SnowPro Specialty: Gen AI Question 218
Single answerLimits on which models can be usedA financial services company uses Snowflake Cortex to let analysts summarize customer support cases. The security team has approved only a subset of foundation models for production due to data governance requirements. An engineer creates a SQL workflow that calls AI_COMPLETE with a model name that is available in Snowflake documentation, but the query fails in the production account. What is the most likely explanation, and what should the engineer do next?
- A
The model may exist in Snowflake generally but is not allowed for that account, region, or feature context; the engineer should verify the models actually available to their environment and switch to an approved supported model.
- B
AI_COMPLETE accepts any model listed by the model provider as long as the prompt is under the token limit; the failure is most likely caused only by prompt size.
- C
If a model appears in Snowflake documentation, it is automatically usable in every Snowflake account; the engineer should retry the statement with a larger warehouse.
- D
Model usage restrictions apply only to Snowsight playground features, not to SQL functions like AI_COMPLETE; the engineer should grant broader warehouse privileges to the role.
Show answer and explanation
Correct answer: A
Explanation
This question tests whether the candidate understands that model choice in Snowflake Cortex is constrained by Snowflake-supported availability, which can vary by account, region, cloud, and feature. In real deployments, teams also layer internal governance rules on top of Snowflake's supported model list. Therefore, seeing a model in documentation is not enough to assume it can be used in a given production environment. Best practice is to validate the models available for the specific Cortex function and deployment environment, then standardize on an approved supported model. This aligns with Snowflake guidance that Cortex model support is environment-dependent rather than universally identical across all accounts.
- A. Correct.
Correct. In practice, the set of models that can be used is constrained by what Snowflake makes available for a given account, region, cloud, and specific Cortex feature. A model documented by Snowflake is not necessarily available everywhere or enabled for every usage context. The right next step is to check which models are supported in that environment and use one that is both available and approved by the organization.
- B. Incorrect.
Incorrect. Token or prompt-size limits can cause failures, but the scenario specifically points to a model name issue in production despite the model being documented. A common misconception is that provider availability alone determines whether AI_COMPLETE can use a model. In Snowflake, model availability is gated by Snowflake-supported model access and environment-specific support, not just provider documentation.
- C. Incorrect.
Incorrect. Snowflake documentation may list models supported across the platform, but that does not mean every listed model is usable in every account or region. Increasing warehouse size does not solve a model availability or authorization/support restriction. Warehouses affect compute for many workloads, but they do not override model support constraints.
- D. Incorrect.
Incorrect. Restrictions on usable models are not limited to playground-style interfaces. SQL-based Cortex functions also depend on supported and available models. Granting more warehouse privileges would not make an unsupported or unavailable model callable through AI_COMPLETE. This distractor reflects confusion between compute permissions and model access/support.