arctic-extract

May 22, 2026: AI_EXTRACT extraction scores (General availability)

Extraction scores for the Snowflake Cortex AI_EXTRACT function are now generally available. Key capabilities include:

• Confidence scores for extracted values: Pass scores => TRUE in an AI_EXTRACT call to get back a scoring object alongside the standard response object, with a score between 0 and 1 indicating the model's certainty about each extracted value.
• Support for entities, lists, and tables: Per-field scores are returned for entity extraction, and aggregate scores are returned for list and table extraction.
• Support for fine-tuned arctic-extract models: Extraction scores work with fine-tuned arctic-extract models in addition to the default model.

Read from the provider's text; the quote is the provider's exact lines.

Apr 20, 2026: Fine-tuning arctic-extract models (General availability)

Fine-tuning arctic-extract models is now generally available. You can adapt the model to your documents and domains for more consistent structured extraction, and run inference with AI_EXTRACT using your fine-tuned model.

Read from the provider's text; the quote is the provider's exact lines.

Jan 28, 2026: Fine-tuning arctic-extract models (Preview)

Snowflake Cortex now supports fine-tuning arctic-extract models to improve document extraction accuracy for your specific formats and domains. Fine-tuning adapts the model to your data, delivering more consistent, structured outputs than zero-shot extraction. Fine-tuned models integrate seamlessly with AI_EXTRACT for production workflows.

Read from the provider's text; the quote is the provider's exact lines.