Unofficial and not affiliated with Anthropic. Written from Anthropic's exam guides for CCAO-F and CCAR-F and Anthropic's public documentation, checked 12 October 2026. Exam details change: confirm everything below against the current exam guide before you book.
Practice with the companion drill: Claude Cert Drill.
Exam facts
| Exam code | CCAO-F |
|---|---|
| Items | 60 |
| Format | Multiple choice and multiple response (each item says how many to pick) |
| Time | 120 minutes, about 2 minutes per item |
| Passing score | 720 on a scaled range of 100 to 1,000 |
| Delivery | Pearson VUE, online proctored or at a test center |
| Fee | $99 list |
| Validity | 12 months |
Two things to know up front.
The credential lasts 12 months, not the two or three years many certifications give. Renewal on time is a free, unproctored assessment covering what changed. Let it lapse and you sit the full exam again. Put the renewal date on your calendar the day you pass.
Multiple-response items are where prepared candidates lose points. Each one says how many answers to pick. Read that number every time.
Architect exam facts
| Exam code | CCAR-F |
|---|---|
| Items | 60 |
| Format | Scenario-based: 4 scenarios drawn at random from a published bank of 6. Multiple choice and multiple response (each item says how many to pick) |
| Time | 120 minutes |
| Passing score | 720 on a scaled range of 100 to 1,000 |
| Delivery | Pearson VUE, online proctored or at a test center |
| Fee | $125 list |
| Validity | 12 months |
| # | Domain | Weight |
|---|---|---|
| 1 | Agentic Architecture and Orchestration | 27% |
| 2 | Tool Design and MCP Integration | 18% |
| 3 | Claude Code Configuration and Workflows | 20% |
| 4 | Prompt Engineering and Structured Output | 20% |
| 5 | Context Management and Reliability | 15% |
The two blueprints test different jobs. The Associate is about everyday use and judgment, while the Architect is scenario-based and about building agents, tools and Claude Code workflows.
The blueprint, and where the risk usually is
| # | Domain | Weight |
|---|---|---|
| 1 | Prompting and Task Execution | 14% |
| 2 | Output Evaluation and Validation | 21% |
| 3 | Product and Model Selection | 12% |
| 4 | Workflow Integration and Solution Design | 16% |
| 5 | Configuration and Knowledge Management | 12% |
| 6 | Governance, Risk, and Responsible Use | 15% |
| 7 | Troubleshooting and Optimization | 10% |
Calibrate first. This is not a security exam and not a developer exam. The candidate it is written for uses Claude every day in operations, marketing, project management or a similar role. Good judgment about work carries you a long way.
Where experienced people slip. Daily use does not guarantee you know the whole feature surface. Domains 3 and 5, 24% together, test exactly that: model tier tradeoffs, context and memory, Projects configuration, connectors and instructions. Most candidates' study hours belong there.
Study plan, in priority order
First: Domains 3 and 5 (24%, the feature surface)
Domain 3, Product and Model Selection. Choosing among Claude's features (Projects, research, chat, artifacts), telling the model tiers apart, matching a model to cost, speed and quality needs, and managing context and memory: when to restart, summarize or carry forward.
- Read the current model overview in Anthropic's documentation. Lineups change, so read it now rather than trusting what you remember.
- For each tier, be able to say in one sentence what kind of task it fits. High-volume, low-complexity drafting belongs on a faster, cheaper tier, not the most capable one.
- Get precise on context limits and memory, and on when to start a fresh conversation versus summarize and carry forward. That judgment shows up as scenario items.
Domain 5, Configuration and Knowledge Management. Setting up Projects with instructions and knowledge, managing uploaded knowledge and connectors, writing effective instructions, and keeping configurations current.
- Go past your personal habits into the full feature surface: what a Project accepts as knowledge, how connectors attach, and how project instructions differ from steering inside one conversation.
- Build one throwaway Project from scratch with instructions, an uploaded file and a connector, just to touch every setting once.
- Don't skim the maintenance objective. Knowing when instructions and knowledge go stale is its own testable idea.
Primary sources: Claude Help Center (opens in a new tab) for app features, and Claude Docs (opens in a new tab) for models and the platform. Review Projects, Artifacts, Memory, Skills and Code Execution in particular.
Second: Domain 2 (21%, the largest domain)
Judging outputs for accuracy and completeness, spotting hallucinations, inconsistency and bias, fact-checking, deciding when a human must review, adapting output for an audience, and choosing the output format.
- Learn hallucination patterns, especially fabricated specifics that look authoritative, like citation numbers or exact figures.
- Treat format choice (an artifact, inline text or structured data) as a judgment call. It is a product-feature question dressed as an evaluation question, and it is easy to under-study.
Third: Domains 1, 4 and 7 (40%, confirm rather than learn)
Prompting, workflow integration and stakeholder communication, and troubleshooting prompts that underperform. Read each objective and rate yourself honestly. Anything unfamiliar becomes a study item; the rest you likely do already.
Last: Domain 6 (15%, verify)
Appropriate and inappropriate uses, data sensitivity and privacy, organizational AI policy and ethics. If you work under an AI use policy already, read the objectives, confirm nothing surprises you and spend the time elsewhere.
Resources, in order of authority
- The official exam guide for CCAO-F. Its blueprint section is the checklist: rate yourself against every objective. Its sample items show the question style.
- Anthropic's prep course for CCAO-F. The primary prep material where you have access to it.
- Anthropic's documentation: the Claude Help Center (opens in a new tab) and Claude Docs (opens in a new tab).
- People who have sat it. They can describe difficulty and emphasis, but not content.
- Third-party practice sets, including the Claude Cert Drill. Treat all of them, mine included, as unofficial and never as a substitute for items 1 to 3.
Before exam day
- Seat booked at Pearson VUE (opens in a new tab)
- Fee confirmed at checkout
- Government photo ID, unexpired, with your name exactly as registered
- Testing online: desk clear of notes, phone and second monitor
- Rated yourself against every objective in all seven domains
- Built one throwaway Project with instructions, knowledge and a connector
- Read the current model overview within the last week
- One "Timed mock" run on the drill, scored
- Renewal date on the calendar, 12 months from the day you pass
About this guide
By Ken Connell, from my own exam prep. Unofficial and not affiliated with Anthropic; nothing here comes from the live exam.
Changelog
- 12 Oct 2026: Architect exam facts added, exam facts re-checked.
- 30 Sep 2026: first published.