The SnowPro Specialty exams are certifications that prove deep, hands-on expertise in a specific Snowflake domain. Where the Core exam tests broad conceptual knowledge, Specialty exams demand applied skills like designing the privilege model for a Native App and choosing the right Cortex LLM Function for the job.
This article walks through hand-picked practice questions from the Specialty exam areas: Snowpark, Native Apps, and Gen AI / Cortex. The Snowpark exam (SPS-C01) ended on November 1, 2026, and can no longer be booked. Every explanation breaks down the technical reason each option is correct or incorrect, not just the answer.
The Native Apps and Gen AI Specialty exams are 55 questions in 85 minutes. That gives you more time per question than the Core exam (100 questions in 115 minutes) — about 93 seconds each — but don't assume the clock is generous. Specialty exams include code-snippet reading and architecture-design judgment calls, and that work eats time quickly.
Here are each exam's domains and weights (official Study Guides, as of October 2026). In 2026, Native Apps moved to a new version (NAS-C02) and Gen AI to GES-C02, and their domains changed.
| Exam | Domains (Weight) | Main Topics |
|---|---|---|
| Snowpark (SPS-C01, ended Nov 1, 2026) | Snowpark Concepts 15% / Snowpark API for Python 30% / Snowpark for Data Transformations 35% / Snowpark Performance Optimization 20% | DataFrame API, UDF/UDTF, stored procedures (Python) |
| Native Apps(NAS-C02) | Design 22% / Build 38% / Deploy 11% / Manage 29% | Architecture and privilege design, manifests and setup scripts, publishing on the Marketplace, versioning and upgrades |
| Gen AI(GES-C02) | Snowflake for Gen AI Overview 18% / Snowflake Gen AI Functions 38% / Snowflake Gen AI Governance 29% / Snowflake Document Processing 15% | Cortex AI functions (AI_COMPLETE, AI_EXTRACT, etc.), Cortex Search, Cortex Analyst, Snowflake Intelligence, AI_PARSE_DOCUMENT |
The Snowpark Python DataFrame API uses lazy evaluation — queries don't actually run until you call an action like collect() or show(). Use the question below to check your understanding of how DataFrames behave and what type each call returns.
Snowpark — DataFrame API
Question 1
Using the Snowpark Python DataFrame API, what does the following code print? df = session.table('orders') df2 = df.filter(col('status') == 'completed') df3 = df2.select('order_id', 'amount') print(type(df3))
Correct answer: B
The Snowpark Python DataFrame API is lazily evaluated. session.table(), filter(), and select() are all transformations — they build up a logical SQL plan but do not send a query to Snowflake. The type of df3 is snowflake.snowpark.DataFrame, a Snowpark-specific object that is distinct from a pandas DataFrame. To actually execute the query and retrieve results you need to call an action method: collect() (returns a list of Rows), show() (prints results to stdout), or to_pandas() (converts to a pandas DataFrame). This lazy-evaluation model is fundamental to working with Snowpark.
In the Snowflake Native App Framework, applications run under the principle of least privilege. After installation, an app cannot reach Consumer data without an explicit grant from the Consumer. Use the question below to check your grasp of this privilege model.
Native Apps — Security Model
Question 2
After installation, what step is required for a Snowflake Native App to access the Consumer's data?
Correct answer: C
Native Apps follow the principle of least privilege and by default have no access to Consumer data. The app can reach Consumer data only after the Consumer explicitly grants it privileges on specific objects (in Snowsight or with GRANT statements) or binds its own objects to the references the app requests. manifest.yml declares the privileges and references the app needs, but the actual grant is a Consumer-side action. Even if the Provider hard-codes table names (A), the query will fail with a SQL error without the grant. Automatic access to every database (B) is impossible under this security model. Native Apps are explicitly designed to access Consumer data when granted, so (D) is also wrong.
Snowflake Cortex LLM Functions let you invoke large language models directly from SQL for text generation, translation, and summarization. The defining feature is that you can call AI from a plain SELECT statement — no Stored Procedure required.
Cortex — LLM Functions
Question 3
Which statement about the COMPLETE function in Snowflake Cortex LLM Functions is correct?
Correct answer: B
Snowflake Cortex's COMPLETE function can be called directly from a SQL SELECT to generate text. You specify the LLM model to use (for example 'llama3.3-70b'; available models vary by region) along with the input prompt. For example, SELECT SNOWFLAKE.CORTEX.COMPLETE('llama3.3-70b', 'What is Snowflake?') works directly in SQL. You can even run COMPLETE in batch against every row of a table. It is not restricted to Stored Procedures (A). Without the options argument it returns the response as a string, so it does not always return JSON (C). Snowflake's edition comparison lists Cortex AI Functions for every edition, including Standard, so it is not limited to Business Critical (D). Note that the current official docs keep SNOWFLAKE.CORTEX.COMPLETE for backward compatibility and recommend its successor, AI_COMPLETE, for new use cases (the GES-C02 Study Guide lists both AI_COMPLETE and COMPLETE).
| Exam | Code | Questions | Duration | Fee | Prerequisite |
|---|---|---|---|---|---|
| Specialty: Snowpark (ended Nov 1, 2026) | SPS-C01 | 55 | 85 min | $225 | None |
| Specialty: Native Apps | NAS-C02 | 55 | 85 min | $225 | None |
| Specialty: Gen AI | GES-C02 | 55 | 85 min | $225 | None |
Native Apps and Gen AI share the same format: 55 questions, 85 minutes, and $225, with no prerequisites (you don't need to pass Core first). What varies between them is the domain content and the question style — that's where the difficulty differences come from. As of November 4, 2026, the official page for the new AI Governance & Security exam (ADS-B01) does not list its question count, time limit, or fee. The Snowpark exam ended on November 1, 2026, and Snowflake recommends Advanced: Data Engineer (which requires an active Core certification) as the next step.
Specialty exams are heavy on scenario-based questions that you simply can't solve by reading docs alone. The most effective study technique is to actually run Native Apps and Cortex functions on the Snowflake 30-day free trial and build muscle memory for how they behave. In particular, running Cortex COMPLETE from SQL and writing a Native App's manifest and setup script are things you only really understand once you've done them.
Specialty exams embed Python and SQL snippets directly in question stems, and you need to read them precisely. Don't just memorize syntax — actually write the code and watch what it does.
Specialty exams deal with relatively new features such as the Native App Framework, Cortex AI functions, and Snowpark Container Services. The 2026 GES-C02 version added Snowflake Intelligence, Cortex Code, MCP (Model Context Protocol), and AI functions such as AI_TRANSCRIBE, AI_REDACT, and AI_FILTER. Check Snowflake's official What's New page and release notes regularly. Note that the GES-C02 Study Guide also lists Public Preview features such as Cortex Fine-tuning, so preview-stage features can be in scope. Confirm the final scope in each exam's latest Study Guide.
"Snowflake Document Processing" (15%), now a domain of its own in GES-C02, covers — per the official Study Guide — AI_PARSE_DOCUMENT's OCR and LAYOUT modes (with page_split and page_limit), AI_EXTRACT response formats and prompting, automating document processing with Streams and Tasks, and troubleshooting. The final version of the earlier GES-C01 had no such domain, so if you studied with materials for the earlier version, give this area extra attention. Snowpipe Streaming and Dynamic Tables, by the way, are not in the Gen AI and Native Apps Study Guides (they are in scope for Core, Advanced: Data Engineer, Architect, and others).
Drill more Specialty practice questions
Practice Native Apps and Gen AI questions in the NicheeLab question bank (built for the NAS-C02 and GES-C02 outlines)
Try free questions →Do I need to pass the Core exam before taking a Specialty exam?
No. Gen AI and Native Apps have no prerequisites, so you can take either without passing Core first. An active Core certification is required only for the Advanced exams, such as Architect and Data Engineer. Gen AI and Native Apps are 55 questions in 85 minutes and cost $225 — fewer questions than Core (100 questions, 115 minutes, $175), but the difficulty is higher because each exam tests deep, hands-on knowledge of its specialty area. As of November 4, 2026, there are three Specialty exams: AI Governance & Security (ADS-B01), Gen AI, and Native Apps. ADS-B01 is new, and its official page does not list the question count, time limit, or fee. The Snowpark exam ended on November 1, 2026.
Which programming languages did the Snowpark Specialty exam cover?
The Snowpark Specialty exam (which ended on November 1, 2026) centered on Python. It used Python code snippets covering DataFrame API operations, UDF/UDTF authoring, and Stored Procedure implementation. Java and Scala are also supported by Snowpark, but the exam leaned heavily on Python.
Did the Native Apps and Gen AI exams change in 2026?
Yes. For Gen AI, the new GES-C02 version launched on May 19, 2026, and the earlier GES-C01 retired on July 20, 2026. Native Apps also switched to a new version, NAS-C02, in 2026, and the earlier NAS-C01 can no longer be scheduled through Pearson VUE. Both are still 55 questions, 85 minutes, and $225 with no prerequisites, but the domains changed. GES-C02 has 4 domains — Snowflake for Gen AI Overview 18%, Snowflake Gen AI Functions 38%, Snowflake Gen AI Governance 29%, and Snowflake Document Processing 15% — and adds Snowflake Intelligence, Cortex Code, MCP (Model Context Protocol), and AI functions such as AI_TRANSCRIBE, AI_REDACT, and AI_FILTER. NAS-C02 has 4 domains: Design 22%, Build 38%, Deploy 11%, and Manage 29%. NicheeLab rebuilt its Native Apps and Gen AI banks for the new outlines (NAS-C02 and GES-C02) in November 2026. Check the latest official Study Guide for the domains before you finalize your study plan.
Related SnowPro Specialty Articles
Snowpark: Complete Guide
Deep dive on the DataFrame API, UDFs, and Stored Procedures
What Are Native Apps?
Application Package layout and distribution model
Cortex AI: Complete Guide
How to use LLM Functions, ML Functions, and Cortex Search
All Snowflake Certifications
Overview, cost, and language availability for each SnowPro exam
Practice with certification-focused question sets
Try free questionsNicheeLab Editorial Team
NicheeLab editorial team focused on data engineering and cloud certification learning. Content is structured around practical study needs and official exam domains.
Snowflake Certifications: All 11 Exams Explained (2026)
Every SnowPro certification on Snowflake's official list — C...
Snowflake Exam Difficulty Ranking: All 11 Certs Compared (2026)
All 11 SnowPro exams compared by difficulty in a ranking and...
Snowflake Study Guide: Fastest Pass Route by Exam (2026)
How to pass SnowPro certifications efficiently — official ma...
SnowPro Core (COF-C03): Complete Exam Guide (2026)
Pass the SnowPro Core exam — five domains, scope, sample que...
SnowPro Associate Platform (SOL-C01): Current Status & What to Take Next (2026)
As of October 2026, SnowPro Associate: Platform (SOL-C01) is...