Amazon AWS AI Practitioner AIF-C01 GenAI Foundations Tokens Embeddings And Inference Economics Practice Test

 

AIF-C01 skills 2.1 | 30 original questions

This AWS Certified AI Practitioner AIF-C01 practice test focuses on genai foundations tokens embeddings and inference economics through original scenario-based questions aligned to AWS Exam Guide version 1.1 published April 30, 2026. Use the full ExamSnap AIF-C01 collection for broader practice across all five current exam domains. For broader exam preparation, review the Amazon AWS Certified AI Practitioner AIF-C01 Exam Dumps page.

Instructions: Select the best answer for each question. Review the rationale after answering. Each distractor includes a brief explanation of why it is not the strongest fit for the stated scenario.

Question 1

For the fraud-review pilot at City Power and Light, stakeholders need to control both token cost and response latency for verbose generations. Which concept, service, or technique most directly addresses this goal? Operational ownership is already assigned, so the team is comparing technical fit rather than staffing models. The control owner requires evidence from 7 test groups before the 677-day release review.

1. Limit output length when long responses are not needed
2. Use prompt caching when supported and repeated context is large
3. Regional deployment choice
4. On-demand inference
5. Reduce unnecessary input tokens

Correct answer: A

Why: Longer generated outputs consume more output tokens and often take longer to produce. It directly addresses the requirement in this scenario.

Option review:

A: Longer generated outputs consume more output tokens and often take longer to produce. It directly addresses the requirement in this scenario.

B: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Region selection can affect compliance, latency, availability, and cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: On-demand consumption is flexible but unit economics and latency can differ from reserved capacity options. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Limit output length when long responses are not needed – Longer generated outputs consume more output tokens and often take longer to produce.

Question 2

Consolidated Messenger is comparing alternatives for its analytics modernization. The risk manager needs to use an attention-based large language architecture trained on extensive text data. Which option is most appropriate while trying to meet a strict latency target? Existing application interfaces can accommodate any of the listed choices, so functional fit is the deciding factor. The project has 4 downstream consumers and a monthly review of approximately 714 sampled interactions.

1. Token
2. Transformer-based LLM
3. Prompt engineering
4. Diffusion model
5. Negative prompt or explicit exclusion

Correct answer: B

Why: Modern LLMs commonly use transformer architectures to model relationships across token sequences. It directly addresses the requirement in this scenario.

Option review:

A: Models commonly process text as tokens, which are subword or character-like units rather than necessarily whole words. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Modern LLMs commonly use transformer architectures to model relationships across token sequences. It directly addresses the requirement in this scenario.

C: Prompt engineering shapes model behavior through carefully structured input without changing model weights. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Diffusion models are widely used for image and other media generation through a denoising process. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Negative constraints help define what the model should not produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Transformer-based LLM – Modern LLMs commonly use transformer architectures to model relationships across token sequences.

Question 3

An architecture review at Nod Publishers has narrowed a compliance-assistant prototype decision to one requirement: lower per-request inference cost when a prompt includes redundant context. What should the security architect select? Assume the required AWS capabilities are available in the selected Region and normal governance controls are in place. The rollout spans 9 application teams, each using the same approved requirement set for the next 751 days.

1. Account for both input and output token rates
2. On-demand inference
3. Reduce unnecessary input tokens
4. Limit output length when long responses are not needed
5. Prompt caching

Correct answer: C

Why: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. It directly addresses the requirement in this scenario.

Option review:

A: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: On-demand consumption is flexible but unit economics and latency can differ from reserved capacity options. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. It directly addresses the requirement in this scenario.

D: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Caching can lower latency and token-processing cost for reusable prompt content. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Reduce unnecessary input tokens – Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost.

Question 4

The risk manager at Fabrikam Health is preparing a recommendation for a forecasting initiative. The recommendation must work with more than one modality such as text and images in a single model interaction. Which choice is the best match? The review committee wants a direct mapping from the requirement to the chosen capability. The evaluation set contains examples from 6 business workflows and 788 recent production cases.

1. Context
2. Prompt engineering
3. Chunking
4. Multimodal model
5. Vector

Correct answer: D

Why: Multimodal models can process or generate combinations of text, image, audio, or other modalities. It directly addresses the requirement in this scenario.

Option review:

A: Context supplies task-relevant information that is not part of the instruction itself. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Prompt engineering shapes model behavior through carefully structured input without changing model weights. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Multimodal models can process or generate combinations of text, image, audio, or other modalities. It directly addresses the requirement in this scenario.

E: A vector is the numeric coordinate representation produced by an embedding model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Multimodal model – Multimodal models can process or generate combinations of text, image, audio, or other modalities.

Question 5

Wingtip Logistics has completed discovery for a customer-support modernization. Before implementation, the security architect must decide how to reduce processing cost or latency for reusable prompt prefixes when the service supports caching. Which choice best satisfies that requirement? The solution will serve multiple internal teams, so the recommendation should be reusable without changing the core requirement. The initial rollout covers 825 internal users across 3 business units.

1. Longer input and output
2. Limit output length when long responses are not needed
3. Regional deployment choice
4. Redundant multi-region design
5. Use prompt caching when supported and repeated context is large

Correct answer: E

Why: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. It directly addresses the requirement in this scenario.

Option review:

A: Token volume directly affects many GenAI cost models and can also affect latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Region selection can affect compliance, latency, availability, and cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Higher availability commonly requires paying for redundant resources and operational complexity. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. It directly addresses the requirement in this scenario.

Learning point: Use prompt caching when supported and repeated context is large – Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes.

Question 6

While planning a agentic workflow trial, Trey Research identifies this requirement: identify the units of text or other input that a model processes for context and billing. Which option should the risk manager prioritize if the goal is to limit exposure of sensitive data? The decision must follow the workload characteristics rather than a preference for the largest model or newest service. The workload processes about 862 requests during its busiest hour and has a documented fallback path.

1. Token
2. Diffusion model
3. Multimodal model
4. Negative prompt or explicit exclusion
5. Foundation model

Correct answer: A

Why: Models commonly process text as tokens, which are subword or character-like units rather than necessarily whole words. It directly addresses the requirement in this scenario.

Option review:

A: Models commonly process text as tokens, which are subword or character-like units rather than necessarily whole words. It directly addresses the requirement in this scenario.

B: Diffusion models are widely used for image and other media generation through a denoising process. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Multimodal models can process or generate combinations of text, image, audio, or other modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Negative constraints help define what the model should not produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Foundation models are trained at scale on broad data and support many tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Token – Models commonly process text as tokens, which are subword or character-like units rather than necessarily whole words.

Question 7

A proof of concept at Bellows College exposed a design decision for the security architect: the solution must estimate inference spend accurately for a conversational application. Which option most directly solves that problem? The security baseline is already defined; the decision here concerns the specific capability described in the requirement. The pilot uses 899 representative records from 5 approved data sources.

1. Redundant multi-region design
2. Account for both input and output token rates
3. Regional deployment choice
4. Custom model
5. Prompt caching

Correct answer: B

Why: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. It directly addresses the requirement in this scenario.

Option review:

A: Higher availability commonly requires paying for redundant resources and operational complexity. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. It directly addresses the requirement in this scenario.

C: Region selection can affect compliance, latency, availability, and cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Customization can improve fit but creates additional lifecycle cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Caching can lower latency and token-processing cost for reusable prompt content. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Account for both input and output token rates – Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered.

Question 8

Blue Yonder Airlines is documenting the target state for a operations automation program. The risk manager needs a solution that can store the numeric representation used for similarity search in embedding-based retrieval. Which option is the strongest fit? The recommendation must solve the stated requirement without introducing unrelated platform complexity. The first release supports 2 departments and is reviewed every 936 days.

1. Diffusion model
2. Negative prompt or explicit exclusion
3. Vector
4. Output schema
5. Prompt engineering

Correct answer: C

Why: A vector is the numeric coordinate representation produced by an embedding model. It directly addresses the requirement in this scenario.

Option review:

A: Diffusion models are widely used for image and other media generation through a denoising process. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Negative constraints help define what the model should not produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: A vector is the numeric coordinate representation produced by an embedding model. It directly addresses the requirement in this scenario.

D: A clear output format reduces ambiguity and makes downstream processing easier. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Prompt engineering shapes model behavior through carefully structured input without changing model weights. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Vector – A vector is the numeric coordinate representation produced by an embedding model.

Question 9

Woodgrove Bank is reviewing a sales-assistant rollout. The security architect has one primary requirement: control both token cost and response latency for verbose generations. Which choice best fits the requirement? The design must remain supportable after launch, but no additional feature is required beyond the stated need. The service has a 973-millisecond internal response target for the affected workflow.

1. Account for both input and output token rates
2. On-demand inference
3. Reduce unnecessary input tokens
4. Limit output length when long responses are not needed
5. Longer input and output

Correct answer: D

Why: Longer generated outputs consume more output tokens and often take longer to produce. It directly addresses the requirement in this scenario.

Option review:

A: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: On-demand consumption is flexible but unit economics and latency can differ from reserved capacity options. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Longer generated outputs consume more output tokens and often take longer to produce. It directly addresses the requirement in this scenario.

E: Token volume directly affects many GenAI cost models and can also affect latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Limit output length when long responses are not needed – Longer generated outputs consume more output tokens and often take longer to produce.

Question 10

During a design review for Wide World Importers, the risk manager must generate media through iterative denoising from a noisy representation. The team also wants to reduce manual handling. What should the team choose? A short pilot window means the team prefers an approach that can be evaluated with clear success criteria. The team is comparing 4 candidate designs after a 50-day proof of concept.

1. Multimodal model
2. Chunking
3. Prompt engineering
4. Context
5. Diffusion model

Correct answer: E

Why: Diffusion models are widely used for image and other media generation through a denoising process. It directly addresses the requirement in this scenario.

Option review:

A: Multimodal models can process or generate combinations of text, image, audio, or other modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Prompt engineering shapes model behavior through carefully structured input without changing model weights. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Context supplies task-relevant information that is not part of the instruction itself. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Diffusion models are widely used for image and other media generation through a denoising process. It directly addresses the requirement in this scenario.

Learning point: Diffusion model – Diffusion models are widely used for image and other media generation through a denoising process.

Question 11

VanArsdel Media is moving a document-intelligence project from pilot to production. The key decision is how to estimate inference spend accurately for a conversational application. Which option is the strongest fit if the team wants to use current managed AWS capabilities? The architecture board will reject a choice that addresses a different problem from the one described. The control owner requires evidence from 9 test groups before the 87-day release review.

1. Account for both input and output token rates
2. Reduce unnecessary input tokens
3. Prompt caching
4. Provisioned throughput
5. Smaller model

Correct answer: A

Why: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. It directly addresses the requirement in this scenario.

Option review:

A: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. It directly addresses the requirement in this scenario.

B: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Caching can lower latency and token-processing cost for reusable prompt content. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Provisioned capacity can improve predictability for steady workloads but introduces commitment cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: A smaller model often uses fewer resources but can sacrifice capability on harder tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Account for both input and output token rates – Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered.

Question 12

A workshop at Datum Dynamics focuses on a single decision: how to design instructions and context to guide a foundation model toward a desired response. Which option should the risk manager recommend? Budget has been approved for the project, but the team still wants to avoid unnecessary recurring consumption. The project has 6 downstream consumers and a monthly review of approximately 124 sampled interactions.

1. Vector
2. Prompt engineering
3. Context
4. Instruction
5. Chunking

Correct answer: B

Why: Prompt engineering shapes model behavior through carefully structured input without changing model weights. It directly addresses the requirement in this scenario.

Option review:

A: A vector is the numeric coordinate representation produced by an embedding model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Prompt engineering shapes model behavior through carefully structured input without changing model weights. It directly addresses the requirement in this scenario.

C: Context supplies task-relevant information that is not part of the instruction itself. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Instructions tell the model what to do. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Prompt engineering – Prompt engineering shapes model behavior through carefully structured input without changing model weights.

Question 13

For the claims-processing redesign at Alpine Ski House, stakeholders need to reduce processing cost or latency for reusable prompt prefixes when the service supports caching. Which concept, service, or technique most directly addresses this goal? The team will validate the result with representative production examples before rollout. The rollout spans 3 application teams, each using the same approved requirement set for the next 161 days.

1. Custom model
2. Prompt caching
3. Use prompt caching when supported and repeated context is large
4. On-demand inference
5. Regional deployment choice

Correct answer: C

Why: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. It directly addresses the requirement in this scenario.

Option review:

A: Customization can improve fit but creates additional lifecycle cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Caching can lower latency and token-processing cost for reusable prompt content. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. It directly addresses the requirement in this scenario.

D: On-demand consumption is flexible but unit economics and latency can differ from reserved capacity options. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Region selection can affect compliance, latency, availability, and cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Use prompt caching when supported and repeated context is large – Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes.

Question 14

Humongous Insurance is comparing alternatives for its personalization program. The risk manager needs to start with a broadly pretrained model that can be adapted or prompted for many downstream tasks. Which option is most appropriate while trying to meet a strict latency target? The pilot has representative data, and the team will measure the selected approach against an agreed acceptance threshold. The evaluation set contains examples from 8 business workflows and 198 recent production cases.

1. Output schema
2. Transformer-based LLM
3. Diffusion model
4. Foundation model
5. Chunking

Correct answer: D

Why: Foundation models are trained at scale on broad data and support many tasks. It directly addresses the requirement in this scenario.

Option review:

A: A clear output format reduces ambiguity and makes downstream processing easier. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Modern LLMs commonly use transformer architectures to model relationships across token sequences. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Diffusion models are widely used for image and other media generation through a denoising process. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Foundation models are trained at scale on broad data and support many tasks. It directly addresses the requirement in this scenario.

E: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Foundation model – Foundation models are trained at scale on broad data and support many tasks.

Question 15

An architecture review at Graphic Design Institute has narrowed a developer-productivity pilot decision to one requirement: lower per-request inference cost when a prompt includes redundant context. What should the security architect select? The team will document the rationale for auditors and wants the recommendation to be defensible from the scenario facts. The initial rollout covers 235 internal users across 5 business units.

1. Limit output length when long responses are not needed
2. Custom model
3. Prompt caching
4. Use prompt caching when supported and repeated context is large
5. Reduce unnecessary input tokens

Correct answer: E

Why: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. It directly addresses the requirement in this scenario.

Option review:

A: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Customization can improve fit but creates additional lifecycle cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Caching can lower latency and token-processing cost for reusable prompt content. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. It directly addresses the requirement in this scenario.

Learning point: Reduce unnecessary input tokens – Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost.

Question 16

The risk manager at Relecloud is preparing a recommendation for a fraud-review pilot. The recommendation must split long source documents into smaller retrieval units for indexing or context assembly. Which choice is the best match? The team wants the least complex technically correct choice that satisfies the requirement. The workload processes about 272 requests during its busiest hour and has a documented fallback path.

1. Chunking
2. Diffusion model
3. Foundation model
4. Multimodal model
5. Instruction

Correct answer: A

Why: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. It directly addresses the requirement in this scenario.

Option review:

A: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. It directly addresses the requirement in this scenario.

B: Diffusion models are widely used for image and other media generation through a denoising process. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Foundation models are trained at scale on broad data and support many tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Multimodal models can process or generate combinations of text, image, audio, or other modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Instructions tell the model what to do. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Chunking – Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt.

Question 17

Adventure Works Manufacturing has completed discovery for a analytics modernization. Before implementation, the security architect must decide how to control both token cost and response latency for verbose generations. Which choice best satisfies that requirement? The workload has passed basic feasibility checks, so the remaining question is which approach best matches the requirement. The pilot uses 309 representative records from 7 approved data sources.

1. Smaller model
2. Limit output length when long responses are not needed
3. Custom model
4. Use prompt caching when supported and repeated context is large
5. Longer input and output

Correct answer: B

Why: Longer generated outputs consume more output tokens and often take longer to produce. It directly addresses the requirement in this scenario.

Option review:

A: A smaller model often uses fewer resources but can sacrifice capability on harder tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Longer generated outputs consume more output tokens and often take longer to produce. It directly addresses the requirement in this scenario.

C: Customization can improve fit but creates additional lifecycle cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Token volume directly affects many GenAI cost models and can also affect latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Limit output length when long responses are not needed – Longer generated outputs consume more output tokens and often take longer to produce.

Question 18

While planning a compliance-assistant prototype, Proseware Services identifies this requirement: represent semantic meaning as a numeric vector so similar content can be compared mathematically. Which option should the risk manager prioritize if the goal is to limit exposure of sensitive data? Stakeholders have ruled out a broad redesign and want the choice that most precisely addresses the stated need. The first release supports 4 departments and is reviewed every 346 days.

1. Output schema
2. Diffusion model
3. Embedding
4. Multimodal model
5. Vector

Correct answer: C

Why: Embeddings map content into vector space where semantic similarity can be measured. It directly addresses the requirement in this scenario.

Option review:

A: A clear output format reduces ambiguity and makes downstream processing easier. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Diffusion models are widely used for image and other media generation through a denoising process. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Embeddings map content into vector space where semantic similarity can be measured. It directly addresses the requirement in this scenario.

D: Multimodal models can process or generate combinations of text, image, audio, or other modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: A vector is the numeric coordinate representation produced by an embedding model. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Embedding – Embeddings map content into vector space where semantic similarity can be measured.

Question 19

A proof of concept at Lucerne Publishing exposed a design decision for the security architect: the solution must reduce processing cost or latency for reusable prompt prefixes when the service supports caching. Which option most directly solves that problem? Operational ownership is already assigned, so the team is comparing technical fit rather than staffing models. The service has a 383-millisecond internal response target for the affected workflow.

1. Prompt caching
2. Regional deployment choice
3. Custom model
4. Use prompt caching when supported and repeated context is large
5. Longer input and output

Correct answer: D

Why: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. It directly addresses the requirement in this scenario.

Option review:

A: Caching can lower latency and token-processing cost for reusable prompt content. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Region selection can affect compliance, latency, availability, and cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Customization can improve fit but creates additional lifecycle cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. It directly addresses the requirement in this scenario.

E: Token volume directly affects many GenAI cost models and can also affect latency. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Use prompt caching when supported and repeated context is large – Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes.

Question 20

Lamna Healthcare is documenting the target state for a customer-support modernization. The risk manager needs a solution that can use an attention-based large language architecture trained on extensive text data. Which option is the strongest fit? Existing application interfaces can accommodate any of the listed choices, so functional fit is the deciding factor. The team is comparing 6 candidate designs after a 420-day proof of concept.

1. Prompt engineering
2. Embedding
3. Negative prompt or explicit exclusion
4. Context
5. Transformer-based LLM

Correct answer: E

Why: Modern LLMs commonly use transformer architectures to model relationships across token sequences. It directly addresses the requirement in this scenario.

Option review:

A: Prompt engineering shapes model behavior through carefully structured input without changing model weights. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Embeddings map content into vector space where semantic similarity can be measured. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Negative constraints help define what the model should not produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Context supplies task-relevant information that is not part of the instruction itself. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Modern LLMs commonly use transformer architectures to model relationships across token sequences. It directly addresses the requirement in this scenario.

Learning point: Transformer-based LLM – Modern LLMs commonly use transformer architectures to model relationships across token sequences.

Question 21

Contoso Retail is reviewing a agentic workflow trial. The security architect has one primary requirement: estimate inference spend accurately for a conversational application. Which choice best fits the requirement? Assume the required AWS capabilities are available in the selected Region and normal governance controls are in place. The control owner requires evidence from 3 test groups before the 457-day release review.

1. Account for both input and output token rates
2. Regional deployment choice
3. Smaller model
4. Provisioned throughput
5. Custom model

Correct answer: A

Why: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. It directly addresses the requirement in this scenario.

Option review:

A: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. It directly addresses the requirement in this scenario.

B: Region selection can affect compliance, latency, availability, and cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: A smaller model often uses fewer resources but can sacrifice capability on harder tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Provisioned capacity can improve predictability for steady workloads but introduces commitment cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Customization can improve fit but creates additional lifecycle cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Account for both input and output token rates – Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered.

Question 22

During a design review for Fourth Coffee, the risk manager must generate media through iterative denoising from a noisy representation. The team also wants to reduce manual handling. What should the team choose? The review committee wants a direct mapping from the requirement to the chosen capability. The project has 8 downstream consumers and a monthly review of approximately 494 sampled interactions.

1. Chunking
2. Diffusion model
3. Foundation model
4. Output schema
5. Token

Correct answer: B

Why: Diffusion models are widely used for image and other media generation through a denoising process. It directly addresses the requirement in this scenario.

Option review:

A: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Diffusion models are widely used for image and other media generation through a denoising process. It directly addresses the requirement in this scenario.

C: Foundation models are trained at scale on broad data and support many tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: A clear output format reduces ambiguity and makes downstream processing easier. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Models commonly process text as tokens, which are subword or character-like units rather than necessarily whole words. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Diffusion model – Diffusion models are widely used for image and other media generation through a denoising process.

Question 23

Margie Travel is moving a operations automation program from pilot to production. The key decision is how to lower per-request inference cost when a prompt includes redundant context. Which option is the strongest fit if the team wants to use current managed AWS capabilities? The solution will serve multiple internal teams, so the recommendation should be reusable without changing the core requirement. The rollout spans 5 application teams, each using the same approved requirement set for the next 531 days.

1. Smaller model
2. Limit output length when long responses are not needed
3. Reduce unnecessary input tokens
4. Prompt caching
5. Redundant multi-region design

Correct answer: C

Why: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. It directly addresses the requirement in this scenario.

Option review:

A: A smaller model often uses fewer resources but can sacrifice capability on harder tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Longer generated outputs consume more output tokens and often take longer to produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. It directly addresses the requirement in this scenario.

D: Caching can lower latency and token-processing cost for reusable prompt content. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Higher availability commonly requires paying for redundant resources and operational complexity. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Reduce unnecessary input tokens – Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost.

Question 24

A workshop at School of Fine Art focuses on a single decision: how to represent semantic meaning as a numeric vector so similar content can be compared mathematically. Which option should the risk manager recommend? The decision must follow the workload characteristics rather than a preference for the largest model or newest service. The evaluation set contains examples from 2 business workflows and 568 recent production cases.

1. Instruction
2. Prompt engineering
3. Chunking
4. Embedding
5. Foundation model

Correct answer: D

Why: Embeddings map content into vector space where semantic similarity can be measured. It directly addresses the requirement in this scenario.

Option review:

A: Instructions tell the model what to do. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Prompt engineering shapes model behavior through carefully structured input without changing model weights. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Embeddings map content into vector space where semantic similarity can be measured. It directly addresses the requirement in this scenario.

E: Foundation models are trained at scale on broad data and support many tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Embedding – Embeddings map content into vector space where semantic similarity can be measured.

Question 25

For the internal search upgrade at Northwind Analytics, stakeholders need to control both token cost and response latency for verbose generations. Which concept, service, or technique most directly addresses this goal? The security baseline is already defined; the decision here concerns the specific capability described in the requirement. The initial rollout covers 605 internal users across 7 business units.

1. Provisioned throughput
2. Regional deployment choice
3. Reduce unnecessary input tokens
4. Account for both input and output token rates
5. Limit output length when long responses are not needed

Correct answer: E

Why: Longer generated outputs consume more output tokens and often take longer to produce. It directly addresses the requirement in this scenario.

Option review:

A: Provisioned capacity can improve predictability for steady workloads but introduces commitment cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Region selection can affect compliance, latency, availability, and cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Many GenAI pricing models distinguish input-token and output-token usage, so both must be considered. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Longer generated outputs consume more output tokens and often take longer to produce. It directly addresses the requirement in this scenario.

Learning point: Limit output length when long responses are not needed – Longer generated outputs consume more output tokens and often take longer to produce.

Question 26

Litware Financial is comparing alternatives for its document-intelligence project. The risk manager needs to work with more than one modality such as text and images in a single model interaction. Which option is most appropriate while trying to meet a strict latency target? The recommendation must solve the stated requirement without introducing unrelated platform complexity. The workload processes about 642 requests during its busiest hour and has a documented fallback path.

1. Multimodal model
2. Transformer-based LLM
3. Foundation model
4. Embedding
5. Context

Correct answer: A

Why: Multimodal models can process or generate combinations of text, image, audio, or other modalities. It directly addresses the requirement in this scenario.

Option review:

A: Multimodal models can process or generate combinations of text, image, audio, or other modalities. It directly addresses the requirement in this scenario.

B: Modern LLMs commonly use transformer architectures to model relationships across token sequences. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Foundation models are trained at scale on broad data and support many tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Embeddings map content into vector space where semantic similarity can be measured. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Context supplies task-relevant information that is not part of the instruction itself. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Multimodal model – Multimodal models can process or generate combinations of text, image, audio, or other modalities.

Question 27

An architecture review at A. Datum Research has narrowed a knowledge-assistant rollout decision to one requirement: lower per-request inference cost when a prompt includes redundant context. What should the security architect select? The design must remain supportable after launch, but no additional feature is required beyond the stated need. The pilot uses 679 representative records from 9 approved data sources.

1. Provisioned throughput
2. Reduce unnecessary input tokens
3. Use prompt caching when supported and repeated context is large
4. Redundant multi-region design
5. Smaller model

Correct answer: B

Why: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. It directly addresses the requirement in this scenario.

Option review:

A: Provisioned capacity can improve predictability for steady workloads but introduces commitment cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. It directly addresses the requirement in this scenario.

C: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Higher availability commonly requires paying for redundant resources and operational complexity. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: A smaller model often uses fewer resources but can sacrifice capability on harder tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Reduce unnecessary input tokens – Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost.

Question 28

The risk manager at Coho Winery is preparing a recommendation for a claims-processing redesign. The recommendation must design instructions and context to guide a foundation model toward a desired response. Which choice is the best match? A short pilot window means the team prefers an approach that can be evaluated with clear success criteria. The first release supports 6 departments and is reviewed every 716 days.

1. Instruction
2. Chunking
3. Prompt engineering
4. Multimodal model
5. Foundation model

Correct answer: C

Why: Prompt engineering shapes model behavior through carefully structured input without changing model weights. It directly addresses the requirement in this scenario.

Option review:

A: Instructions tell the model what to do. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Chunking divides content into manageable segments that can be embedded, retrieved, or inserted into a prompt. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Prompt engineering shapes model behavior through carefully structured input without changing model weights. It directly addresses the requirement in this scenario.

D: Multimodal models can process or generate combinations of text, image, audio, or other modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Foundation models are trained at scale on broad data and support many tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Prompt engineering – Prompt engineering shapes model behavior through carefully structured input without changing model weights.

Question 29

Lucerne Retail has completed discovery for a personalization program. Before implementation, the security architect must decide how to reduce processing cost or latency for reusable prompt prefixes when the service supports caching. Which choice best satisfies that requirement? The architecture board will reject a choice that addresses a different problem from the one described. The service has a 753-millisecond internal response target for the affected workflow.

1. Prompt caching
2. Reduce unnecessary input tokens
3. Redundant multi-region design
4. Use prompt caching when supported and repeated context is large
5. Smaller model

Correct answer: D

Why: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. It directly addresses the requirement in this scenario.

Option review:

A: Caching can lower latency and token-processing cost for reusable prompt content. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Token-priced services generally charge based on processed input and output tokens, so removing irrelevant context can reduce cost. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: Higher availability commonly requires paying for redundant resources and operational complexity. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes. It directly addresses the requirement in this scenario.

E: A smaller model often uses fewer resources but can sacrifice capability on harder tasks. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

Learning point: Use prompt caching when supported and repeated context is large – Prompt caching can avoid reprocessing repeated context and can improve economics for recurring large prefixes.

Question 30

While planning a developer-productivity pilot, Tailspin Toys identifies this requirement: start with a broadly pretrained model that can be adapted or prompted for many downstream tasks. Which option should the risk manager prioritize if the goal is to limit exposure of sensitive data? Budget has been approved for the project, but the team still wants to avoid unnecessary recurring consumption. The team is comparing 8 candidate designs after a 790-day proof of concept.

1. Multimodal model
2. Transformer-based LLM
3. Output schema
4. Negative prompt or explicit exclusion
5. Foundation model

Correct answer: E

Why: Foundation models are trained at scale on broad data and support many tasks. It directly addresses the requirement in this scenario.

Option review:

A: Multimodal models can process or generate combinations of text, image, audio, or other modalities. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

B: Modern LLMs commonly use transformer architectures to model relationships across token sequences. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

C: A clear output format reduces ambiguity and makes downstream processing easier. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

D: Negative constraints help define what the model should not produce. This can be appropriate in another scenario, but it does not most directly satisfy the requirement described here.

E: Foundation models are trained at scale on broad data and support many tasks. It directly addresses the requirement in this scenario.

Learning point: Foundation model – Foundation models are trained at scale on broad data and support many tasks.