Claude Certification Price, Domains, and What You Get

The claude certification price is $99 per attempt, and that single figure is the starting point for every budget and study plan. But the cost of the exam is only one variable. The Claude Certified Architect, Foundations (CCA-F) exam, launched 12 March 2026, is Anthropic's first professional certification, and understanding what that $99 buys you, what the exam actually tests, and how to prepare without wasting attempts is the real question this post answers.

What is the claude certification price and what does it include?

At $99 per attempt, the CCA-F sits in the mid-range of cloud and AI certifications. Tiered Anthropic partners receive a discounted first attempt through the Claude Partner Network, a $100M programme, so check your organisation's partnership tier before paying full price. The fee covers one sitting of 60 scenario-based multiple-choice questions, delivered either online-proctored or at a test centre.

There is no annual maintenance fee published at this time, and Anthropic has not announced a recertification cadence. What you are paying for is access to a credential that, as of 3 June 2026, fewer than 10,000 individuals worldwide hold, across a partner ecosystem of 40,000+ applicant firms.

How is the exam scored and what is the passing mark?

The exam is scored on a scale of 100 to 1000. The passing score is 720. Anthropic does not publish the raw-to-scaled conversion formula, so we will not speculate on an exact question count required to pass. What we can say is that 720 on a 100-to-1000 scale represents a meaningful but achievable bar, and the exam consistently rewards candidates who favour deterministic solutions over probabilistic ones when stakes are high.

Each question presents one correct answer and three plausible distractors. The distractors are engineered to catch candidates who have memorised patterns rather than understood the underlying reasoning, which is why scenario-based preparation outperforms flashcard drilling.

What are the five domains and their weights?

The exam covers five domains and 30 task statements. Domain weights determine where to invest study time first.

Domain 1, Agentic Architecture & Orchestration, carries the largest single weight at 27%. Candidates who underestimate it, treating it as a niche topic, consistently score below 720. The domain tests hub-and-spoke coordinator patterns, subagent context isolation, parallel spawning, and loop termination logic, all in scenario form.

Domains 3 and 4 are tied at 20% each, making the combined prompt-and-configuration block the largest study surface at 40% of the exam. Domain 5, Context Management & Reliability, is the smallest domain by weight but is the one most candidates neglect, and neglecting 15% of a 60-question exam is a costly error.

What does Domain 1 actually test at 27%?

Agentic Architecture & Orchestration is the exam's heaviest domain because production Claude deployments are almost always multi-agent systems. The task statements cover coordinator responsibilities, dynamic subagent selection, structured context passing, and error routing between agents.

A representative scenario might describe a coordinator that spawns three subagents in parallel, one of which returns a malformed result. The question asks which recovery path is correct: retry the subagent, escalate to a human, or proceed with partial data. The exam rewards proportionate fixes and root-cause tracing rather than blanket retries or silent suppression.

Our Agentic Loop Anti-Patterns concept covers the failure modes the exam tests most frequently in this domain, including premature termination and unbounded retry loops.

How does Domain 2 test Tool Design and MCP Integration at 18%?

Tool Design & MCP Integration tests whether candidates understand that tool descriptions are the primary selection mechanism Claude uses to route calls. A poorly written description causes misrouting; the fix is almost always a description rewrite rather than a system-prompt patch.

The domain also covers the MCP isError flag pattern and structured error metadata. Exam questions distinguish between an access failure (the tool could not reach a resource) and a valid empty result (the resource exists but contains no data). Conflating the two is a common distractor trap.

Structured error responses on the exam typically include fields such as errorCategory and isRetryable. A minimal compliant error payload looks like this:

{
  "isError": true,
  "errorCategory": "access_failure",
  "isRetryable": false,
  "message": "Database connection timed out after 30 s"
}

The isRetryable flag is load-bearing: the exam tests whether candidates know that a transient network error should set it to true, while a permissions failure should set it to false.

What does Domain 3 cover for Claude Code at 20%?

Claude Code Configuration & Workflows tests the three-level CLAUDE.md hierarchy: global user settings, project-level configuration, and directory-level overrides. The exam distinguishes between plan mode, where Claude proposes changes for human approval, and direct execution mode, where it acts immediately.

CI/CD scenarios are common. A typical question describes a pipeline where Claude is invoked with the -p flag for non-interactive execution and asks which configuration scope applies. The answer depends on whether a project-level CLAUDE.md is present and whether .claude/rules/ YAML frontmatter restricts the relevant path.

The Three-Level Configuration Hierarchy concept maps every scope and precedence rule the exam tests in this domain.

How does Domain 4 test Prompt Engineering at 20%?

Prompt Engineering & Structured Output tests schema design, few-shot example construction, and output reliability under adversarial inputs. The exam treats few-shot examples as the highest-leverage technique for ambiguous edge cases, and questions often ask candidates to choose between zero-shot, one-shot, and multi-shot approaches for a given scenario.

A common question type presents a prompt that produces inconsistent JSON and asks which single change would most reliably fix it. The correct answer is almost always adding a typed schema constraint or a well-chosen few-shot example, not increasing temperature or adding more instructions.

Structured output questions also test field-level confidence calibration: when should a field be null versus omitted versus populated with a low-confidence value? The exam has a clear preference for explicit null over silent omission.

What does Domain 5 cover for Context Management at 15%?

Context Management & Reliability tests how candidates handle long conversations, large-codebase exploration, and session continuity. The core tension is between context preservation (keeping relevant history available) and context dilution (filling the window with noise that degrades attention).

The exam tests three session strategies: resume an existing session, fork a session for divergent exploration, and start fresh with a summary injection. Each strategy has a correct application scenario, and the exam presents all three as distractors for each other.

The Version Control Implications concept is relevant here too: context strategy intersects with how session state is persisted across CI runs.

Is the exam closed-book, and how does that affect preparation?

Yes, the CCA-F is a closed-book, proctored exam with no AI assistance permitted. There is a certain irony in sitting an AI certification without AI help, but the constraint is deliberate: Anthropic is testing internalised understanding, not the ability to prompt a model for answers.

That constraint shapes the right preparation strategy. Passive reading of documentation does not build the pattern recognition the exam requires. Scenario-based practice, where you commit to an answer before seeing the explanation, is the mechanism that transfers to exam performance.

How does AI Skill Certs help with preparation?

AI Skill Certs (https://aiskillcerts.com) is an independent adaptive prep platform for the CCA-F. We are not affiliated with or endorsed by Anthropic. Our adaptive engine uses Bayesian Knowledge Tracing with a 0.90 mastery threshold, which means the platform continues serving questions in a concept area until your demonstrated mastery crosses that bar, not until you have simply seen every question once.

Practice exams are 60 questions, scored 100 to 1000 on the same scale as the real exam, with 720 as the passing bar. Our concept library at /concepts covers 174 atomic concepts mapped to all five domains and 30 task statements.

Archie, our Socratic tutor, never gives the answer directly. It guides with graduated hints, which builds the reasoning process rather than answer recall. That distinction matters on an exam where every question is a scenario and every distractor is plausible.

What is the return on the $99 investment?

As of 3 June 2026, the CCA-F is the only professional Claude certification in existence. Anthropic has announced further architect, developer, and seller certifications planned for later in 2026, which means early holders of the CCA-F will have a credential that predates the broader certification ecosystem.

The $99 cost is low relative to the credential's current scarcity: 10,000+ certified individuals against 40,000+ partner applicant firms implies significant demand for certified practitioners within the partner network. Whether that translates to career value depends on your specific role and organisation, and we will not invent salary figures to argue the point. What we can say is that the exam is priced to encourage attempts, and the preparation investment is the larger variable.

The table above is the complete cost-and-context picture. The $99 is the easy part. The preparation is where candidates differentiate themselves.