Claude Certifications: All Four Tracks, One Decision Guide

The Claude Partner Network launched four claude certifications on 12 March 2026, each a proctored Pearson VUE exam scored on a 100-to-1000 scale. As of 3 June 2026, more than 10,000 individuals hold at least one credential and more than 40,000 firms have applied to join the programme. The two developer-relevant Foundations exams, CCAR-F and CCDV-F, both cost $125 and pass at 720. This guide maps all four tracks, examines the CCDV-F domain weights from Anthropic's 8 July 2026 exam guide, and explains how to allocate study time based on what the blueprint actually says.

What are the four Claude Partner Network certification tracks?

All four exams are scored on a 100-to-1000 scale with a passing mark of 720. Credentials are valid for 12 months from the award date and can be taken online-proctored or at a Pearson VUE test centre.

Item counts for CCAO-F and CCAR-P are not published in available sources. CCAR-F and CCDV-F details come from Anthropic's official exam guides. Tiered Claude Partner Network partners receive discounted first attempts on eligible exams. Anthropic has stated more tracks are planned for later in 2026 with no dates confirmed. The certifications form part of the Claude Partner Network, a $100M programme.

AI Skill Certs is an independent prep platform, not affiliated with or endorsed by Anthropic. We currently support CCAR-F and CCDV-F preparation.

Which certification should a developer take first?

CCDV-F (Developer Foundations, $125) targets implementation and integration: API usage, agent construction, prompt engineering, and tool contracts. CCAR-F (Architect Foundations, $125) targets system design: orchestration patterns, multi-agent architecture, and cross-cutting reliability decisions.

If you write production code against the Claude API, CCDV-F is the direct credential for your work. If your role involves designing systems that use Claude rather than building them, CCAR-F fits better. The two tracks are complementary rather than sequential; there is no formal prerequisite between them, and both carry equal exam cost and passing bar.

How do CCAR-F and CCDV-F differ structurally?

The absence of a scenario bank in CCDV-F has practical consequences for preparation. CCAR-F candidates can identify the six scenarios in the bank and practise against them specifically. CCDV-F candidates cannot: items are written directly against the skills in each domain. Comprehensive domain coverage is the only reliable strategy. Do not plan CCDV-F preparation around a scenario bank; there is none.

What domains does the CCDV-F exam test?

The eight CCDV-F domains and their weights come from Anthropic's official exam guide dated 8 July 2026. Earlier prep materials may reflect a prior blueprint; verify your source's date before using it.

Domain 2 alone accounts for 33.1% of the exam. Domains 2, 5, and 1 together represent 64.6% of items. A candidate who masters those three domains is already in passing territory before addressing the remaining five.

Where should you focus CCDV-F study time?

Work the weights. Below is a proportional study sequence based on the eight domain percentages from Anthropic's July 2026 exam guide:

Domain 2  Applications and Integration     33.1%
Domain 5  Model Selection and Optimization 16.8%
Domain 1  Agents and Workflows             14.7%
Domain 6  Prompt and Context Engineering   11.0%
Domain 8  Tools and MCPs                  10.6%
Domain 7  Security and Safety              8.1%
Domain 3  Claude Code                      3.1%
Domain 4  Eval, Testing, and Debugging     2.6%

If you have ten study hours, roughly six should go to the top three domains and two more to Domains 6 and 8. Claude Code and Eval, Testing, and Debugging are worth no more than 30 minutes each for a candidate with production API experience, given their combined weight of 5.7%.

What does the Applications and Integration domain test?

Domain 2 (33.1%) tests practical judgment about building Claude-powered applications: SDK usage, message construction, streaming versus synchronous responses, error handling, and integration into existing application stacks. Items present realistic engineering decisions rather than syntax recall.

A minimal synchronous API call the exam assumes you can read and reason about:

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Summarise this document in three sentences."}
    ]
)
print(message.content)

Items often ask when to stream versus wait for a complete response, or how to handle a partial failure in an integration pipeline. The exam rewards proportionate solutions: the simplest integration path that meets the stated requirement is the correct answer. For prompt construction patterns that feed directly into integration scenarios, Prompt Engineering & Structured Output covers the core mechanics tested in Domain 2.

What does the Model Selection and Optimization domain cover?

Domain 5 (16.8%) is the second-largest domain and covers choosing the right Claude model for a given task, optimising for cost, latency, and quality, and understanding when to use extended thinking or streaming. Items ask candidates to select the model tier most appropriate for a production scenario, factoring in throughput requirements, context window needs, and cost constraints.

A typical item: a team runs a classification pipeline processing a high volume of documents per hour with a tight cost budget. The question asks which model and configuration are most appropriate. The correct answer balances throughput, cost-per-token, and task complexity, rather than defaulting to the most capable model for every workload. Model selection decisions connect directly to Domain 2 because integration design choices, such as streaming versus batch or synchronous versus asynchronous, depend partly on model latency and token-rate behaviour.

How much does the Agents and Workflows domain matter?

Domain 1 (14.7%) focuses on agent loop control, termination conditions, orchestration patterns, and when to use multi-agent systems versus simpler pipelines. A candidate must be able to read an agent loop, identify where it could run indefinitely, and choose the fix that addresses the structural cause rather than masking it.

messages = [{"role": "user", "content": task}]

while True:
    response = client.messages.create(
        model="claude-opus-5",
        tools=tools,
        messages=messages,
        max_tokens=4096,
    )
    if response.stop_reason == "end_turn":
        break
    messages = append_tool_results(messages, response)

Understanding stop_reason values and loop exit conditions is a recurring theme in Domain 1 items. The Agentic Architecture & Orchestration concept library covers hub-and-spoke topology, prerequisite gate design, and parallel subagent spawning, all of which appear in agentic scenarios on the exam.

What do candidates need to know about Tools and MCPs?

Domain 8 (10.6%) tests tool contract design, MCP server integration, structured error responses, and the build-versus-adopt decision for MCP servers. Items frequently present a poorly described tool that causes misrouting, then ask which single change is highest leverage.

A well-formed tool definition the exam expects candidates to recognise:

{
  "name": "get_customer_record",
  "description": "Retrieves a customer record by confirmed ID. Returns null if the ID does not exist. Do not call this tool unless a validated customer ID is available.",
  "input_schema": {
    "type": "object",
    "properties": {
      "customer_id": {
        "type": "string",
        "description": "The unique customer identifier."
      }
    },
    "required": ["customer_id"]
  }
}

Tool description quality directly determines routing accuracy. The Tool Design & MCP Integration concept library covers the isError flag pattern, the four error categories, and structured error metadata, each of which maps to Domain 8 task statements.

How security-focused is the CCDV-F exam?

Domain 7 (Security and Safety, 8.1%) tests prompt injection defences, blast-radius reduction, safe handling of untrusted content, and when to apply programmatic versus prompt-level controls. Items are almost always structural: a pipeline ingests user-supplied text, and the correct fix involves architectural isolation rather than a sharper prompt instruction.

The exam consistently favours solutions that minimise the blast radius of a single failure. A constrained, scoped tool that limits what an agent can do with untrusted input is the correct answer over a broad tool guarded only by a prompt instruction. Context Management & Reliability covers the isolation and degradation patterns that span both Domain 6 and Domain 7, since context flow and injection risk share structural roots in how information moves through a pipeline.

How much Claude Code knowledge does CCDV-F require?

Domain 3 (Claude Code) is 3.1% of the exam. At 53 items, that equates to roughly one or two items per sitting. You need to understand what Claude Code does in a development workflow and how its configuration layers operate, but deep expertise is not required to pass. Spend no more than 30 minutes reviewing CLAUDE.md configuration patterns and Claude Code's role in CI/CD pipelines. That allocation is proportionate to the domain's weight; over-investing here comes at the cost of the domains that actually decide the outcome.

What is the best strategy for claude certifications in 2026?

Work the domain weights, use prep materials aligned to Anthropic's 8 July 2026 exam guide, and validate your readiness on timed practice exams before booking a sitting. We offer adaptive study, Archie tutoring (our Socratic tutor that guides with graduated hints rather than giving answers directly), and full-length practice exams scored on the 100-to-1000 scale for both CCDV-F and CCAR-F. The platform uses Bayesian Knowledge Tracing with a 0.90 mastery threshold. The concept library at /concepts maps 174 atomic concepts to CCAR-F's five domains, with CCDV-F adaptive study available alongside it.

At $125 per attempt, the cost of an unprepared retake exceeds the cost of a structured prep investment. Candidates who reach 720 consistently on adaptive practice with the 0.90 mastery threshold pass on their first sitting at a markedly higher rate than those who book without benchmarking.