Snowflake SnowPro Specialty - Gen AI (GES - C01) Questions 2026 Snowflake SnowPro Specialty - Gen AI (GES - C01) Questions 2026 Contains 500+ exam questions to pass the exam in first attempt. SkillCertPro offers real exam questions for practice for all major IT certifications. • For a full set of 510 questions. Go to https://skillcertpro.com/product/snowflake - snowpro - specialty - gen - ai - ges - c01 - exam - questions/ • SkillCertPro offers detailed explanations to each question which helps to understand the concepts better. • It is recommended to score above 85% in SkillCertPro exams before attempting a real exam. • SkillCertPro updates exam questions every 2 weeks. • You will get life time access and life time free updates • SkillCertPro assures 100% pass guarantee in first attempt. Below are the free 10 sample questions. Question 1: A financial analytics team is developing an application to extract specific, structured financial data (e.g., company name, revenue, profit margin) from various news articles using Snowflake Cortex LLM functions. They require the output to strictly conform to a predefined JSON schema and want to ensure robust error handling. Which of the following statements are crucial considerations for achieving this goal? A.The AI_COMPLETE function should be used with the response_format argument, supplying a JSON schema object that defines the required structure, data types, and constraints for the output. B.To guarantee that critical fields like ‘company name‘ and ‘revenue‘ are always extracted, these properties must be explicitly listed within the ‘required‘ array of the JSON schema provided to AI_COMPLETE. C.For enhanced reliability in production pipelines, the team should wrap their AI_COMPLETE calls within TRY_COMPLE TE, as it returns a structured error object if the model fails to adhere to the schema, allowing for detailed debugging. D.Setting the temperature option to 0 in the AI_COMPLETE call is essential for obtaining the most consistent and accurate structured JSON outputs, regardless of task complexity or model used. E.The complexity of the JSON schema provided to AI_COMPLETE has no impact on compute costs, as only the input text and generated content tokens are billed. Answer: A, B and D Explanation: A. AI_COMPLETE Structured Outputs allows specifying a JSON schema via the response_format argument. AI_COMPLETE Structured Outputs allows you to specify a JSON schema using the response_format argument, ensuring responses adhere to a defined structure, data types, and constraints. B. Using the required field in the JSON schema ensures that specified properties are extracted. The required field ensures that the specified properties are included in the response. If they cannot be extracted, COMPLETE raises an error, making the extraction of critical information more reliable. D. Setting the temperature to 0 improves JSON adherence. For the most consistent results and optimal JSON schema adherence, it is recommended to set the temperature option to 0 when calling COMPLETE (or AI_COMPLETE). Incorrect: C. TRY_COMPLETE returns a structured error object for detailed debugging. TRY_COMPLETE performs the same operation as COMPLETE (or AI_COMPLETE) but returns NULL instead of raising an error when the operation c annot be performed. It does not return a structured error object for debugging. E. Schema complexity does not affect token usage or cost. The number of tokens processed (and billed) increases with schema complexity. Larger and more complex schemas consume more input and output tokens, resulting in higher compute costs. Question 2: A data scientist is implementing a Retrieval Augmented Generation (RAG) system in Snowflake for a legal document repository. They need to convert legal document chunks into vector embeddings and efficiently find the most relevant document chunks based on a user’s query. Which of the following statements accurately describe the process and best practices for creating and using these vector embeddings with Snowflake Cortex LLM functions? A.To create the embeddings for legal document chunks, the SNOWFLAKE.CORTEX.EMBED_TEXT_768 function should be used, specifying an appropriate model like ‘snowflake - arctic - embed - m‘. B.When comparing a user‘s query embedding with document embeddings, the VECTOR_L1_DISTANCE function is generally recommended over VECTOR_COSINE_SIMILARITY for RAG applications to ensure optimal semantic relevance. C.For best search results in RAG, legal document text should be split into chunks of no more than 512 tokens, as sma ller chunks typically lead to higher retrieval quality. D.Embedding functions like EMBED_TEXT_768 and EMBED_TEXT_1024 incur compute costs based on both input and output tokens, similar to how AI_COMPLETE is billed. E.The VECTOR data type in Snowflake only supports integer elements, meaning legal document embeddings must be converted to integer vectors before storage. Answer: A and C Explanation: Option A. To create the embeddings for legal document chunks, the SNOWFLAKE.CORTEX.EMBED_TEXT_768 function should be used, specifying an appropriate model like ‘snowflake - arctic - embed - m’. This is correct because EMBED_TEXT_768 (or similar embedding functions) generates vector embeddings from text. Using a model like snowflake - arctic - embed - m ensures semantic representation of legal text chunks, enabling efficient similarity search. Option C. For best search results in RAG, legal document text should be sp lit into chunks of no more than 512 tokens, as smaller chunks typically lead to higher retrieval quality. This is correct because chunking text into manageable sizes improves retrieval accuracy. Smaller chunks reduce noise and allow embeddings to capture more precise semantic meaning, which is critical in legal contexts. Incorrect Option B. When comparing a user’s query embedding with document embeddings, the VECTOR_L1_DISTANCE function is generally recommended over VECTOR_COSINE_SIMILARITY for RAG appl ications to ensure optimal semantic relevance. Incorrect because cosine similarity is the standard metric for semantic relevance in embedding ‑ based retrieval. L1 distance (Manhattan distance) is not optimal for capturing semantic similarity in high ‑ dimensional vector spaces. Option D. Embedding functions like EMBED_TEXT_768 and EMBED_TEXT_1024 incur compute costs based on both input and output tokens, similar to how AI_COMPLETE is billed. Incorrect because embedding functions are billed based on input to kens only, not output tokens. Unlike generative functions (e.g., AI_COMPLETE), embeddings produce fixed ‑ size vectors, so output token billing does not apply. Option E. The VECTOR data type in Snowflake only supports integer elements, meaning legal document embeddings must be converted to integer vectors before storage. Incorrect because Snowflake’s VECTOR data type supports floating ‑ point values, which are required for embeddings. Converting embeddings to integers would destroy semantic meaning and make similarity search ineffective. Question 3 : A Snowflake user attempts to access the Cortex Playground to experiment with LLM functions but receives an error indicating insufficient privileges. They have an active Snowflake session. Which of the following steps must be taken to ensure they can successfully use the Cortex Playground? A.The user‘s role must be granted the SNOWFLAKE.CORTEX_USER database role, which includes the necessary privileges to call Snowflake Cortex LLM functions. B.The CORTEX_MODELS_ALLOWLIST account parameter must be explicitly set to include the desired LLM, even if it is a Snowflake - hosted model. C.The user must first create a dedicated compute pool of instance family GPU_NV_M specifically for Cortex Playground operations. D.The CORTEX_ENABLED_CROSS_REGION parameter must be set to ANY_REGION if the user‘s account is not in one of the natively supported regions for Cortex LLM functions. E.The user needs to manually install the TruLens SDK in their environment, as it is a prerequisite for all Cortex AI Studio features. Answer: A Explanation: A. The user‘s role must be granted the SNOWFLAKE.CORTEX_USER database role... To access Snowflake Cortex AI features (including the graphical Cortex Playground in Snowsight), a user's active role requires specific permissions. By default, access to Snowf lake Cortex AI is granted to the PUBLIC role via the SNOWFLAKE.CORTEX_USER database role. If an administrator locks down governance by revoking this broad access, users will hit privilege errors. To fix this, an administrative role must explicitly grant the SNOWFLAKE.CORTEX_USER database role to the custom role the user is actively leveraging. Incorrect: B. The CORTEX_MODELS_ALLOWLIST account parameter must be explicitly set... CORTEX_MODELS_ALLOWLIST is a governance parameter used by administrators to restrict or narrow down which specific LLMs can be utilized within an account. It is not required to be explicitly set just to use the playground natively; by default, if it is unset, all standard Snowflake - hosted models available in that region are accessible. C. The user must first create a dedicated compute pool... Snowflake Cortex AI functions and the Cortex Playground run as fully - managed serverless features. They do not require the user to configure or pay for custom Snowpark Container Services (S PCS) compute pools or dedicated GPU instances. The system relies entirely on standard virtual warehouses to process metadata queries and the backend serverless engine for inference. D. The CORTEX_ENABLED_CROSS_REGION parameter must be set to ANY_REGION... While cross - region parameters exist for multi - region configurations, the Cortex Playground dropdown only dynamically exposes models that are natively supported and active within the local region of the specific Snowflake account being used. You cannot ov erride the playground's UI restrictions by changing cross - region settings. E. The user needs to manually install the TruLens SDK... While TruLens is integrated with Snowflake's AI observability capabilities for programmatic evaluation workflows, it is an external SDK and is absolutely not a prerequisite for opening or using the built - in Cortex Playground UI inside Snowsight. Question 4 : A data analyst is using Snowflake Copilot in Snowsight to generate SQL queries for a new dataset containing customer PII. Which of the following statements accurately describes how Snowflake Copilot operates with respect to data access, governance, and model interaction? A.Snowflake Copilot directly accesses and processes the raw data within customer tables to understand its content and generate SQL. B.Snowflake Copilot is powered by a fine - tuned model that runs securely inside Snowflake Cortex, leveraging only database/schema/table/column names and data types, ensuring data remains within Snowflake‘s governance boundary and respects RBAC. C.To protect sensitive information, Snowflake Copilot transmits sampled PII data to an external LLM for schema understanding before generating SQL. D.Snowflake Copilot requires explicit column - level grants for direct data access, similar to how a human analyst would query specific data points. E.While Snowflake Copilot generates SQL based on metadata, the generated SQL queries are executed in an isolated environment that does not respect existing Snowflake RBAC policies. Answer: B Explanation: B. Snowflake Copilot does not have access to the data inside your tables. Snowflake Copilot is an LLM - powered assistant that uses a model fine - tuned by Snowflake and runs securely within Snowflake Cortex. It ensures that your enterprise data and metadata remain securely inside Snowflake while fully respecting role - based access control (RBAC). Importantly, Snowflake Copilot does not have access to the data stored inside your tables. Instead, it generates responses based on metadata such as the names of databases, schemas, tables, columns, and their data types. Incorrect Options: A. Copilot does not directly access customer data. It relies only on metadata to generate responses. C. Copilot does not transmit customer data outside of Snowflake, ensuring enterprise data remains secure. D. This option incorrectly implies that Copilot directly accesses table data, which it does not. E. Copilot fully integrates with and respects Snowflake's RBAC policies; it does not bypass or ignore access controls. Question 5 : A team of data application developers is leveraging Snowflake Copilot to streamline the creation of analytical SQL queries within their Streamlit in Snowflake application. They observe that Copilot sometimes struggles with complex joins or provides suboptimal queries when dealing with a newly integrated, deeply nested dataset. Based on Snowflake‘s best practices and known limitations, which actions or considerations would help improve Copilot‘s performance in this scenario? A.Implement curated views with descriptive and easy - to - understand names for views and columns, appropriate data types, and pre - define common/complex joins to simplify the underlying schema for Copilot. B.Enable the CORTEX_MODELS_ALLOWLIST parameter to restrict Copilot to only use the largest available LLMs, thereby guaranteeing higher accuracy for complex queries. C.Break down complex requests into simpler, multi - turn questions, as Copilot is designed to build complex queries through conversational refine ment and follow - up questions. D.Grant Copilot direct access to the raw data using ACCOUNTADMIN privileges, allowing it to infer schema relationships more effectively from data content. E.Ensure that a database and schema are explicitly selected for the current session, and that column names are meaningful, to provide Copilot with better context for query generation. Answer: A, C Explanation: To improve Snowflake Copilot's performance, it is a best practice to create curated views with descriptive names, appropriate data types, and predefined common or complex joins. Copilot can also build complex SQL queries through a conversation by asking follow - up questions. Additionally, it relies on the names of databases, schemas, tables, columns, and their data types to understand the available metadata, so using meaningful object names and ensuring the correct session context are essential for generat ing accurate and relevant responses. Incorrect: B. Configuring CORTEX_MODELS_ALLOWLIST The CORTEX_MODELS_ALLOWLIST parameter controls access to specific large language models (LLMs), but it does not improve the accuracy of Snowflake Copilot's SQL generation. D. Granting ACCOUNTADMIN privileges Snowflake Copilot does not have access to the data inside tables; it operates solely on metadata. Granting ACCOUNTADMIN privileges does not change this behavior and violates the principle of least privilege. • For a full set of 510 questions. Go to https://skillcertpro.com/product/snowflake - snowpro - specialty - gen - ai - ges - c01 - exam - questions/ • SkillCertPro offers detailed explanations to each question which helps to understand the concepts better. • It is recommended to score above 85% in SkillCertPro exams before attempting a real exam. • SkillCertPro updates exam questions every 2 weeks. • You will get life time access and life time free updates • SkillCertPro assures 100% pass guarantee in first attempt. Question 6 : A data science team is fine - tuning a mistral - 7b model within Snowflake Cortex using proprietary customer interaction logs. Which of the following principles and practices apply to this fine - tuning process concerning data privacy, model ownership, and subsequent inference? A.The proprietary customer interaction logs used for fine - tuning are leveraged by Snowflake to improve the base mistral - 7b model for all customers. B.The resulting fine - tuned model is exclusively available to the data science team and cannot be accessed by other Snowflake customers. C.The fine - tuning process occurs entirely within Snowflake‘s security and governance boundaries, ensuring the data never leaves the Snowflake environment. D.The fine - tuned model, which is a CORTEX_FINETUNED type, can be shared with other Snowflake accounts using secure data sharing. E.The fine - tuned model is fully managed by the Snowflake Model Registry API, allowing programmatic management of its lifecycle. Answer: B, C and D Explanation: Cortex Fine - tuning is a fully managed service that allows you to fine - tune popular large language models (LLMs) using your own data, entirely within Snowflake. Your usage data and customer data (including inputs and outputs) are not used to train, retrain , or fine - tune models made available to other customers. Fine - tuned models created using your data are exclusive to your organization. Correct Options: B. Fine - tuned models are exclusive to the customer who created them. C. The entire fine - tuning process is managed securely within Snowflake's environment. D. Models created with Cortex Fine - tuning (of type CORTEX_FINETUNED) can be shared using Data Sharing. Incorrect Options: A. Your usage and customer data are not used to train or improve models for o ther customers. E. Although Cortex Fine - Tuned LLMs appear in the Snowsight Model Registry UI, they are not managed through the Model Registry API. Question 7 : A data scientist fine - tuned a mistral - 7b model in Snowflake for a specific customer support response generation task, naming it my_custom_responder_model. They now want to make this model available for AI_COMPLETE calls in production, ensuring proper acce ss control and regional availability. Which of the following statements is true regarding the deployment and management of this fine - tuned model in Snowflake? A.To make my_custom_responder_model available for inference, the data scientist must register it in the Snowflake Model Registry using the snowflake.ml.registry.Registry().log_model() Python API, as it is a custom model. B.If my_custom_responder_model was fine - tuned in AWS US West 2 (Oregon) but needs to be used for inference in an Azure East US 2 (Virginia) account, cross - region inference will be automatically handled by Snowflake Cortex without any explicit parameter configuration. C.Administrators can restrict access to my_custom_responder_model and other specific LLMs within the account by setting the CORTEX_MODELS_ALLOWLIST account parameter, which controls model availability for Cortex LLM functions. D.Any user role calling AI_COMPLETE with my_custo m_responder_model must have the SNOWFLAKE.DOCUMENT_INTELLIGENCE_CREATOR database role granted to access fine - tuned models. E.To update my_custom_responder_model with new training data and improve its performance, the existing model can be directly updated using an ALTER SNOWFLAKE.CORTEX.FINETUNE ‘UPDATE‘ SQL command, specifying the new training data. Answer: C Explanation: The CORTEX_MODELS_ALLOWLIST account parameter can be used by administrators to restrict access to specific large language models (LLMs), including custom fine - tuned models, for use with Cortex LLM functions such as COMPLETE and TRY_COMPLETE. Incorrect Options: A. Although Cortex Fine - Tuned LLMs appear in the Snowsight Model Registry UI, they are not managed through the Model Registry API. Fine - tuned models are created using the FINETUNE function. B. Cross - region inference for Cortex LLM functions, inclu ding fine - tuned models, is not always automatic. If the required inference region is not natively supported, the CORTEX_ENABLED_CROSS_REGION parameter must be explicitly configured. D. To use Cortex LLM functions such as AI_COMPLETE, user roles must be granted the SNOWFLAKE.CORTEX_USER database role, not the SNOWFLAKE.DOCUMENT_INTELLIGENCE_CREATOR role. E. Fine - tuned models in Snowflake Cortex are immutable. To modify a model, it must be deleted, and a new fine - tuning job must be created and trained. Question 8 : A Gen AI Specialist in Snowflake Cortex aims to fine - tune an LLM for enhanced task - specific performance. When creating a fine - tuning job using SNOWFLAKE.CORTEX.FINETUNE(‘CREATE‘, ...), which statement accurately describes the required training data format and a supported base model, aligning with Snowflake‘s Gen AI principles for leveraging LLMs? A.Training data must contain input and output columns, and the llama4 - maverick model is a supported base model for fine - tuning. B.Training data must exclusively contain JSON objects, and the claude - 4 - opus model is a supported base model for fine - tuning. C.Training data can be any format as long as it‘s a Snowflake table, and openai - gpt - 4.1 is a supported base model for fine - tuning. D.The training data query must return a single text column that will be automatically split into prompts and completions, and snowflake - arctic is a supported base model for fine - tuning. E.The SQL query for training data must return prompt and completion columns, and mistral - 7b is a supported base model for fine - tuning. Answer: E Explanation: E. The SQL query for training data must return prompt and completion columns, and mistral - 7b is a supported base model for fine - tuning. According to Snowflake Cortex documentation, the training_data_query provided to SNOWFLAKE.CORTEX.FINETUNE('CREATE', . ..) must evaluate to a result set containing exactly two specifically named columns: prompt (the input context provided to the model) and completion (the expected ideal text output). Furthermore, mistral - 7b is natively listed as a supported foundational base model for Cortex's fully - managed parameter - efficient fine - tuning (PEFT) framework, making this statement entirely accurate. Incorrect: A. Training data must contain input and output columns, and the llama4 - maverick model is a supported base model... While conceptually representing inputs and outputs, the columns must be literally aliased or named as prompt and completion (not input and output). Additionally, llama4 - maverick is not a valid, supported Snowflake Cortex model. C. Training data can be any format as long as it‘s a Snowflake table, and openai - gpt - 4.1 is a supported base model... The training data cannot be in "any format" — it strictly requires a text - based format structured with designated prompt and completion columns. Additionally, Snowfla ke Cortex hosts open - weight foundation models locally within its secure infrastructure boundary; it does not support fine - tuning proprietary external models like OpenAI's gpt - 4 series via this function. D. The training data query must return a single text column... and snowflake - arctic is a supported base model... A single text column will cause an immediate query error, as the backend cannot automatically deduce where the prompt ends and the completion begins. Moreover, while snowflake - arctic is a founda tional model developed by Snowflake, it is utilized for general inference/embedding tasks and is not a supported base model for the managed CORTEX.FINETUNE engine (which prioritizes models like mistral - 7b or llama3 - 8b). B. Training data must exclusively contain JSON objects, and the claude - 4 - opus model is a supported base model... The function expects an executed SQL query returning regular relational columns (prompt, completion), not raw text - serialized JSON files. Additionally, claude - 4 - opus is an exter nal, proprietary model from Anthropic and is not supported for fine - tuning within Snowflake Cortex. Question 9 : A data application developer, adhering to Snowflake‘s Gen AI best practices for deploying LLMs, needs to perform inference with a newly fine - tuned llama3.1 - 70b model via AI_COMPLETE and expects a structured JSON output. Which of the following statements accurately describe how to configure this inference and potential limitations within Snowflake Cortex? A.The fine - tuned model can be called directly by its name in AI_COMPLETE, but structured output requires a separate post - processing step outside of Snowflake. B.Structured outputs with AI_COMPLETE can be achieved by specifying a JSON schema in the response_format argument. When using OpenAI (GPT) models, the additionalProperties field must be set to false in every node of the schema. C.The developer should use TRY_COMPLETE instead of AI_COMPLETE to ensure structured output. Cortex LLM functions, including those for fine - tuned models, do not support dynamic tables. D.For consistent results with structured outputs, the temperature option in AI_COMPLETE should be set to 0. The schema definition should place object definitions under the $defs key for strict enforcement by Snowflake‘s validation mechanism. E.Fine - tuned models from Cortex Fine - tuning are managed by the Snowflake Model Registry API, and their inference can be monitored through the CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY view. Answer: B and D Explanation: A — Incorrect. AI_COMPLETE can call fine ‑ tuned models directly by name and can return structured objects when you supply a response_format JSON schema to AI_COMPLETE, so you do not necessarily need an external post ‑ processing step to get structured output from within Snowflake.docs.snowflake B — Correct (with detail). AI_COMPLETE supports a response_format parameter that accepts a JSON schema to request structured outputs, which lets the function return an object following that format when used correctly . The documentation also notes model ‑ specific requirements for schema enforcement; for OpenAI (GPT) style models some schema characteristics (for example, controlling extra properties) are required to get strict validation, and using additionalProperties=false is a way to disallow unexpected fields in schema nodes. This option correctly calls out both response_format usage and the need to restrict additionalProperties for strict schema enforcement with OpenAI/GPT models.docs.snowflake C — Incorrect. TRY_ COMPLETE is a different testing/debug helper and is not required (nor generally recommended) to obtain structured outputs; AI_COMPLETE is the supported Cortex LLM function for producing completions and accepts response_format for structured JSON. Also, Cortex LLM functions currently have limited support for dynamic tables (and behavior depends on the function and model), but the blanket statement “do not support dynamic tables” is inaccurate as phrased.docs.snowflake D — Mostly correct but with nuance. S etting temperature to 0 improves determinism and is recommended when you need consistent structured outputs from AI_COMPLETE, and Snowflake’s response_format JSON schema enforcement expects schema definitions in standard JSON Schema constructs; placing reusable object definitions under $defs is the correct JSON ‑ Schema pattern for definitions and helps strict validation. However, whether Snowflake strictly enforces $defs vs inline schemas can depend on the model and guardrails settings, so the statement is generally accurate as guidance but remember enforcement depends on model capabilities and parameters.docs.snowflake E — Partially correct. Fine ‑ tuned models created and managed via Snowflake Cortex Fine ‑ tuning are tracked and accessible via Snowflake ’ s model/registry features, and usage monitoring for Cortex functions is available through account usage views such as CORTEX_FUNCTIONS_USAGE_HISTORY (note: documentation shows this view was GA and that account ‑ usage monitoring of Cortex functions exists), b ut the exact API/registry names and which account_usage view to query can vary over time (Snowflake also documents CORTEX_AISQL_USAGE_HISTORY as current in some places). So the option is broadly correct that fine ‑ tuned models are managed and their inference usage can be monitored via the Cortex usage views, but check the current Account Usage view name and notes in the docs for the latest recommended view.docs.snowflake+1 Question 10 : A data platform administrator needs to retrieve a consolidated overview of credit consumption for all Snowflake Cortex AI functions (e.g., LLM functions, Document AI, Cortex Search) across their entire account for the past week. They are interested in the aggregated daily credit usage rather than specific token counts per query. Which Snowflake account usage views should the administrator primarily leverage to gather this information? A.The SNOWFLAKE.ACCOUNT_USAGE.CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY view to get detailed token usage for each LLM function call, then aggregate manually. B.The SNOWFLAKE.CORTEX.COUNT_TOKENS function to re - calculate estimated costs for all past queries that used Cortex AI functions. C.The SNOWFLAKE.ACCOUNT_USAGE.CORTEX_DOCUMENT_PROCESSING_USAGE_HIST ORY view for Document AI costs, and SNOWFLAKE.ACCOUNT_USAGE.CORTEX_SEARCH_DAILY_USAGE_HISTORY for Cortex Search costs, then combine them. D.Only the SNOWFLAKE.AC COUNT_USAGE.QUERY_HISTORY view, analyzing the EXECUTION_STATUS and TOTAL_ELAPSED_TIME columns for queries involving Cortex functions. E.The SNOWFLAKE.ORGANIZATION_USAGE.METERING_DAILY_HISTORY view, specifically filtering for SERVICE_TYPE = ‘AI_SERVICES‘. Answer: E Explanation: E. The SNOWFLAKE.ORGANIZATION_USAGE.METERING_DAILY_HISTORY view provides daily credit usage for an account. By filtering on SERVICE_TYPE = 'AI_SERVICES', administrators can view aggregated credit consumption for all AI services, including Cortex LLM Funct ions, Cortex Analyst, and Document AI. This provides the consolidated daily overview requested. Incorrect: A. CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY This view provides detailed, per - query token usage for Cortex LLM functions. While useful for granular analysis, it does not provide an aggregated daily view across all AI services. C. AI service - specific usage views Although these views track usage for specific AI services (such as Document AI and Cortex Search), METERING_DAILY_HISTORY filtered by SERVICE_TY PE = 'AI_SERVICES' offers a broader, consolidated view across all AI services, making it a more efficient solution for this requirement. D. QUERY_HISTORY QUERY_HISTORY focuses on general query execution details and does not directly report AI service - specific credit consumption. B. COUNT_TOKENS The COUNT_TOKENS function is used to estimate token usage and potential costs before execution. It is not intended for analyzing historical AI service costs or usage. • For a full set of 510 questions. Go to https://skillcertpro.com/product/snowflake - snowpro - specialty - gen - ai - ges - c01 - exam - questions/ • SkillCertPro offers detailed explanations to each question which helps to understand the concepts better. • It is recommended to score above 85% in SkillCertPro exams before attempting a real exam. • SkillCertPro updates exam questions every 2 weeks. • You will get life time access and life time free updates • SkillCertPro assures 100% pass guarantee in first attempt.