calls)

Thinking / Reasoning​

Claude 3.7 Sonnet, Claude 4 Opus, and DeepSeek R1 on Cortex support extended thinking. LiteLLM translates reasoning_effort to the provider's thinking parameter.

from litellm import completion

response = completion(
    model="snowflake/claude-3-7-sonnet",
    messages=[{"role": "user", "content": "Solve: what is 127 * 389?"}],
    reasoning_effort="low",
)
print(response.choices[0].message.content)

Prompt Caching​

Snowflake Cortex supports prompt caching to reduce costs:

• OpenAI models: Implicit caching for prompts ≥ 1,024 tokens (no code changes needed)
• Claude models: Explicit caching via cache_control breakpoints

Cached input tokens are billed at 10% of the regular input rate (90% discount) when ≥ 1,024 tokens are cached.

See Cortex REST API Billing & Cost Analysis for details.

Embeddings​

from litellm import embedding
import os

os.environ["SNOWFLAKE_API_KEY"] = "pat/<your-pat>"
os.environ["SNOWFLAKE_API_BASE"] = "https://<account>.snowflakecomputing.com/api/v2/cortex/v1"

response = embedding(
    model="snowflake/snowflake-arctic-embed-l-v2.0",
    input=["Snowflake Cortex provides LLM inference"],
)
print(response.data[0]["embedding"][:5])

Supported Models​

All models are available through the snowflake/ prefix.

Chat Completion Models​

Embedding Models​