What changes in the CCAR-P exam on December 7, 2026

Published November 18, 2026. Based on Anthropic's two public CCAR-P exam guides: the guide for exams taken through December 6, 2026, and the guide effective December 7, 2026. Both are linked from the Claude certifications page. CERTain·exam is not affiliated with Anthropic.

Short answer. From December 7, 2026, the Claude Certified Architect – Professional (CCAR-P) exam has 70 items instead of 63. 60 items are scored and 10 are unscored and not marked. Items are described as scenario-based multiple choice, the domain weights change, and every domain's objectives are rewritten. The 120-minute time limit, the passing score of 720 on a 100–1,000 scale, the $175 USD fee and the 12-month validity stay the same.

Which exam guide applies to my exam date?

The date you take the exam decides which guide applies. The updated guide is effective December 7, 2026. Both guides are linked from Anthropic's Claude certifications page, one for exams through December 6, 2026 and one for exams from December 7, 2026.

If you test on December 6, 2026 or earlier, prepare from the guide for exams through December 6, 2026. If you test on December 7, 2026 or later, prepare from the guide effective December 7, 2026.

How many items are on the CCAR-P exam from December 7, 2026?

70 items. The guide effective December 7, 2026 says 60 of them are scored and 10 are unscored. Unscored items are not identified and do not affect your score. The guide explains that they are being evaluated for future exam forms and asks you to allow time for every item.

Through December 6, 2026, the exam has 63 items. The time limit is 120 minutes in both versions. That is about 1 minute 54 seconds per item through December 6, and about 1 minute 43 seconds per item from December 7.

What changes in the item format?

The guide for exams through December 6, 2026 describes "multiple-choice and multiple-response items". The guide effective December 7, 2026 describes "scenario-based multiple-choice items". Both guides add that "each item states how many responses to select".

The updated guide also says exam scenarios present artifacts, such as architecture diagrams, configuration files, logs and traces, evaluation results and code, and expect you to reason from them.

How do the CCAR-P domain weights change?

The weight is the approximate share of scored items that comes from each domain.

DomainThrough December 6, 2026From December 7, 2026Change
1. Solution Design and Architecture17%18%+1 point
2. Claude Models, Prompting, and Context Engineering13%13%no change
3. Integration19%21%+2 points
4. Evaluation, Testing, and Optimization16%17%+1 point
5. Governance, Safety, and Risk Management14%16%+2 points
6. Stakeholder Communication and Lifecycle Management14%7%−7 points
7. Developer Productivity and Operational Enablement7%8%+1 point

The biggest change is Domain 6, Stakeholder Communication and Lifecycle Management. Its weight halves, from 14% to 7%, and it becomes the smallest domain.

Integration stays the largest domain and grows from 19% to 21%. Governance, Safety, and Risk Management grows from 14% to 16%.

Integration, Solution Design and Evaluation together make up 52% of scored items through December 6, 2026, and 56% from December 7, 2026.

Did the domain names change?

Only the punctuation. The guide effective December 7, 2026 writes "and" where the earlier guide used "&", for example "Solution Design and Architecture" instead of "Solution Design & Architecture". There are still seven domains, in the same order.

What is new in the objectives of each domain?

The guide for exams through December 6, 2026 lists short tasks for each domain. The guide effective December 7, 2026 lists named skills, each with one sentence on what you analyze, evaluate or select, usually under stated constraints. Below are the new or clearly expanded areas, summarized in our own words.

  • Domain 1 (18%). New: deciding which parts of a business process go to Claude, stay with people or are shared, judged by reversibility, stakes and accountability. New: choosing a deployment surface (Anthropic API, Amazon Bedrock, Google Vertex AI or Microsoft Foundry) and an agent runtime (Messages API loop, self-hosted Agent SDK or headless Claude Code, or Claude Managed Agents). Skills-based patterns join workflow, agentic and augmented-LLM patterns.
  • Domain 2 (13%). New: reasoning effort and context lifecycle for long-running agents (context editing, memory tool, compaction). New: prompted reasoning versus native thinking, and structured output, as prompting choices. Expanded: diagnosing an underspecified system prompt and placing cache breakpoints so stable content comes before dynamic content.
  • Domain 3 (21%). New: production reliability controls (rate-limit handling, retry and backoff, idempotency for tools with side effects, fallback routing). Retrieval now compares a pre-built RAG pipeline, agentic search with tools and full-corpus context. Tool access is judged against least privilege and minimal footprint.
  • Domain 4 (17%). New: LLM-as-judge design and its biases (position, verbosity, self-preference, judge drift). New: migrating to a new model version within a deprecation timeline. New: experiment design, including how many trials per case separate a real difference between configurations from run-to-run variance.
  • Domain 5 (16%). New: prompt injection through retrieved documents, tool results, MCP server responses and third-party Skills. New: ranking deployment risks, including token and quota exhaustion and supply-chain risk. Compliance is framed as choosing architectural controls (data residency, zero data retention, PII redaction, audit logging, tenant isolation); the earlier guide named regulations such as GDPR, HIPAA and FedRAMP.
  • Domain 6 (7%). Narrower: three skills instead of five tasks. Spotting commitments in requirements or a draft SLA that the architecture cannot meet, reviewing architecture decision records, and checking that an operational handoff package is complete.
  • Domain 7 (8%). New: packaging and distributing a Claude Code workflow across an organization (plugins with skills, hooks, subagents, commands and MCP configuration, a marketplace and managed settings). Expanded: configuring Claude Code and the Agent SDK, including permission modes, hooks, subagents and headless or CI use. New: keeping human ownership of AI-generated code.

What else changes in the updated guide?

  • The intended candidate works daily with the Claude Developer Platform, Agent Skills, Claude Code and the Claude Agent SDK, the Model Context Protocol and Claude Managed Agents, and knows how the Anthropic API differs from Amazon Bedrock, Google Vertex AI and Microsoft Foundry.
  • The typical candidate has at least six months of hands-on experience with Claude in production. The earlier guide said Claude or comparable LLM-based systems.
  • Claude Certified Architect – Foundations is now "recommended, but not required, preparation", and the updated guide treats the Foundations scope as assumed knowledge. It is still not a prerequisite.
  • The preparation advice adds practice in reviewing artifacts produced by others: architecture diagrams, configuration files, logs and traces, evaluation results and code.
  • The guide has new sample items. We don't reproduce them here; read them in the guide.

What stays the same?

Through December 6, 2026From December 7, 2026
Time limit120 minutes120 minutes
Passing score720 on a 100–1,000 scale720 on a 100–1,000 scale
Exam fee$175 USD$175 USD
Validity12 months12 months
ResultPass/fail, scaled score, percent correct by domainPass/fail, scaled score, percent correct by domain
DeliveryProctored, online or at a test centerProctored, online or at a test center
Domains77

Registration, rescheduling, retake and renewal rules read the same in both guides. For details, see CCAR-P exam: cost, registration, and format.

What does this mean for your preparation?

  • Check your exam date first, then study from the guide for that date.
  • If you test from December 7, 2026, self-assess against the updated skills. Give extra attention to the areas that are new: deployment surfaces and agent runtimes, production reliability controls, LLM-as-judge evaluation, prompt injection and distributing Claude Code workflows.
  • Plan your pace for 70 items in 120 minutes. Unscored items are not marked, so give every item the same care.
  • Practice reading artifacts: logs, traces, configuration files and architecture diagrams.
  • Don't skip Domain 6. Its weight falls to 7%, but it is still scored.

For a study plan by domain, see How to prepare for CCAR-P. For practice exams and sample questions, see CCAR-P practice exams.

Questions

When does the CCAR-P exam change?

On December 7, 2026. Exams taken through December 6, 2026 follow the earlier guide; exams taken on December 7, 2026 or later follow the updated guide.

How many items are on the CCAR-P exam after December 7, 2026?

70 items, of which 60 are scored and 10 are unscored. Unscored items are not identified and do not affect your score. Through December 6, 2026, the exam has 63 items.

Do the time limit, passing score or fee change?

No. Both guides list a 120-minute time limit, a passing score of 720 on a 100–1,000 scale, a $175 USD fee and 12 months of validity.

Which CCAR-P domain changes the most?

Stakeholder Communication and Lifecycle Management falls from 14% to 7%. Integration grows from 19% to 21% and stays the largest domain.

Do I need Claude Certified Architect – Foundations first?

No. The updated guide calls it recommended, but not required, preparation, and treats its scope as assumed knowledge.

Where can I download both CCAR-P exam guides?

On Anthropic's Claude certifications page in Anthropic Partner Academy. It links the guide for exams through December 6, 2026 and the guide for exams on December 7, 2026 or later.

Sources

Get your free practice exam

First, sign in with your email.