Kimi K3 model available in Public Preview in Snowflake Cortex AI

The kimi-k3 model is now part of the Snowflake Cortex AI ecosystem. It can be accessed through built-in AI functions or directly via the Cortex REST API.

Use cases

Imagine you need to integrate advanced language capabilities into your data pipelines or applications within Snowflake, using the kimi-k3 model to process text or handle complex prompts.

• Run inferences with the kimi-k3 model using Cortex SQL functions.
• Integrate the model via the Cortex REST API into external applications.
• Leverage cross-region inference support when required.

How to use

To use the model in SQL queries, you can invoke Cortex functions while configuring cross-region inference settings if your environment requires it.

SELECT SNOWFLAKE.CORTEX.COMPLETE(
  'kimi-k3',
  'Write a short summary about data engineering in tables.'
);

What happens in this code:

1. The SNOWFLAKE.CORTEX.COMPLETE function is called.
2. The kimi-k3 model identifier is passed as the first argument.
3. The prompt text is passed as the second argument to generate the response.

Limitations and requirements

This feature is in Public Preview. Using the model requires cross-region inference (CORTEX_ENABLED_CROSS_REGION) to be configured to values such as AWS_GLOBAL or ANY_REGION, depending on your cloud topology and region.