Four tiers in one programme
The four tiers, their prices and the exam format are all documented on Anthropic's official certification page. Associate Foundations costs $99. Developer Foundations and Architect Foundations cost $125 each. Architect Professional, the senior tier, costs $175. Partner discounts are tier- and date-limited: Global Premier partners pay nothing through 31 December 2026, other qualifying partner firms get 50% off. Select and Pref tier firms receive that 50% automatically at checkout, Global Premier reverts to the standard 50% from 1 January 2027, and none of it is free for someone paying their own way.
The format is not identical across the track. Question counts and domain weightings differ between the Foundations and Professional tiers; the Professional exam is documented at 63 questions within a 120-minute limit. Scenario-based items are drawn from four of six scenario banks, and every tier is scored against a scaled pass mark of 720 out of 1,000. Because the score is scaled, 720 is not a raw percentage of correct answers.
All of it sits behind a partner gate, described in full further down. The practical consequence is that most engineers will meet this syllabus as documentation rather than as an exam — which is why the domain weights matter more than the badge.
What the Architect Foundations exam covers
Five domains make up the Foundations exam: Agentic Architecture & Orchestration, Tool Design & MCP Integration, Claude Code Configuration & Workflows, Prompt Engineering & Structured Output, and Context Management & Reliability. Orchestration carries the largest weight at 27% of the total score.
The Professional tier adds two areas and covers seven domains in total, among them Governance, Stakeholder Communication and Developer Enablement. The senior exam therefore grades how a team documents decisions and hands work to other developers, alongside the API-level work. Preparation for both tiers runs through the same five domains, with the Professional exam assuming you can already build and operate them.
All of the scenario material sits on the same stack: Claude Code, the Claude Agent SDK, Model Context Protocol servers and the Claude API. A practical guide by Sathish Kraju works through that stack with runnable code and follows the exam's scenario style.
Why orchestration outweighs prompting
Anthropic's format treats the LLM as a stateless reasoning engine. Anything that has to survive between two model calls is written by the engineer: stored conversation state, the routing rule that selects a tool, the retry that fires when a tool returns malformed JSON. Those are the decisions the exam scores.
That matches what we see in the pipelines behind this magazine — article generation, transcription and TTS. Swapping a model is a config change. The retry logic, the context budget and the tool schemas are the parts that page someone at three in the morning.
Token prices reinforce the same point. Picking a different model for each step is a routing decision, and the orchestration layer is what makes it. Claude Haiku 4.5 sits at $1 per million input tokens and $5 per million output tokens on Anthropic's published pricing page, which is cheap enough to take the classification and extraction steps that dominate a long pipeline. A harness that cannot route that work away from an expensive reasoning model is burning budget on every request. We measure the same trade-offs on local and hosted models in AI Arena, where tokens per second, time to first token and VRAM use are logged per run.
Prices in euros
The tier prices are quoted in US dollars. At roughly €0.92 to the dollar, that works out to about €91, €115 and €161 before tax. VAT treatment depends on how the exam is sold in your country, so treat those figures as floors rather than final numbers.
Access is the harder problem. Registration runs through the Claude Partner Network and requires a verified company domain email, so membership is something an employer holds rather than an individual. An independent developer in Prague, Lisbon or Warsaw cannot simply book a slot, and the exams are sold through partner organisations. For a freelancer or a two-person agency, the barrier is therefore not the fee but the procurement path: someone has to have a partner relationship and a reason to sponsor the seat. Companies that qualify get the discount automatically at checkout; everyone else is looking at a purchase they cannot make.
The EU AI Act angle
Article 4 of the EU AI Act requires providers and deployers of AI systems to ensure a sufficient level of AI literacy among the staff who work with those systems. No regulator asks for a Claude badge. But when an employer has to show that its engineers understand context limits, tool permissions and failure modes, a vendor exam is one of the few documents that exists on the market.
Badges are valid for 12 months, which is shorter than a typical enterprise governance cycle. Vendor certification also carries no regulatory status. Since 2 August 2025 the European AI Office has held direct enforcement powers over general-purpose AI providers, including fines under Article 101 of the Act of up to €15 million or 3% of global turnover, and none of that exposure is addressed by passing an Anthropic exam.
Five domains, one conclusion
Read the weights and a picture emerges that contradicts the usual prompt-engineering advice. Prompting is one of five domains. Orchestration, tool design, context management and failure handling make up the rest, and the scenario format pushes candidates toward decisions they would have to defend in a code review.
For anyone preparing without access to the exam, the syllabus is still public. Building one MCP server, one tool-routing layer with retries and one context-trimming routine covers most of what the Foundations domains describe, and it is the same code the Professional tier assumes you already run in production.
The limits are plain: the partner gate set out above decides who can sit the exams at all, and awarded badges expire after 12 months.
Can a freelancer take the Claude Certified Architect exam?
Not under the current terms — the exams are sold through the Claude Partner Network, so a developer without a partner employer cannot book a slot. The published syllabus is the practical alternative.
What do the exams cost once a partner discount applies?
List prices are $99, $125 and $175, roughly €91, €115 and €161 at a rate of about €0.92 per dollar, before any VAT. Global Premier partners pay nothing through 31 December 2026 and fall back to the standard 50% discount from 1 January 2027; Select and Pref tier partner firms get an automatic 50% discount at checkout. Anyone paying their own way pays the full list price.
