Snowflake · GES-C01
SnowPro Specialty: Gen AI (GES-C01) Practice Test
Validates expertise in applying generative AI and large language model (LLM) capabilities within the Snowflake platform. Assesses the ability to build, govern, and scale Gen AI solutions using Snowflake Cortex, Document AI, and related features.
Exam Details
Questions
600
Duration
85 minutes
Passing Score
750/1000
Difficulty
SpecialtyLast Updated
Jun 2026
Topics Covered
Snowflake for Gen AI OverviewSnowflake Gen AI & LLM FunctionsSnowflake Cortex LLMs and Cortex SearchSnowflake Gen AI GovernanceDocument AIModel Registry and SnowparkCortex Analyst and Agents
SnowPro Specialty: Gen AI (GES-C01) Practice Exam Preparation
Use this GES-C01 practice exam to prepare for SnowPro Specialty: Gen AI (GES-C01) with realistic questions, detailed explanations, and focused study modes. The practice bank includes 600 questions for Snowflake GES-C01, so you can review the exam steadily instead of relying on one long cram session.
As you practice, pay extra attention to recurring topics such as Snowflake for Gen AI Overview, Snowflake Gen AI & LLM Functions, Snowflake Cortex LLMs and Cortex Search, Snowflake Gen AI Governance, and Document AI. Start with short sessions to identify weak areas, then move into timed quizzes once your accuracy is consistent.
The explanations are especially useful when you want to connect exam wording to the responsibilities and scenarios described in the official certification guidance. Use the free preview first, then unlock the full question bank when you are ready to build a complete study routine.
Exam Domain Breakdown
Snowflake for Gen AI Overview26%
Snowflake Gen AI & LLM Functions40%
Snowflake Gen AI Governance22%
Document AI12%
Exam Overview
The SnowPro Specialty: Gen AI (GES-C01) certification validates specialized knowledge, skills, and best practices for leveraging generative AI and large language model (LLM) capabilities within the Snowflake Data Cloud. It assesses a candidate's ability to implement Snowflake Cortex AI features—including Cortex LLMs, Cortex Search, Cortex Analyst, and Cortex Agents—alongside Document AI, the Snowflake Model Registry, and Snowpark Container Services to build, govern, and scale production-grade Gen AI solutions.
The exam spans the full Gen AI development lifecycle on Snowflake: from applying task-specific and general-purpose LLM functions, constructing retrieval-augmented generation (RAG) pipelines and text-to-SQL workflows, to enforcing AI governance through access controls, guardrails, cost monitoring, and observability tooling. Candidates are also expected to understand how to run open-source and third-party models within the Snowflake environment. Note that GES-C01 is in the process of being retired in favor of its successor, GES-C02; candidates should review Snowflake's certification FAQs for transition details.
Who Should Take This Exam
This certification is designed for practitioners who build and operationalize AI-powered data applications on Snowflake. Primary target roles include AI/ML engineers constructing intelligent data pipelines, data scientists productionizing machine learning models, data engineers integrating Gen AI workflows into existing Snowflake architectures, and application developers adding AI-powered features to enterprise products.
Candidates are expected to have at least one year of hands-on Gen AI experience with Snowflake in an enterprise environment, along with foundational data engineering skills and SQL proficiency. Python coding experience is strongly beneficial, particularly for working with Snowpark and the Model Registry.
Prerequisites
Snowflake recommends that candidates hold an active SnowPro Associate: Platform or SnowPro Core certification before attempting GES-C01. This ensures a baseline understanding of Snowflake's core architecture, data loading, transformation patterns, and access control model before tackling Gen AI-specific content.
Beyond the certification prerequisite, candidates should have at least one year of practical Gen AI experience within Snowflake's enterprise environment. Proficiency in SQL and data engineering principles is assumed throughout the exam, and familiarity with Python is strongly advised for domains covering Snowpark, the Model Registry, and container-based model deployment.
Exam Format
The GES-C01 exam consists of 55 scored questions delivered in 85 minutes, yielding a per-question pace of roughly 90 seconds. Question types include multiple-choice, multiple-select, and scenario-based items that test applied judgment rather than rote recall. The exam is administered through Pearson VUE and is available as an online proctored or in-person test center experience.
Scoring is on a scale of 0–1000, with a passing score of 750 required. The exam fee is $225 USD (regional pricing variations may apply). The resulting certification is valid for two years, after which recertification is required. No unscored pilot questions have been publicly disclosed for this exam.
Skills Measured
- 1.Domain 1 – Snowflake for Gen AI Overview (26%): Covers foundational Gen AI principles on Snowflake, the Cortex platform ecosystem (LLMs, Search, Analyst, Fine-tuning, Agents, Copilot), security and privacy controls, role-based access control (RBAC) applied to AI resources, and available interfaces including SQL, REST API, and the Snowflake AI playground.
- 2.Domain 2 – Snowflake Gen AI & LLM Functions (40%): The most heavily weighted domain; focuses on applying Cortex general-purpose and task-specific LLM functions, vector operations and embeddings, helper functions, building RAG pipelines, text-to-SQL with Cortex Analyst, constructing chat interfaces, integrating Gen AI functions into data pipelines, and running third-party or open-source models via Snowpark Container Services and the Model Registry.
- 3.Domain 3 – Snowflake Gen AI Governance (22%): Emphasizes model access controls and policy enforcement, implementing AI guardrails, cost optimization strategies, monitoring LLM consumption via the CORTEX_FUNCTIONS_USAGE_HISTORY view, and leveraging AI observability tooling to audit and trace model behavior.
- 4.Domain 4 – Document AI (12%): Covers Snowflake Document AI capabilities for extracting structured information from unstructured documents, integrating document processing into Snowflake workflows, and applying governance controls to document-based AI pipelines.
Study Tips
- Download and read the official SnowPro Specialty: Gen AI Study Guide (PDF) from learn.snowflake.com before doing anything else—it defines the exact domain boundaries and lists the specific Cortex features, functions, and views you are expected to know.
- Prioritize Domain 2 (Snowflake Gen AI & LLM Functions, 40%) by completing hands-on labs in a Snowflake trial account: practice calling COMPLETE, SUMMARIZE, EXTRACT_ANSWER, SENTIMENT, and TRANSLATE functions directly in SQL, and build at least one end-to-end RAG pipeline using Cortex Search.
- For Domain 3 governance, study the CORTEX_FUNCTIONS_USAGE_HISTORY view in the Snowflake documentation and practice writing queries against it; also review how network rules, masking policies, and RBAC roles apply specifically to Cortex AI features.
- Use Snowflake University courses and the free-to-audit Snowflake courses on Coursera to cover Cortex Analyst, Cortex Agents, and the Model Registry—these newer features appear in both Domain 1 and Domain 2 and are underrepresented in third-party prep materials.
- For Document AI (Domain 4), work through Snowflake's official Quickstart guides on Document AI to understand how to configure build and query models for document extraction, since this domain's concepts are not covered by most generic ML study materials.
- Practice scenario-based questions by reviewing Snowflake developer blog posts and solution accelerators that show real enterprise Gen AI architectures—the exam favors applied judgment over memorization of parameter names.
- In the final week before the exam, take at least two full-length timed practice exams to calibrate your 85-minute pacing; flag any questions touching Cortex Fine-tuning or Snowpark Container Services for focused review, as these are frequently cited as weak spots.
Career Benefits
Earning the GES-C01 positions professionals at the intersection of enterprise data engineering and applied AI, a combination in high demand as organizations move from LLM experimentation to production deployment. Roles commonly associated with this certification include AI/ML Engineer, Data Science Engineer, Snowflake Platform Architect, and AI Solutions Engineer—all of which command strong compensation. Entry-level professionals with this credential can expect salaries in the $120,000–$155,000 range annually, mid-level specialists $155,000–$210,000, and senior practitioners $210,000 or higher, with top figures concentrated in major tech hubs such as San Francisco and Seattle.
Beyond compensation, the certification demonstrates platform-specific mastery that distinguishes candidates from those holding general cloud ML certifications (such as AWS MLS-C01 or Google Professional ML Engineer), as it validates hands-on ability with Snowflake's proprietary Cortex stack rather than generic ML frameworks. As Snowflake continues to expand its AI product surface—including Cortex Agents and fine-tuning—professionals certified on the platform are well positioned to lead Gen AI adoption initiatives within data-heavy enterprises.
Sample Questions
5 sample questions with answers and explanations. Start a practice session to test yourself across all 600 questions.
Preview — answers shown1. A platform engineer at Woodgrove Enterprises is deploying a Cortex Agent to automate multi-step employee onboarding workflows involving document retrieval, structured data lookup, and form generation. Leadership is concerned about AI tasks consuming unpredictable amounts of compute and potentially running indefinitely during high-load periods. Which configuration capability built into Cortex Agents most directly addresses this operational concern? (Select one!)
AConfiguring an account-level resource monitor with a credit quota and threshold alerts covering all Cortex function usage across the account
BConfiguring the agent's budget with a time limit and a token limit so the agent automatically stops when either threshold is reached
CSetting a STATEMENT_TIMEOUT_IN_SECONDS parameter on the virtual warehouse where the agent executes to cap the total session duration
DSetting the max_tokens parameter within each individual COMPLETE function call inside the agent's tool definitions to cap the length of each response
Explanation
Cortex Agents natively support budget configuration that includes both a time limit and a token limit. The agent monitors both thresholds throughout its multi-step execution and automatically stops when either is reached, preventing runaway processing during high-load periods without requiring external intervention. This is a built-in lifecycle control mechanism within the Cortex Agent framework. Setting STATEMENT_TIMEOUT_IN_SECONDS on the warehouse controls individual SQL statement execution time, not the full lifecycle of a multi-step agent workflow that may span many sequential tool calls. Account-level resource monitors track credit consumption and send alerts but do not directly halt an in-progress agent execution when a threshold is crossed mid-workflow. Setting max_tokens within individual COMPLETE calls limits the token length of each single LLM response but does not constrain the total number of tool invocations or the cumulative duration of the agent's end-to-end execution.
2. A data engineering team at Proseware Analytics has a Snowflake account deployed in AWS ap-southeast-1 (Singapore). The mistral-large2 model is not available in their region, so they enable cross-region inference. When they execute a batch of SNOWFLAKE.CORTEX.COMPLETE calls, Snowflake routes the inference to AWS us-east-1 where the model is available. How are Snowflake credits charged for this cross-region processing? (Select one!)
ACredits are charged at a premium cross-region surcharge rate applied on top of standard per-token inference costs
BCredits are split equally between the requesting region and the processing region
CCredits are consumed in AWS ap-southeast-1, the requesting region, with no egress charges applied
DCredits are charged in AWS us-east-1 at that region's credit rate, plus data egress fees for transferring results back to ap-southeast-1
Explanation
Snowflake's cross-region inference billing model charges credits in the requesting region regardless of where model inference actually occurs. Because the Snowflake account is deployed in AWS ap-southeast-1, all credits are consumed there. Snowflake does not apply data egress fees for cross-region inference — traffic travels over Snowflake's internal network infrastructure. There is no split billing between regions, and no premium surcharge is added on top of standard inference credit consumption. This billing design ensures organizations face no unexpected additional costs for accessing models in other regions beyond the standard per-token inference rate.
3. A data engineer at Tailspin Toys is building an automated product review processing pipeline using SNOWFLAKE.CORTEX.COMPLETE(). The downstream ingestion system requires every response to be a valid JSON object with exactly three fields: product_category, sentiment_label, and improvement_suggestion. The team must eliminate all manual post-processing needed to clean or reformat model output. Which approach guarantees that COMPLETE() always returns output conforming to the required structure? (Select one!)
AInclude explicit formatting instructions in the prompt such as 'Return only valid JSON containing product_category, sentiment_label, and improvement_suggestion' and rely on model compliance
BUse TRY_COMPLETE() instead of COMPLETE() to prevent malformed responses from propagating into downstream tables
CPass a JSON schema definition within the response_format key of the options parameter provided to COMPLETE()
DReplace COMPLETE() with SNOWFLAKE.CORTEX.AI_EXTRACT() because it natively produces multi-field structured key-value output designed for extraction tasks
Explanation
Passing a JSON schema through the response_format key inside the options parameter of COMPLETE() enforces constrained structured output at the token generation level. Snowflake verifies each generated token against the schema definition, making it impossible for the response to contain free-form text outside the required structure or fields that deviate from the schema. Prompt instructions requesting JSON-formatted output are probabilistic rather than deterministic -- the model may prefix the JSON with commentary, vary field names slightly, or produce subtly malformed structures under certain inputs, making prompt-only approaches unreliable when downstream ingestion requires strict schema conformance. AI_EXTRACT is a scalar function that accepts a single context passage and one natural language extraction question, returning the value corresponding to that specific question -- it cannot produce multi-field JSON objects and provides no schema compliance enforcement across complex unstructured review text. TRY_COMPLETE() is the error-safe variant of COMPLETE() that returns NULL instead of raising an exception on generation failure; it applies no output structure constraints and performs no validation on the generated response content.
4. A developer at Contoso Financial is deploying a Cortex Agent to answer complex research questions by orchestrating calls to both a structured database via Cortex Analyst and document repositories via Cortex Search. During testing, the agent occasionally enters long-running execution loops that far exceed acceptable response times and accumulate unexpected token costs. Which two configuration options should the developer apply to prevent runaway executions and enforce cost boundaries? (Select two!)
Multiple correct answersASet a time limit in the agent budget configuration to cap the maximum execution duration
BSet a token limit in the agent budget configuration to stop the agent when its token budget is consumed
CEnable Cortex Guard to intercept excessive token consumption and terminate the agent session
DAdd a LIMIT clause to all SQL queries generated internally by the Cortex Analyst tool
ESet the max_output_tokens parameter on each COMPLETE call invoked within the agent workflow
Explanation
Cortex Agents support a budget configuration that accepts two independent thresholds: a time limit that caps the maximum wall-clock duration of the agent execution, and a token limit that stops the agent once the total token budget is consumed. The agent halts as soon as either threshold is reached, providing dual protection against runaway executions. Cortex Guard is a safety feature designed to detect and filter harmful or unsafe content in responses — it is not a cost or time budget control mechanism. Adding LIMIT clauses only constrains SQL query result sizes and has no effect on the agent's cumulative token usage or total execution duration across multiple orchestrated steps. Setting max_output_tokens on individual COMPLETE calls limits single-call output but does not govern the agent's overall token consumption across the full multi-step workflow.
5. A data analyst at Contoso Financial needs to process thousands of customer support emails using SNOWFLAKE.CORTEX.COMPLETE() and insert the results into a relational table with three specific columns: issue_category, urgency_level, and resolution_recommended. A downstream stored procedure will fail with a parse error if the model response is not valid JSON containing exactly those field names. What is the most reliable way to ensure every model response conforms to the required structure? (Select one!)
APrefix the prompt with explicit JSON formatting instructions and rely on the model's instruction-following capability to produce consistent structure
BCall SNOWFLAKE.CORTEX.EXTRACT_ANSWER() in a separate invocation for each of the three fields, then assemble the results into a JSON object before inserting
CInclude a response_format parameter containing a JSON schema definition in the COMPLETE function's options object to enforce structured output
DSet the temperature option to 0 in the COMPLETE function's options object to maximize token selection determinism and reduce output variability
Explanation
The COMPLETE function's options object accepts a response_format parameter that takes a JSON schema definition specifying required field names and their data types. When this parameter is provided, Snowflake instructs the underlying model to produce output that strictly conforms to the defined schema on every invocation, making downstream structured inserts reliable without requiring custom parsing or error-handling logic. Setting temperature to 0 increases determinism in token selection but does not impose any schema constraint — the model can still produce free-form prose or inconsistently structured JSON, which will still break a strict parser. Adding explicit JSON formatting instructions to the prompt relies on model cooperation, which is inherently non-deterministic at production scale. Models frequently introduce extra commentary, missing fields, or formatting deviations that cause downstream failures. EXTRACT_ANSWER is purpose-built to return a single natural-language answer to a question extracted from a passage of text. It is not designed to emit multi-field schema-enforced JSON objects, and invoking it in a loop for each field would be both inefficient and unreliable for structured extraction at scale.
$7.99
One-time access to this exam
Full access to all 600 questions
Or $15/mo for all 253 exams
Detailed explanations
Free preview stays available