Anthropic will soon have two professional-level Claude certifications at the same $175 fee. Claude Certified Architect - Professional (CCAR-P) is open for registration today. Claude Certified Developer - Professional is listed in Anthropic's catalog as a preview, with its format and domains published but no exam guide or launch date yet. Both cover integration, evaluation and safety, so they can look interchangeable on paper. They certify different jobs: the architect decides how a Claude system should be built and governed and brings stakeholders along, and the professional developer keeps that system healthy, measured and secure once it runs.
Status on November 16, 2026
CCAR-P is open for registration, and Anthropic updates its content on December 7, 2026 (updated exam guide on November 17). Developer - Professional is a preview entry in the Partner Academy catalog and cannot be booked yet. See the Developer - Professional exam guide for everything Anthropic has listed.
The two professional Claude certifications (as listed by Anthropic, November 2026)
| Architect - Professional (CCAR-P) | Developer - Professional | |
|---|---|---|
| What Anthropic says it validates | Advanced architecture at scale: enterprise solution design, integrations, evaluation and optimization, governance and risk, stakeholder leadership across the lifecycle, developer productivity and operations | Running Claude-based systems in production at scale: observability, incident response, release evaluation, resilient integrations, cost and latency across a fleet, layered guardrails, durable multi-agent systems, coding-agent fleets |
| Status | Open for registration | Preview (no registration yet) |
| Questions and time | 63 in 120 minutes, all scored | 70 in 135 minutes, 60 scored |
| Fee | $175 | $175 |
| Languages | English and Japanese | English (as listed) |
| Exam guide | Published (updated guide due November 17, 2026) | Not published |
Where the domains overlap
Domain by domain
| Architect - Professional (weight) | Developer - Professional (weight) | How they differ |
|---|---|---|
| Integration (19%) | Production Integration, Reliability, and Knowledge Grounding (20%) | The architect chooses the integration pattern and platform; the developer keeps it up under rate limits, overloads and long requests, and grounds answers in data |
| Evaluation, Testing & Optimization (16%) | Evaluation Engineering and Quality Assurance (12%) | The architect defines what success means and how it is tested; the developer builds the eval suite that gates every release |
| Governance, Safety & Risk Management (14%) | Runtime Guardrails and Production Security (11%) | The architect sets policy and risk controls; the developer implements layered defenses in the running agent |
| Developer Productivity & Operational Enablement (7%) | Agent Fleet Operations for Software Delivery (10%) | The architect enables teams; the developer operates the coding agents those teams run |
| Solution Design & Architecture (17%) | Advanced Agent Architecture and Orchestration (17%) | Solution design covers the whole system; the developer domain is about agents that survive long runs and failures |
| Claude Models, Prompting & Context Engineering (13%) | Cost and Latency Optimization at Scale (8%) | Model and context choices meet the bill and the latency budget at volume |
| Stakeholder Communication & Lifecycle Management (14%) | (no counterpart) | Leading stakeholders is architect work |
| (no counterpart) | Observability and Production Monitoring (12%) | Instrumenting the system is developer work |
| (no counterpart) | Advanced Debugging and Incident Response (10%) | Finding the cause of failures and leading the response |
Two domains on each side have no counterpart, and they define the split. CCAR-P spends 14% on stakeholder communication and lifecycle management, the conversations and decisions that happen before and around a build. Developer - Professional spends 22% on observability plus debugging and incident response, the work that starts after launch.
Try 15 CCDV-P practice questions free
They come from the CCDV-P practice tests, and each option shows its explanation once you answer.
- 6 timed practice exams
- Every answer explained
- A score for each domain
- $19.99 one-time
One system, two jobs
Take a bank that wants Claude to draft replies to customer complaints, with a rule from compliance: nothing that admits fault or offers compensation goes out without a human approval.
The architect's questions come first. Where does Claude run, given the bank's data rules? What counts as a good reply, and how will the team prove the system meets that bar before launch and after every model change? Where does the approval rule live so that no prompt edit can bypass it? Who signs off, and what happens when the model the system is pinned to is deprecated?
The professional developer's questions start once it runs. Are replies being drafted within the latency target during the Monday peak? Which requests were refused, and why? Did last Tuesday's prompt change move the eval scores? Why did cost per reply rise 30% after a deploy that touched only the tool list? When a draft offering compensation slipped through, which layer failed?
The two questions below are written at each level on this same system. They are our own examples: Anthropic publishes sample questions for CCAR-P in its exam guide and none yet for Developer - Professional.
The same system at each professional level
Before launch, the bank's head of compliance asks how the team will know the complaint-reply system still meets its accuracy and approval requirements after Anthropic releases newer models and the current one is eventually deprecated. Which plan answers the question?
The question is about lifecycle and stakeholders: agreed criteria, a repeatable gate, evidence for audit and a migration plan. A breaks when the model is deprecated, C ships unverified changes, and D relies on the model grading itself.
After the team moved the complaint-reply system to a newer model, the share of responses with stop_reason: "refusal" on legitimate complaints rose from 0.2% to 3%, and those customers received no reply. What is the best first response?
A refusal arrives as a successful response with stop_reason: "refusal", and stop_details names the policy category. Anthropic's guidance is to retry on a fallback model or rephrase. The incident response protects customers first, then finds the cause with evidence. A loops on a non-error, C changes behavior without evidence, and D confuses refusal with max_tokens.
Full CCDV-P set: 6 timed exams, every answer explained, $19.99 one-time.
See the practice testsWhich one fits you
Take Architect - Professional if you design Claude solutions for an enterprise or for clients, choose platforms and patterns, write the governance and risk plan, and present to stakeholders. It is open now, and the CCAR-P complete guide covers it.
Plan for Developer - Professional if you are the engineer on call for a Claude system, run Claude Code or the Agent SDK for many developers, or own the cost and reliability numbers. Start with the domain-by-domain breakdown while the exam guide is pending.
Consider both if you are a technical lead at a Claude Partner Network firm who designs systems and then runs them for clients. The overlap in integration, evaluation and safety means preparing for one covers a fair share of the other, and the two domains each side has alone are the ones to add.
On partner credit: Anthropic counts Developer - Foundations, Architect - Foundations and Architect - Professional toward an organization's standing in the Claude Partner Network. It has not said whether Developer - Professional will count.
Frequently asked questions
Sources: