Timo Labs › Claude Certification › Exam changes
Claude certification exam changes from 7 December 2026
Three Claude certification exams change on 7 December 2026: Claude Certified Associate Foundations (CCAO-F), Claude Certified Developer Foundations (CCDV-F) and Claude Certified Architect Professional (CCAR-P). Anthropic published the new exam guides for all three on 17 November 2026.
From 7 December, each of the three exams has 70 questions, the weight of each domain changes, and every sample question in the guide is new. A domain is a topic area with a set share of the scored questions, and that share is its weight. The time limit (120 minutes), the pass mark (720 on a scale of 100 to 1,000) and the exam fees stay the same. Claude Certified Architect Foundations (CCAR-F, also written CCA-F) does not change.
This page compares the old and new guide for each exam. Every figure was checked against Anthropic's guides on 18 November 2026.
Last checked: 18 November 2026
Which exam guide applies to your exam date
| Your exam date | Guide to study from |
|---|---|
| On or before 6 December 2026 | The guide labelled "For exams through Dec 6" |
| On or after 7 December 2026 | The guide labelled "For exams from Dec 7" |
Both guides are numbered version 1.1, so the version number does not tell them apart. Use the date label on the guide. Each guide is on its exam's row in Anthropic's Partner Certifications list.
If you move your exam from before 7 December to on or after it, or the other way, study from the guide for the new date.
What changes in all three exams
| Until 6 December | From 7 December | |
|---|---|---|
| Questions, CCAO-F | 60 | 70 (60 scored, 10 unscored) |
| Questions, CCDV-F | 53 | 70 (60 scored, 10 unscored) |
| Questions, CCAR-P | 63 | 70 (60 scored, 10 unscored) |
| Question type | Multiple choice and multiple response | Scenario-based multiple choice: each question describes a work situation, and a question can still ask for more than one answer |
| Sample questions in the guide | 3 per exam | 3 per exam, all new |
| Topics | Lists of knowledge and tasks | Skills tested through work situations |
Unscored questions. An unscored question is one Anthropic is trying out for future exams. It does not count towards your score. The exam does not show which 10 questions are unscored, so answer all 70.
Less time per question. The time limit stays at 120 minutes while the question count rises, so the average time per question falls:
| Exam | Until 6 December | From 7 December |
|---|---|---|
| CCAO-F | 2 minutes | 1 minute 43 seconds |
| CCDV-F | 2 minutes 16 seconds | 1 minute 43 seconds |
| CCAR-P | 1 minute 54 seconds | 1 minute 43 seconds |
Questions with more than one correct answer can still appear. The new guides still say that each question states how many answers to select.
What does not change
- Time limit: 120 minutes.
- Pass mark: 720 on a scale of 100 to 1,000. This is a scaled score, not 72% of questions correct.
- Exam fees: US$99 for CCAO-F, US$125 for CCDV-F, US$175 for CCAR-P, before any partner discount.
- Where you take it: with Pearson, online with a remote proctor or at a test centre.
- Results: pass or fail, a scaled score, and the percentage correct in each domain.
- Validity: 12 months.
- Retakes: a wait of 14 days after the first failed attempt, 30 days after the second and 90 days after the third, with up to 4 attempts in 12 months for each exam.
- Cancelling or rescheduling: up to 1 hour before an online appointment, or up to 48 hours before a test-centre appointment.
- Prerequisites: none for any of the three exams.
Claude Certified Associate Foundations (CCAO-F): what changes
The exam goes from seven domains to six. Troubleshooting and Optimization was a separate domain worth 10%. From 7 December its skills are tested inside the prompting, output evaluation, workflow and configuration domains.
| Until 6 December | Weight | From 7 December | Weight |
|---|---|---|---|
| Prompting and Task Execution | 14% | Task Definition and Prompting | 15% |
| Output Evaluation and Validation | 21% | Output Evaluation and Quality Control | 20% |
| Product and Model Selection | 12% | Capability, Model, and Context Selection | 15% |
| Workflow Integration and Solution Design | 16% | Workflow Design and Integration | 18% |
| Configuration and Knowledge Management | 12% | Configuration and Knowledge Management | 15% |
| Governance, Risk, and Responsible Use | 15% | Responsible Use, Data Protection, and Governance | 17% |
| Troubleshooting and Optimization | 10% | No separate domain |
Named in the new guide and not in the old one:
- Skills, which are reusable instructions Claude loads for a task.
- Connectors, which link Claude to Drive, Gmail and calendar.
- Analysis that runs code, for exact calculations on a file.
- Memory: reviewing, editing and deleting what Claude has saved about you, and using incognito chats, which are not saved to memory.
- Model effort levels as well as model tiers. A tier is the size of the model, and the old guide named the Haiku, Sonnet and Opus families. An effort level is a setting for how much reasoning Claude does before it answers.
- Removing personal details from data before uploading it.
- The AI Fluency Framework, Anthropic's four skills for working with AI: Delegation, Description, Discernment and Diligence.
What to practise. Practise these because the new guide names them: set up a Project with instructions, knowledge files and a Skill; connect one workplace data source; check every figure, citation and quote in an output against its source; review and delete saved memories.
Output evaluation stays the largest domain at 20%. In Timo practice tests up to 7 November 2026, 18.7% of answers in this domain were wrong, the highest of any CCAO-F domain, so give this domain extra study time.
Related: Claude Certified Associate Foundations exam guide and the free CCAO-F practice test.
Claude Certified Developer Foundations (CCDV-F): what changes
The eight domains stay. Their weights change:
| Domain | Until 6 December | From 7 December |
|---|---|---|
| Agents and Workflows | 14.7% | 17% |
| Applications and Integration | 33.1% | 20% |
| Claude Code | 3.1% | 8% |
| Eval, Testing, and Debugging | 2.6% | 10% |
| Model Selection and Optimization | 16.8% | 12% |
| Prompt and Context Engineering | 11.0% | 15% |
| Security and Safety | 8.1% | 8% |
| Tools and MCP | 10.6% | 10% |
MCP is the Model Context Protocol, the standard way to connect Claude to outside tools and data.
Three domains grow the most. Eval, Testing, and Debugging goes from 2.6% to 10%, about 6 of the 60 scored questions. Claude Code goes from 3.1% to 8%. Prompt and Context Engineering goes from 11% to 15%. Applications and Integration falls from 33.1% to 20%.
The old guide gave a weight for each of its 25 skills. The new guide gives weights for the eight domains only.
Named in the new guide and not in the old one:
- Evaluation design. This covers a golden set, which is a fixed set of test cases with known correct answers. It covers regression tests, which are tests you run again after each change to check that nothing that worked has broken. It also covers using a model to grade outputs after checking the model's grades against human grades.
- Integration failures: how a streamed response ends, rate-limit and overload errors, and retry and fallback logic.
- Extending Claude Code: choosing between Skills, plugins, rules, hooks, subagents and commands.
- Grounding: answering from supplied documents with citations, and saying when the answer is not in them.
- Cost: finding which part of a request drives cost, and using prompt caching. Prompt caching lets Claude reuse the part of a prompt that repeats across requests, which lowers cost. A cache breakpoint marks where the reusable part ends.
No longer named: the agent frameworks Strands, LangGraph and PydanticAI, and general software topics such as REST, JSON and websockets as separate skills.
What to practise. Practise these because the new guide names them: build and ship one agent that uses tools; write a small evaluation with a golden set; read a trace, which is the step-by-step record of what an agent did, to find why a run failed; set cache breakpoints and check how often requests reuse the cache.
Security and Safety stays at 8%. In Timo practice tests up to 7 November 2026, 27.2% of answers in this domain were wrong, the highest of any CCDV-F domain, so give this domain extra study time.
Related: Claude Certified Developer Foundations exam guide and the free CCDV-F practice test.
Claude Certified Architect Professional (CCAR-P): what changes
The seven domains stay. Stakeholder Communication and Lifecycle Management halves, and its weight moves to five other domains:
| Domain | Until 6 December | From 7 December |
|---|---|---|
| Solution Design and Architecture | 17% | 18% |
| Claude Models, Prompting, and Context Engineering | 13% | 13% |
| Integration | 19% | 21% |
| Evaluation, Testing, and Optimization | 16% | 17% |
| Governance, Safety, and Risk Management | 14% | 16% |
| Stakeholder Communication and Lifecycle Management | 14% | 7% |
| Developer Productivity and Operational Enablement | 7% | 8% |
Named in the new guide and not in the old one:
- Choosing where to run Claude: the Anthropic API, Amazon Bedrock, Google Vertex AI or Microsoft Foundry. It also covers choosing the agent runtime, which is the environment that runs the agent.
- Production reliability: rate limits, retries, fallback routing, and making sure a repeated request does not repeat a payment or other action.
- Moving to a new model version: reading regression results and deciding whether to stay on the current version, adjust prompts or delay.
- Bias in evaluations graded by a model, such as favouring longer answers.
- Prompt injection, which is an instruction hidden in content Claude reads. The guide covers where it can enter: retrieved documents, tools, MCP responses and third-party Skills.
- Keeping human review and ownership of AI-generated code.
The new guide tests compliance through controls instead of naming regulations. The old guide named GDPR, HIPAA and FedRAMP. The new guide tests the controls: where data is stored, how long it is kept, removing personal data, audit logs and keeping each customer's data separate.
Architect Foundations is now recommended preparation. It is still not a prerequisite. The new guide says the scope of CCAR-F is assumed knowledge for CCAR-P.
What to practise. Practise these because the new guide's preparation section lists them: review architecture diagrams, configuration files, logs, traces and evaluation results, and write down the trade-off behind each design decision.
Related: Claude Certified Architect Professional exam guide and the free CCAR-P practice test.
Claude Certified Architect Foundations (CCAR-F) does not change
CCAR-F, also written CCA-F, has no syllabus change on 7 December. Its guide was updated in November for registration, identification and support wording only. It keeps its five domains and 60 questions.
What the new sample questions cover
Each guide has three sample questions. All nine are new. The old ones asked you to pick a feature or pattern. The new ones describe a fault or a constraint and ask what to do about it.
| Exam | Old sample topics | New sample topics |
|---|---|---|
| CCAO-F | Checking a cited regulation; choosing a faster model; hiding account numbers before upload | Checking figures, citations and a quote in a brief for a client; a Project using old pricing and another client's terms; removing personal details from an HR dataset before upload |
| CCDV-F | Batch processing for cost; prompt injection in a web page; an MCP server for a shared API | A streamed answer that ends early; prompt caching that never reuses the cache; a tool that receives a customer name instead of an order ID |
| CCAR-P | Removing tools an agent does not need; prompt caching; a retrieval fault after a data refresh | Choosing where to deploy under a data-residency rule; retries that pay a reimbursement twice; approval steps that slow every tool call |
Questions about the December exam changes
How many questions are on the Claude certification exams from 7 December 2026?
CCAO-F, CCDV-F and CCAR-P each have 70 questions: 60 scored and 10 unscored. CCAR-F stays at 60.
Is the pass mark changing?
No. It stays at 720 on a scale of 100 to 1,000.
Do the exam fees change?
No. CCAO-F is US$99, CCDV-F is US$125 and CCAR-P is US$175, before any partner discount.
Does the Claude Certified Architect Foundations exam change?
No. CCAR-F is not part of the 7 December change.
I am already certified. Do I have to take the new exam?
A certification is valid for 12 months and is renewed with a free assessment in Anthropic Partner Academy. The guides add that if exam content changes significantly, Anthropic may require holders to retake the full exam to recertify. The guides do not say whether that applies to this change.
Which guide do I use if my exam is on 7 December?
The guide labelled "For exams from Dec 7".
Sources
- Anthropic, Claude Certified Associate Foundations, Developer Foundations and Architect Professional exam guides, version 1.1, "For exams through Dec 6" and "For exams from Dec 7". Checked 18 November 2026.
- Anthropic Partner Academy, Partner Certifications list: https://anthropic-partners.skilljar.com/page/partner-certifications
- Pearson, Claude Certification Program page: https://www.pearsonvue.com/us/en/anthropic.html
By Amotion AI