Amazon AWS AIP-C01 GenAI Cost Optimization Practice Test

 

Topic 16 focuses on GenAI Cost Optimization and Resource Efficiency for the AWS Certified Generative AI Developer – Professional certification and the AIP-C01 exam, using Amazon Bedrock and AWS generative AI services and architecture scenarios where relevant. For broader exam preparation, review the AWS Certified Generative AI Developer – Professional AIP-C01 Exam Dumps page. Each question includes a concise explanation of the correct answer and the technical reason the other choices are incorrect.

Question 1

The research assistant application is prioritizing a requirement to measure prompt and completion token consumption before and after optimization. Which implementation is most appropriate?

1. Bedrock provisioned throughput optimization
2. tiered model routing
3. context pruning
4. token estimation and tracking

Correct Answer: D

 

Correct Answer

Answer D is correct because token estimation and tracking is designed to measure prompt and completion token consumption before and after optimization. It creates a quantitative baseline for cost, context size, and usage trends instead of optimizing blindly.

Incorrect Answers

Answer A is incorrect because Bedrock provisioned throughput optimization is primarily used to reserve model capacity when predictable sustained demand justifies dedicated throughput, which is a different requirement from the one being tested.

Answer B is incorrect because tiered model routing is primarily used to send simple requests to lower-cost models and reserve premium models for complex work, which is a different requirement from the one being tested.

Answer C is incorrect because context pruning is primarily used to remove irrelevant retrieved or conversational content before inference, which is a different requirement from the one being tested.

 

Question 2

The AI product engineering team has a design goal to remove irrelevant retrieved or conversational content before inference. What should the team choose?

1. cost-capability tradeoff evaluation
2. semantic caching
3. prompt compression
4. context pruning

Correct Answer: D

 

Correct Answer

Answer D is correct because context pruning is designed to remove irrelevant retrieved or conversational content before inference. It reduces input token volume while retaining information needed for the current request.

Incorrect Answers

Answer A is incorrect because cost-capability tradeoff evaluation is primarily used to choose the least expensive model that still meets the quality and latency target, which is a different requirement from the one being tested.

Answer B is incorrect because semantic caching is primarily used to reuse prior responses for requests with equivalent meaning, which is a different requirement from the one being tested.

Answer C is incorrect because prompt compression is primarily used to express required instructions and context with fewer tokens without materially changing intent, which is a different requirement from the one being tested.

 

Question 3

The supply-chain analytics group needs an architecture that can express required instructions and context with fewer tokens without materially changing intent. Which choice best meets that need?

1. batched inference
2. response length control
3. deterministic request-hash caching
4. prompt compression

Correct Answer: D

 

Correct Answer

Answer D is correct because prompt compression is designed to express required instructions and context with fewer tokens without materially changing intent. It reduces recurring prompt overhead and can lower inference cost for high-volume workloads.

Incorrect Answers

Answer A is incorrect because batched inference is primarily used to improve throughput efficiency when multiple independent requests can be processed together, which is a different requirement from the one being tested.

Answer B is incorrect because response length control is primarily used to prevent models from generating unnecessarily long completions, which is a different requirement from the one being tested.

Answer C is incorrect because deterministic request-hash caching is primarily used to reuse results only when the normalized request is exactly equivalent, which is a different requirement from the one being tested.

 

Question 4

The enterprise developer platform needs an architecture that can prevent models from generating unnecessarily long completions. Which choice best meets that need?

1. capacity planning and auto scaling
2. response length control
3. tiered model routing
4. prompt caching

Correct Answer: B

 

Correct Answer

Answer B is correct because response length control is designed to prevent models from generating unnecessarily long completions. It uses maximum output settings or explicit concise-response instructions to bound completion token cost.

Incorrect Answers

Answer A is incorrect because capacity planning and auto scaling is primarily used to align serving resources with changing GenAI traffic patterns, which is a different requirement from the one being tested.

Answer C is incorrect because tiered model routing is primarily used to send simple requests to lower-cost models and reserve premium models for complex work, which is a different requirement from the one being tested.

Answer D is incorrect because prompt caching is primarily used to reuse supported prompt-prefix computation across repeated requests, which is a different requirement from the one being tested.

 

Question 5

The customer identity platform is prioritizing a requirement to send simple requests to lower-cost models and reserve premium models for complex work. Which implementation is most appropriate?

1. Bedrock provisioned throughput optimization
2. cost-capability tradeoff evaluation
3. tiered model routing
4. token estimation and tracking

Correct Answer: C

 

Correct Answer

Answer C is correct because tiered model routing is designed to send simple requests to lower-cost models and reserve premium models for complex work. It classifies request complexity or quality needs before selecting an inference tier.

Incorrect Answers

Answer A is incorrect because Bedrock provisioned throughput optimization is primarily used to reserve model capacity when predictable sustained demand justifies dedicated throughput, which is a different requirement from the one being tested.

Answer B is incorrect because cost-capability tradeoff evaluation is primarily used to choose the least expensive model that still meets the quality and latency target, which is a different requirement from the one being tested.

Answer D is incorrect because token estimation and tracking is primarily used to measure prompt and completion token consumption before and after optimization, which is a different requirement from the one being tested.

 

Question 6

The business intelligence application needs an architecture that can choose the least expensive model that still meets the quality and latency target. Which choice best meets that need?

1. semantic caching
2. context pruning
3. batched inference
4. cost-capability tradeoff evaluation

Correct Answer: D

 

Correct Answer

Answer D is correct because cost-capability tradeoff evaluation is designed to choose the least expensive model that still meets the quality and latency target. It compares measured business value and model performance rather than selecting solely by absolute capability.

Incorrect Answers

Answer A is incorrect because semantic caching is primarily used to reuse prior responses for requests with equivalent meaning, which is a different requirement from the one being tested.

Answer B is incorrect because context pruning is primarily used to remove irrelevant retrieved or conversational content before inference, which is a different requirement from the one being tested.

Answer C is incorrect because batched inference is primarily used to improve throughput efficiency when multiple independent requests can be processed together, which is a different requirement from the one being tested.

 

Question 7

The healthcare document platform is prioritizing a requirement to improve throughput efficiency when multiple independent requests can be processed together. Which implementation is most appropriate?

1. batched inference
2. deterministic request-hash caching
3. prompt compression
4. capacity planning and auto scaling

Correct Answer: A

 

Correct Answer

Answer A is correct because batched inference is designed to improve throughput efficiency when multiple independent requests can be processed together. It groups compatible work to reduce per-request overhead and improve resource utilization.

Incorrect Answers

Answer B is incorrect because deterministic request-hash caching is primarily used to reuse results only when the normalized request is exactly equivalent, which is a different requirement from the one being tested.

Answer C is incorrect because prompt compression is primarily used to express required instructions and context with fewer tokens without materially changing intent, which is a different requirement from the one being tested.

Answer D is incorrect because capacity planning and auto scaling is primarily used to align serving resources with changing GenAI traffic patterns, which is a different requirement from the one being tested.

 

Question 8

The e-commerce search team must align serving resources with changing GenAI traffic patterns. Which approach is the strongest fit?

1. prompt caching
2. Bedrock provisioned throughput optimization
3. capacity planning and auto scaling
4. response length control

Correct Answer: C

 

Correct Answer

Answer C is correct because capacity planning and auto scaling is designed to align serving resources with changing GenAI traffic patterns. It uses measured demand and utilization to avoid both chronic overprovisioning and sustained undercapacity.

Incorrect Answers

Answer A is incorrect because prompt caching is primarily used to reuse supported prompt-prefix computation across repeated requests, which is a different requirement from the one being tested.

Answer B is incorrect because Bedrock provisioned throughput optimization is primarily used to reserve model capacity when predictable sustained demand justifies dedicated throughput, which is a different requirement from the one being tested.

Answer D is incorrect because response length control is primarily used to prevent models from generating unnecessarily long completions, which is a different requirement from the one being tested.

 

Question 9

The risk analytics team has a design goal to reserve model capacity when predictable sustained demand justifies dedicated throughput. What should the team choose?

1. Bedrock provisioned throughput optimization
2. token estimation and tracking
3. tiered model routing
4. semantic caching

Correct Answer: A

 

Correct Answer

Answer A is correct because Bedrock provisioned throughput optimization is designed to reserve model capacity when predictable sustained demand justifies dedicated throughput. It matches provisioned model capacity to steady production usage instead of paying for unnecessary reservation or suffering on-demand constraints.

Incorrect Answers

Answer B is incorrect because token estimation and tracking is primarily used to measure prompt and completion token consumption before and after optimization, which is a different requirement from the one being tested.

Answer C is incorrect because tiered model routing is primarily used to send simple requests to lower-cost models and reserve premium models for complex work, which is a different requirement from the one being tested.

Answer D is incorrect because semantic caching is primarily used to reuse prior responses for requests with equivalent meaning, which is a different requirement from the one being tested.

 

Question 10

The fraud detection engineering team must reuse prior responses for requests with equivalent meaning. Which approach is the strongest fit?

1. context pruning
2. cost-capability tradeoff evaluation
3. semantic caching
4. deterministic request-hash caching

Correct Answer: C

 

Correct Answer

Answer C is correct because semantic caching is designed to reuse prior responses for requests with equivalent meaning. It compares semantic similarity so repeated intents can avoid another model invocation when policy permits.

Incorrect Answers

Answer A is incorrect because context pruning is primarily used to remove irrelevant retrieved or conversational content before inference, which is a different requirement from the one being tested.

Answer B is incorrect because cost-capability tradeoff evaluation is primarily used to choose the least expensive model that still meets the quality and latency target, which is a different requirement from the one being tested.

Answer D is incorrect because deterministic request-hash caching is primarily used to reuse results only when the normalized request is exactly equivalent, which is a different requirement from the one being tested.

 

Question 11

The technical documentation assistant needs an architecture that can reuse results only when the normalized request is exactly equivalent. Which choice best meets that need?

1. batched inference
2. prompt compression
3. deterministic request-hash caching
4. prompt caching

Correct Answer: C

 

Correct Answer

Answer C is correct because deterministic request-hash caching is designed to reuse results only when the normalized request is exactly equivalent. It creates a stable cache key from controlled request content so identical inputs map predictably to stored output.

Incorrect Answers

Answer A is incorrect because batched inference is primarily used to improve throughput efficiency when multiple independent requests can be processed together, which is a different requirement from the one being tested.

Answer B is incorrect because prompt compression is primarily used to express required instructions and context with fewer tokens without materially changing intent, which is a different requirement from the one being tested.

Answer D is incorrect because prompt caching is primarily used to reuse supported prompt-prefix computation across repeated requests, which is a different requirement from the one being tested.

 

Question 12

The content intelligence platform needs an architecture that can reuse supported prompt-prefix computation across repeated requests. Which choice best meets that need?

1. response length control
2. capacity planning and auto scaling
3. token estimation and tracking
4. prompt caching

Correct Answer: D

 

Correct Answer

Answer D is correct because prompt caching is designed to reuse supported prompt-prefix computation across repeated requests. It reduces repeated processing for stable prompt context when the model service supports prompt caching.

Incorrect Answers

Answer A is incorrect because response length control is primarily used to prevent models from generating unnecessarily long completions, which is a different requirement from the one being tested.

Answer B is incorrect because capacity planning and auto scaling is primarily used to align serving resources with changing GenAI traffic patterns, which is a different requirement from the one being tested.

Answer C is incorrect because token estimation and tracking is primarily used to measure prompt and completion token consumption before and after optimization, which is a different requirement from the one being tested.

 

Question 13

The AI governance program needs to identify a capability with this behavior: creates a quantitative baseline for cost, context size, and usage trends instead of optimizing blindly. What is the best match?

1. Bedrock provisioned throughput optimization
2. token estimation and tracking
3. context pruning
4. tiered model routing

Correct Answer: B

 

Correct Answer

Answer B is correct because the description directly matches token estimation and tracking. token estimation and tracking creates a quantitative baseline for cost, context size, and usage trends instead of optimizing blindly.

Incorrect Answers

Answer A is incorrect because Bedrock provisioned throughput optimization is primarily used to reserve model capacity when predictable sustained demand justifies dedicated throughput, which is a different requirement from the one being tested.

Answer C is incorrect because context pruning is primarily used to remove irrelevant retrieved or conversational content before inference, which is a different requirement from the one being tested.

Answer D is incorrect because tiered model routing is primarily used to send simple requests to lower-cost models and reserve premium models for complex work, which is a different requirement from the one being tested.

 

Question 14

The sales operations automation team documents this GenAI behavior: reduces input token volume while retaining information needed for the current request. Which capability matches it?

1. semantic caching
2. cost-capability tradeoff evaluation
3. context pruning
4. prompt compression

Correct Answer: C

 

Correct Answer

Answer C is correct because the description directly matches context pruning. context pruning reduces input token volume while retaining information needed for the current request.

Incorrect Answers

Answer A is incorrect because semantic caching is primarily used to reuse prior responses for requests with equivalent meaning, which is a different requirement from the one being tested.

Answer B is incorrect because cost-capability tradeoff evaluation is primarily used to choose the least expensive model that still meets the quality and latency target, which is a different requirement from the one being tested.

Answer D is incorrect because prompt compression is primarily used to express required instructions and context with fewer tokens without materially changing intent, which is a different requirement from the one being tested.

 

Question 15

The knowledge retrieval engineering team documents this GenAI behavior: reduces recurring prompt overhead and can lower inference cost for high-volume workloads. Which capability matches it?

1. deterministic request-hash caching
2. batched inference
3. prompt compression
4. response length control

Correct Answer: C

 

Correct Answer

Answer C is correct because the description directly matches prompt compression. prompt compression reduces recurring prompt overhead and can lower inference cost for high-volume workloads.

Incorrect Answers

Answer A is incorrect because deterministic request-hash caching is primarily used to reuse results only when the normalized request is exactly equivalent, which is a different requirement from the one being tested.

Answer B is incorrect because batched inference is primarily used to improve throughput efficiency when multiple independent requests can be processed together, which is a different requirement from the one being tested.

Answer D is incorrect because response length control is primarily used to prevent models from generating unnecessarily long completions, which is a different requirement from the one being tested.

 

Question 16

The enterprise content repository team describes a component that uses maximum output settings or explicit concise-response instructions to bound completion token cost. Which capability is being described?

1. prompt caching
2. capacity planning and auto scaling
3. tiered model routing
4. response length control

Correct Answer: D

 

Correct Answer

Answer D is correct because the description directly matches response length control. response length control uses maximum output settings or explicit concise-response instructions to bound completion token cost.

Incorrect Answers

Answer A is incorrect because prompt caching is primarily used to reuse supported prompt-prefix computation across repeated requests, which is a different requirement from the one being tested.

Answer B is incorrect because capacity planning and auto scaling is primarily used to align serving resources with changing GenAI traffic patterns, which is a different requirement from the one being tested.

Answer C is incorrect because tiered model routing is primarily used to send simple requests to lower-cost models and reserve premium models for complex work, which is a different requirement from the one being tested.

 

Question 17

The manufacturing analytics team describes a component that classifies request complexity or quality needs before selecting an inference tier. Which capability is being described?

1. token estimation and tracking
2. Bedrock provisioned throughput optimization
3. cost-capability tradeoff evaluation
4. tiered model routing

Correct Answer: D

 

Correct Answer

Answer D is correct because the description directly matches tiered model routing. tiered model routing classifies request complexity or quality needs before selecting an inference tier.

Incorrect Answers

Answer A is incorrect because token estimation and tracking is primarily used to measure prompt and completion token consumption before and after optimization, which is a different requirement from the one being tested.

Answer B is incorrect because Bedrock provisioned throughput optimization is primarily used to reserve model capacity when predictable sustained demand justifies dedicated throughput, which is a different requirement from the one being tested.

Answer C is incorrect because cost-capability tradeoff evaluation is primarily used to choose the least expensive model that still meets the quality and latency target, which is a different requirement from the one being tested.

 

Question 18

The developer productivity team needs to identify a capability with this behavior: compares measured business value and model performance rather than selecting solely by absolute capability. What is the best match?

1. cost-capability tradeoff evaluation
2. semantic caching
3. context pruning
4. batched inference

Correct Answer: A

 

Correct Answer

Answer A is correct because the description directly matches cost-capability tradeoff evaluation. cost-capability tradeoff evaluation compares measured business value and model performance rather than selecting solely by absolute capability.

Incorrect Answers

Answer B is incorrect because semantic caching is primarily used to reuse prior responses for requests with equivalent meaning, which is a different requirement from the one being tested.

Answer C is incorrect because context pruning is primarily used to remove irrelevant retrieved or conversational content before inference, which is a different requirement from the one being tested.

Answer D is incorrect because batched inference is primarily used to improve throughput efficiency when multiple independent requests can be processed together, which is a different requirement from the one being tested.

 

Question 19

The business process automation team describes a component that groups compatible work to reduce per-request overhead and improve resource utilization. Which capability is being described?

1. capacity planning and auto scaling
2. batched inference
3. deterministic request-hash caching
4. prompt compression

Correct Answer: B

 

Correct Answer

Answer B is correct because the description directly matches batched inference. batched inference groups compatible work to reduce per-request overhead and improve resource utilization.

Incorrect Answers

Answer A is incorrect because capacity planning and auto scaling is primarily used to align serving resources with changing GenAI traffic patterns, which is a different requirement from the one being tested.

Answer C is incorrect because deterministic request-hash caching is primarily used to reuse results only when the normalized request is exactly equivalent, which is a different requirement from the one being tested.

Answer D is incorrect because prompt compression is primarily used to express required instructions and context with fewer tokens without materially changing intent, which is a different requirement from the one being tested.

 

Question 20

Within the regulatory reporting platform’s architecture, which capability matches this technical description: uses measured demand and utilization to avoid both chronic overprovisioning and sustained undercapacity?

1. capacity planning and auto scaling
2. prompt caching
3. response length control
4. Bedrock provisioned throughput optimization

Correct Answer: A

 

Correct Answer

Answer A is correct because the description directly matches capacity planning and auto scaling. capacity planning and auto scaling uses measured demand and utilization to avoid both chronic overprovisioning and sustained undercapacity.

Incorrect Answers

Answer B is incorrect because prompt caching is primarily used to reuse supported prompt-prefix computation across repeated requests, which is a different requirement from the one being tested.

Answer C is incorrect because response length control is primarily used to prevent models from generating unnecessarily long completions, which is a different requirement from the one being tested.

Answer D is incorrect because Bedrock provisioned throughput optimization is primarily used to reserve model capacity when predictable sustained demand justifies dedicated throughput, which is a different requirement from the one being tested.

 

Question 21

The privacy engineering function documents this GenAI behavior: matches provisioned model capacity to steady production usage instead of paying for unnecessary reservation or suffering on-demand constraints. Which capability matches it?

1. semantic caching
2. Bedrock provisioned throughput optimization
3. token estimation and tracking
4. tiered model routing

Correct Answer: B

 

Correct Answer

Answer B is correct because the description directly matches Bedrock provisioned throughput optimization. Bedrock provisioned throughput optimization matches provisioned model capacity to steady production usage instead of paying for unnecessary reservation or suffering on-demand constraints.

Incorrect Answers

Answer A is incorrect because semantic caching is primarily used to reuse prior responses for requests with equivalent meaning, which is a different requirement from the one being tested.

Answer C is incorrect because token estimation and tracking is primarily used to measure prompt and completion token consumption before and after optimization, which is a different requirement from the one being tested.

Answer D is incorrect because tiered model routing is primarily used to send simple requests to lower-cost models and reserve premium models for complex work, which is a different requirement from the one being tested.

 

Question 22

The logistics optimization team documents this GenAI behavior: compares semantic similarity so repeated intents can avoid another model invocation when policy permits. Which capability matches it?

1. semantic caching
2. context pruning
3. deterministic request-hash caching
4. cost-capability tradeoff evaluation

Correct Answer: A

 

Correct Answer

Answer A is correct because the description directly matches semantic caching. semantic caching compares semantic similarity so repeated intents can avoid another model invocation when policy permits.

Incorrect Answers

Answer B is incorrect because context pruning is primarily used to remove irrelevant retrieved or conversational content before inference, which is a different requirement from the one being tested.

Answer C is incorrect because deterministic request-hash caching is primarily used to reuse results only when the normalized request is exactly equivalent, which is a different requirement from the one being tested.

Answer D is incorrect because cost-capability tradeoff evaluation is primarily used to choose the least expensive model that still meets the quality and latency target, which is a different requirement from the one being tested.

 

Question 23

Within the service reliability organization’s architecture, which capability matches this technical description: creates a stable cache key from controlled request content so identical inputs map predictably to stored output?

1. prompt compression
2. deterministic request-hash caching
3. batched inference
4. prompt caching

Correct Answer: B

 

Correct Answer

Answer B is correct because the description directly matches deterministic request-hash caching. deterministic request-hash caching creates a stable cache key from controlled request content so identical inputs map predictably to stored output.

Incorrect Answers

Answer A is incorrect because prompt compression is primarily used to express required instructions and context with fewer tokens without materially changing intent, which is a different requirement from the one being tested.

Answer C is incorrect because batched inference is primarily used to improve throughput efficiency when multiple independent requests can be processed together, which is a different requirement from the one being tested.

Answer D is incorrect because prompt caching is primarily used to reuse supported prompt-prefix computation across repeated requests, which is a different requirement from the one being tested.

 

Question 24

A model service can reuse processing of an unchanged prompt prefix across requests. Which capability reduces repeated work for this stable context?

1. prompt caching
2. response length control
3. token estimation and tracking
4. capacity planning and auto scaling

Correct Answer: A

 

Correct Answer

Answer A is correct because the description directly matches prompt caching. prompt caching reduces repeated processing for stable prompt context when the model service supports prompt caching.

Incorrect Answers

Answer B is incorrect because response length control is primarily used to prevent models from generating unnecessarily long completions, which is a different requirement from the one being tested.

Answer C is incorrect because token estimation and tracking is primarily used to measure prompt and completion token consumption before and after optimization, which is a different requirement from the one being tested.

Answer D is incorrect because capacity planning and auto scaling is primarily used to align serving resources with changing GenAI traffic patterns, which is a different requirement from the one being tested.

 

Question 25

The risk and controls team has proposed token estimation and tracking for its design. Which requirement best justifies it?

1. Send simple requests to lower-cost models and reserve premium models for complex work
2. Measure prompt and completion token consumption before and after optimization
3. Reserve model capacity when predictable sustained demand justifies dedicated throughput
4. Remove irrelevant retrieved or conversational content before inference

Correct Answer: B

 

Correct Answer

Answer B is correct because token estimation and tracking is specifically used to measure prompt and completion token consumption before and after optimization. It creates a quantitative baseline for cost, context size, and usage trends instead of optimizing blindly.

Incorrect Answers

Answer A is incorrect because that requirement aligns with tiered model routing, not token estimation and tracking.

Answer C is incorrect because that requirement aligns with Bedrock provisioned throughput optimization, not token estimation and tracking.

Answer D is incorrect because that requirement aligns with context pruning, not token estimation and tracking.

 

Question 26

Why would the enterprise search architecture group introduce context pruning into the GenAI architecture?

1. Remove irrelevant retrieved or conversational content before inference
2. Reuse prior responses for requests with equivalent meaning
3. Choose the least expensive model that still meets the quality and latency target
4. Express required instructions and context with fewer tokens without materially changing intent

Correct Answer: A

 

Correct Answer

Answer A is correct because context pruning is specifically used to remove irrelevant retrieved or conversational content before inference. It reduces input token volume while retaining information needed for the current request.

Incorrect Answers

Answer B is incorrect because that requirement aligns with semantic caching, not context pruning.

Answer C is incorrect because that requirement aligns with cost-capability tradeoff evaluation, not context pruning.

Answer D is incorrect because that requirement aligns with prompt compression, not context pruning.

 

Question 27

The insurance automation team is considering prompt compression. What problem is this choice primarily meant to solve?

1. Prevent models from generating unnecessarily long completions
2. Express required instructions and context with fewer tokens without materially changing intent
3. Improve throughput efficiency when multiple independent requests can be processed together
4. Reuse results only when the normalized request is exactly equivalent

Correct Answer: B

 

Correct Answer

Answer B is correct because prompt compression is specifically used to express required instructions and context with fewer tokens without materially changing intent. It reduces recurring prompt overhead and can lower inference cost for high-volume workloads.

Incorrect Answers

Answer A is incorrect because that requirement aligns with response length control, not prompt compression.

Answer C is incorrect because that requirement aligns with batched inference, not prompt compression.

Answer D is incorrect because that requirement aligns with deterministic request-hash caching, not prompt compression.

 

Question 28

The procurement automation group is considering response length control. What problem is this choice primarily meant to solve?

1. Send simple requests to lower-cost models and reserve premium models for complex work
2. Align serving resources with changing GenAI traffic patterns
3. Prevent models from generating unnecessarily long completions
4. Reuse supported prompt-prefix computation across repeated requests

Correct Answer: C

 

Correct Answer

Answer C is correct because response length control is specifically used to prevent models from generating unnecessarily long completions. It uses maximum output settings or explicit concise-response instructions to bound completion token cost.

Incorrect Answers

Answer A is incorrect because that requirement aligns with tiered model routing, not response length control.

Answer B is incorrect because that requirement aligns with capacity planning and auto scaling, not response length control.

Answer D is incorrect because that requirement aligns with prompt caching, not response length control.

 

Question 29

The observability engineering group has proposed tiered model routing for its design. Which requirement best justifies it?

1. Choose the least expensive model that still meets the quality and latency target
2. Send simple requests to lower-cost models and reserve premium models for complex work
3. Reserve model capacity when predictable sustained demand justifies dedicated throughput
4. Measure prompt and completion token consumption before and after optimization

Correct Answer: B

 

Correct Answer

Answer B is correct because tiered model routing is specifically used to send simple requests to lower-cost models and reserve premium models for complex work. It classifies request complexity or quality needs before selecting an inference tier.

Incorrect Answers

Answer A is incorrect because that requirement aligns with cost-capability tradeoff evaluation, not tiered model routing.

Answer C is incorrect because that requirement aligns with Bedrock provisioned throughput optimization, not tiered model routing.

Answer D is incorrect because that requirement aligns with token estimation and tracking, not tiered model routing.

 

Question 30

The managed services operations team has proposed cost-capability tradeoff evaluation for its design. Which requirement best justifies it?

1. Choose the least expensive model that still meets the quality and latency target
2. Reuse prior responses for requests with equivalent meaning
3. Remove irrelevant retrieved or conversational content before inference
4. Improve throughput efficiency when multiple independent requests can be processed together

Correct Answer: A

 

Correct Answer

Answer A is correct because cost-capability tradeoff evaluation is specifically used to choose the least expensive model that still meets the quality and latency target. It compares measured business value and model performance rather than selecting solely by absolute capability.

Incorrect Answers

Answer B is incorrect because that requirement aligns with semantic caching, not cost-capability tradeoff evaluation.

Answer C is incorrect because that requirement aligns with context pruning, not cost-capability tradeoff evaluation.

Answer D is incorrect because that requirement aligns with batched inference, not cost-capability tradeoff evaluation.