Exams / MLA-C02 / Study guide / Choosing foundation models in Bedrock

Choosing foundation models in Bedrock: MLA-C02 study notes and practice questions

What to know about Choosing foundation models in Bedrock for the MLA-C02 exam, from the ML Model and Foundation Model (FM) Development domain (24% of the exam guide), then 1 free question to test it. Part of the MLA-C02 study guide.

Last verified October 2026 against current AWS documentation ·How we write our questions

What to know for the exam

Model choice questions describe a generative AI use case and ask which model or approach to pick and how to decide.

Criteria

Criterion	Questions to ask
Task quality	How well does it do on our prompts and data? Test, don’t assume
Modality	Text, images, documents, audio, video, embeddings
Context window	How much input (documents, chat history) must fit
Latency	Interactive chat needs fast first tokens; batch jobs don’t
Cost	Price per input and output token; smaller models are much cheaper
Customization	Does it support fine-tuning, continued pre-training or distillation in Bedrock?
Languages and Region	Supported languages and availability in your Regions
Features	Tool use, structured output, guardrail and knowledge base support

How to compare

Bedrock model evaluation runs your prompt dataset through candidate models with automatic metrics, an LLM as a judge, or human reviewers.
The playground helps with quick side-by-side checks.
Start with a capable model to prove the task, then try smaller models to cut cost and latency.

Ways to reduce cost without switching models

Prompt caching for repeated long context.
Batch inference at a discount for non-urgent jobs.
Intelligent prompt routing sends simple prompts to a smaller model in the same family.
Distillation to a smaller model.
Cross-Region inference improves throughput and availability.

Common traps

Always choosing the largest model. It costs more and responds slower.
Picking on a public leaderboard instead of your own data.
Training a model from scratch when a foundation model with prompting or RAG works.

Try the 1 practice question

Practice Choosing foundation models in Bedrock Take the free practice exam

Question 1

Medium ·Link to this question

An esports league runs a fan chatbot that answers short questions about schedules, rules and team rosters. It expects about 200,000 questions a day, and most answers must start appearing within one second.

Answer quality must stay high for these simple questions, and the monthly model cost must stay low. The team plans to choose a foundation model in Amazon Bedrock and test it on its own question set.

Which approach should the team take to choose the model?

Choose one answer.

A. Shortlist smaller Bedrock models, run each on the team's labeled question set, and pick the cheapest one that meets the quality bar at the latency target.
B. Choose the largest foundation model available in Bedrock, since the biggest model gives the best answers for every question.
C. Train a custom language model from scratch in SageMaker AI on the league's past chat logs, and test foundation models only afterward.
D. Use the model with the largest context window, because a larger window always reduces latency for short questions.

Show the answer and explanation

Answer: A

A. Right: Correct. Evaluating candidate models on the team's own questions measures quality for the real workload, and comparing latency and cost narrows the choice. This matches how Bedrock model choice should be made. Learn more: Evaluate the performance of Amazon Bedrock resources (opens in a new tab)
B. Wrong: The largest model usually responds more slowly and costs more per request, which conflicts with the one-second and cost goals. Size alone does not show quality on these simple questions.
C. Wrong: Training from scratch is costly and slow, and it is unnecessary for simple questions that existing models already answer well. The stem asks for a fast, cheap choice from available models.
D. Wrong: A large context window does not make short questions faster, and the window size is unrelated to answer quality on this task. Context size should match the amount of text the app sends.

Explanation

Model choice for a Bedrock workload should weigh quality on the team's own data against latency and cost. The largest model or a from-scratch model adds cost and delay, and context window size does not speed up short answers.

Study

Want more on Choosing foundation models in Bedrock? The 5 full-length practice sets have 8 more Choosing foundation models in Bedrock questions, 325 in all, none of them in the free exam. One payment of $14.99, no subscription. See the practice sets

Study Choosing foundation models in Bedrock

Amazon Bedrock overview