Snowflake Cortex meters serverless foundation model usage using token-based billing metrics. The accounting methodology distinguishes between functions that generate new textual responses and functions that output non-generative scalar values or dense embeddings:
Dual-Token Accounting (Input + Output): Functions that process an incoming prompt and generate variable-length natural language text bill for both the tokens ingested (input) and the completion tokens generated (output). This category includes generative task-specific functions such as SUMMARIZE, TRANSLATE, and CLASSIFY_TEXT (alongside general COMPLETE invocations).
Input-Only Token Accounting: Embedding generation functions—including EMBED_TEXT_768 and EMBED_TEXT_1024—only meter input tokens, as their output is a fixed-dimension vector rather than a generated natural language token stream.
Alternative Billing Models: SENTIMENT evaluates text to return a single floating-point scalar score and is billed according to specialized execution consumption rates rather than standard generative input/output token accounting.
References & Study Guide Topics:
Snowflake Gen AI Functions: Consumption Metering and Credit Computation
Token Accounting Framework: Input vs. Output Token Sizing across Cortex NLP Utilities
===============
Contribute your Thoughts:
Chosen Answer:
This is a voting comment (?). You can switch to a simple comment. It is better to Upvote an existing comment if you don't have anything to add.
Submit