CCA-F Domain 3 Study Guide: Product and Model Selection

Domain weight: 12% (approximately 7 of 60 items). Based on Exam Guide Version 1.0, effective July 2026; study-guide edition updated August 1, 2026.

Domain purpose

Domain 3 tests whether an Associate can choose the right Claude surface, feature, and model for a task while balancing quality, speed, cost, continuity, and context. It is a selection domain: the best answer fits the actual work instead of automatically choosing the most powerful option.

This guide does not repeat prompt construction or output validation. It focuses on where the work should happen, which capability should support it, and when context should be continued, summarized, persisted, or restarted.

Official objectives

Candidates are expected to:

• •Select appropriate Claude product features: Projects, research mode, chat, and artifacts.
• •Differentiate between Claude model types: Haiku, Sonnet, and Opus.
• •Align model selection with task requirements: cost, speed, and quality.
• •Understand and manage context limitations and memory considerations, including when to restart, summarize, or persist.

Selection framework: surface

Use SURFACE:

S - Scope: one question, recurring work, deep research, or reusable deliverable?

U - Urgency: immediate response or time for deeper work?

R - Reasoning: straightforward or complex and ambiguous?

F - Files and facts: supplied documents, connected sources, or current web information?

A - Audience and artifact: conversational answer or standalone output?

C - Continuity: should knowledge persist across related work?

E - Economy: what quality, time, and usage budget is justified?

SURFACE is a study framework, not an official Anthropic acronym.

Choose the product feature

4.1 Chat

Choose ordinary chat for:

• •a one-time question;
• •quick drafting or explanation;
• •short analysis with supplied context;
• •exploratory conversation;
• •work that does not need a persistent project knowledge base.

Chat is often the simplest correct choice. Do not create a Project for every isolated task.

4.2 Projects

Anthropic describes Projects as self-contained workspaces with their own chat histories and knowledge bases. Use a Project when several conversations share a stable purpose, instructions, files, or team context.

Good examples:

• •recurring campaign work using brand guidelines;
• •a semester course with syllabus and reading material;
• •an ongoing operations initiative;
• •repeated analysis of an approved policy collection.

Projects reduce repeated setup, but their knowledge must be maintained. A Project filled with obsolete or unrelated files can reduce usefulness.

4.3 Research

Use Research for deep, multi-step investigation that benefits from broad searching and synthesis across web or connected sources. It takes longer and may consume more usage than quick search or chat.

Use it when:

• •the question is complex and evidence-heavy;
• •current information matters;
• •several sources must be compared;
• •a documented research trail is valuable.

Do not use deep Research for a simple fact that can be found quickly or for a task answerable from supplied documents.

4.4 Artifacts

Use an Artifact when Claude is creating substantial standalone content that the user will edit, reuse, share, or interact with. Examples include reports, tools, visualizations, prototypes, and structured content.

Use an inline response for a short answer. Use an Artifact or downloadable file when separation from the conversation materially helps the work.

4.5 Files and file creation

Claude can create and edit common business files when the relevant capability is enabled. Select file creation when the required deliverable is actually a spreadsheet, presentation, document, or PDF rather than prose to be copied manually.

4.6 Enterprise search versus Research

Current help guidance distinguishes quick retrieval from connected internal sources from deep, multi-step Research. Choose quick enterprise search for locating everyday internal knowledge; choose Research when the problem requires extended investigation and synthesis.

Model families and avoiding stale tables

The official blueprint names Haiku, Sonnet, and Opus. These are durable model-family concepts:

Haiku:

Generally optimized for speed and cost on straightforward work.

Sonnet:

Generally offers a strong balance of intelligence, speed, and cost for broad professional work.

Opus:

Generally intended for the most complex reasoning and demanding work where quality justifies additional time or cost.

As of August 1, 2026, Anthropic's public model overview also lists newer offerings and generations, including Fable. Availability can vary by product, plan, and platform. The exam guide's explicit objective remains differentiation among Haiku, Sonnet, and Opus, so learn the selection principle rather than memorizing a fragile list of model IDs or prices.

Always verify the live model selector and current official comparison page before making a production or budget decision.

Match model to task

6.1 Start with requirements

Define:

• •minimum acceptable quality;
• •complexity and ambiguity;
• •response-time need;
• •volume and usage budget;
• •consequence of error;
• •file or context size;
• •need for repeated iteration.

6.2 Typical choices

High-volume, straightforward drafts or classifications:

Start by evaluating a faster, economical model such as Haiku.

General professional analysis, writing, and routine complex work:

Sonnet is often the balanced starting point.

The hardest reasoning, complex synthesis, or high-value demanding work:

Evaluate Opus when improved quality justifies additional resource use.

These are starting hypotheses, not guarantees. Test representative tasks. A smaller model that reliably meets the requirement is usually more efficient; a stronger model may reduce rework on difficult tasks.

6.3 Do not optimize one dimension alone

Cheapest is not best if errors require extensive correction. Most capable is not best if the task is simple and high-volume. Fastest is not best if quality falls below the business threshold.

Use total task cost:

model usage + employee review time + rework + delay + cost of error.

6.4 Product availability matters

A model or feature may not be available on every plan, region, cloud, or interface. Check the user's actual environment rather than recommending a theoretical option.

Context and memory decisions

7.1 Context is the information available for the current work

It may include the current conversation, uploaded files, Project knowledge, connected sources, and memory. Excess context can become stale or irrelevant; insufficient context produces generic or incorrect work.

7.2 Continue the same chat when

• •the current history is directly relevant;
• •you are refining the same deliverable;
• •prior decisions must remain visible;
• •the conversation is still focused and coherent.

7.3 Summarize when

• •the conversation is long but key decisions remain useful;
• •you need a clean handoff or checkpoint;
• •background details can be compressed into current facts, decisions, and open issues.

A good summary preserves objectives, constraints, decisions, unresolved questions, and next actions.

7.4 Restart when

• •the new task is unrelated;
• •old instructions are interfering;
• •the conversation contains conflicting directions;
• •repeated correction is failing because the history is polluted;
• •sensitive work should be isolated according to policy.

7.5 Persist when

• •information should support repeated related work;
• •the source is stable and approved;
• •team members need shared context;
• •future sessions should not depend on one chat.

Persist through an appropriate Project, maintained source file, or approved knowledge system. Do not treat conversational memory as the sole system of record.

7.6 Memory scope

Current Claude help guidance distinguishes standalone memory from Project memory. Project memory is scoped to that Project. Incognito chats do not use or contribute to memory. Availability and behavior depend on plan and settings.

Exam trap: memory is not the same as a verified knowledge base, and a new chat is not automatically context-free if memory is enabled.

Common traps

• •Choosing Opus for every task.
• •Choosing Haiku solely because it is cheaper without checking quality.
• •Using deep Research for a simple lookup.
• •Using chat for recurring work that needs stable shared knowledge.
• •Creating an Artifact for a two-sentence answer.
• •Copying content manually when file creation matches the deliverable.
• •Continuing a polluted chat indefinitely.
• •Restarting when important context has not been summarized or persisted.
• •Assuming memory is complete, current, or authoritative.
• •Memorizing model prices and IDs as permanent facts.
• •Ignoring plan and feature availability.
• •Measuring model cost but not rework and review time.

Seven original practice questions

These are original and not from the live exam.

• •A user needs 500 short, routine reply drafts and prioritizes speed and cost. Best starting choice?
• •A: Always Opus B. Evaluate Haiku C. Deep Research D. Artifact

Answer: B. Test the economical model against the quality threshold.

• •A team repeatedly creates campaigns using the same approved brand files. Best surface?
• •A: New unrelated chat each time B. Project C. Incognito chat D. Deep Research only

Answer: B. A Project provides persistent related knowledge and instructions.

• •A manager needs a multi-source investigation of a current market change with citations. Best feature?
• •A: Research B. Calculator only C. Memory only D. Static Artifact without research

Answer: A.

• •Select TWO reasons to start a new chat:
• •A: New task is unrelated B. Old conflicting instructions keep affecting results C. The same draft needs one more revision D. Prior decisions are still essential

Answers: A and B.

• •A short policy definition is needed immediately. Best output location?
• •A: Inline chat B. Interactive Artifact C. Full presentation D. New Project necessarily

Answer: A.

• •A difficult board-level synthesis repeatedly fails on a faster model. Best next step?
• •A: Continue solely because it is cheaper B. Evaluate a more capable model on representative cases C. Remove sources D. Ignore review cost

Answer: B.

• •Select TWO items a useful context summary should preserve:
• •A: Current decisions and constraints B. Open issues and next actions C. Every irrelevant conversational detail D. Obsolete instructions as active rules

Answers: A and B.

Seven-day study plan

Day 1: Map ten tasks to chat, Project, Research, Artifact, or file creation.

Day 2: Compare Haiku, Sonnet, and Opus using current official descriptions.

Day 3: Build a cost-speed-quality decision matrix for five workplace tasks.

Day 4: Practice continue, summarize, restart, and persist decisions.

Day 5: Test one representative task on available models and record quality, time, and rework.

Day 6: Answer the seven questions and create seven variations.

Day 7: Review live product/model documentation and create a one-page selection sheet.

Use related blogs for additional scenarios, but update model and feature claims from official sources before study or publication.

Readiness checklist

• •I can choose chat, Project, Research, Artifact, or file creation.
• •I understand the durable Haiku, Sonnet, and Opus trade-offs.
• •I balance quality, speed, cost, volume, risk, and rework.
• •I verify current availability rather than relying on stale tables.
• •I know when to continue, summarize, restart, or persist.
• •I understand that memory scope and availability vary.
• •I can justify a selection in one clear sentence.

Sources

Primary: Claude Certified Associate - Foundations Exam Guide, Version 1.0, effective July 2026.

Models overview:

https://platform.claude.com/docs/en/about-claude/models/overview

Choosing the right model:

https://platform.claude.com/docs/en/about-claude/models/choosing-a-model

Projects:

https://support.claude.com/en/articles/9517075-what-are-projects

Project management and memory:

https://support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects

Chat search and memory:

https://support.claude.com/en/articles/11817273-use-claude-s-chat-search-and-memory-to-build-on-previous-context

Artifacts:

https://support.claude.com/en/articles/9487310-what-are-artifacts-and-how-do-i-use-them

Create and edit files:

https://support.claude.com/en/articles/12111783-create-and-edit-files-with-claude

Note: Product features and models change quickly. Verify the current help center and model comparison page.