Claude Academy: 289 Resources, 22 Courses, and a Certificate That Certifies Attendance
Anthropic's learning platform launched today. Every course, with real lesson counts and durations; the difference between the free badge and the $125 proctored exam you probably cannot sit; and the four hours actually worth taking.
Anthropic launched Claude Academy on 20 August 2026 at academy.claude.com. It is a rebuild and rebrand of Anthropic Academy, which had run since March 2026 at anthropic.skilljar.com. The learning management system underneath is still Skilljar. anthropic.com/learn now returns a 308 redirect to the new domain.
The catalogue reports 289 resources split into three kinds: courses, tutorials and use cases. Everything is free. You need an account, and no payment method is requested anywhere in the flow.
This is what is actually in it, what the certificate is worth, and what is worth your time.
The three content types
Courses are multi-lesson sequences with a quiz at the end, running from 45 minutes to 9 hours. Tutorials are short single-topic walkthroughs. Use cases are roughly ten-minute one-page briefs pinned to a job role — Account research and Account tracking are both filed under Sales.
The platform tracks completion, recommends courses based on stated interests, and issues badges. Anthropic states five design principles behind it: increase human agency, cultivate mindsets over features, extend education beyond the AI interaction itself, encourage practice, and use AI as a learning partner.
The 22 courses
Lesson counts, quiz counts and durations are as published on the catalogue page.
Three courses carry the platform: the API course and the two cloud courses are 8 to 9 hours each and account for 198 of the roughly 380 lessons on offer. Everything Claude Code related is short — Claude Code 101 and Claude Code in Action are an hour apiece, and the subagents course is 45 minutes.
Note that Claude with Amazon Bedrock is AWS-created content sitting in Anthropic's catalogue, not Anthropic-authored material.
AI Fluency and the 4D framework
Nine of the 22 courses are AI Fluency variants. They share one framework, applied to different audiences: educators, students, nonprofits, small businesses, builders, pK–12, and a train-the-trainer track.
The framework is the 4Ds:
It was authored by Prof. Rick Dakan (Ringling College of Art and Design) and Prof. Joseph Feller (University College Cork) in collaboration with Anthropic, and released under a Creative Commons BY-NC-SA 4.0 licence. Anthropic teaches it to its own employees during onboarding, which is the strongest thing anyone can say for it.
The most-cited praise for the course is not about prompting. It is that the Delegation module spends real time on when not to use AI.
The certificate is not the certification
This is the part most write-ups get wrong, and it matters if you are planning to put something on LinkedIn.
What the Academy gives you is a completion certificate and a badge. There is no exam, no proctor, and no identity check. It records that you watched the lessons and passed the in-course quiz. As one commenter put it: these are “completion certificates for official vendor training, not proctored exams”, and overselling them as elite credentials “will hurt you”.
Anthropic does run a real certification programme. It is paid, proctored, and it is not on the public Academy at all.
Delivery moved to Pearson on 30 June 2026, either online-proctored through OnVUE or at a test centre. Exams run 120 minutes, multiple-choice and scenario-based, and are closed book: no notes, no documentation, no translation tools, and no AI assistants. Scoring is scaled 100–1000 with a pass mark of 720 on all four. Certifications are valid 12 months; renewal is a free, non-proctored assessment if you do it on time, and a full paid retake if you let it lapse.
Architect — Foundations is 60 questions across five weighted domains:
The catch: it is partner-only. You have to work at a Claude Partner Network organisation and register with a recognised company email domain. Personal addresses are rejected. Partner tiers get 50% off, and Global Premier partners pay nothing at all through 31 August 2026.
So the free thing anyone can take gives a certificate that certifies attendance, and the thing that would actually signal competence is gated behind an employer relationship. A third-party prep industry has already grown around the exam anyway — Udemy practice-exam courses, freeCodeCamp prep articles, and sites selling mock exams to people who in many cases cannot sit it.
What people say is wrong with it
The criticism is consistent and specific.
Depth. Claude Code in Action gets called “a basic course”. One developer review rated the catalogue from 9/10 down to 3/10 for MCP: Advanced Topics, citing outdated demos and orchestration topics covered only cursorily. If you already use these tools daily, most of the catalogue is below your level.
Overlap. A ranked review of the full catalogue found substantial repetition across the API, cloud and AI Fluency tracks, and concluded: “take fewer courses, but take the right ones.” The nine AI Fluency variants teach one framework nine times.
Drift. Anthropic ships faster than it can re-record video. Demo footage dates quickly, and a 67-lesson API course is expensive to keep current.
The praise is narrower and more credible for being so: the AI Fluency material is better than expected, and hands-on API work beats the theory sections.
What to actually take
Two paths, and neither is “all of it”.
If you write code: Claude Code 101 then Claude Code in Action — two hours total, and the fastest way to stop using the tool as autocomplete. Then Introduction to Model Context Protocol (1 hr) only if you are going to build a server; skip the advanced MCP course until it is re-recorded. Introduction to agent skills and Introduction to subagents are 45–60 minutes each and are the highest ratio of useful-to-elapsed on the platform. Take Building with the Claude API only if you are starting from zero on the API — nine hours of video against reading the docs is a poor trade if you are already shipping.
If you do not write code: Claude 101 (2.5 hr), then AI Fluency: Framework & Foundations (4 hr), then the one role variant that matches your job. Not the other eight. AI Capabilities and Limitations is the one to add if you want to understand why the thing fails the way it does rather than just how to drive it.
If you lead a team: the 4D framework is the reusable artefact here, and it is Creative Commons licensed. You can build your own internal training on it without asking anyone.
What this is not
Two other Anthropic learning resources exist and neither is the Academy.
github.com/anthropics/courses holds five hands-on notebook courses — API fundamentals, the prompt engineering interactive tutorial, real world prompting, prompt evaluations, and tool use. It has 22.7k stars, is public and not archived, and its README does not point at the Academy. It still favours Claude 3 Haiku to keep student API costs down, which tells you how current it is. For anyone who learns by running code rather than watching video, it is still the better artefact.
Coursera also carries Claude material, including AI Fluency. Be careful there: Intro to Claude AI on Coursera is published by Scrimba, not Anthropic.
Verdict
It is free, first-party, and the catalogue is genuinely large. That is worth something, and the AI Fluency framework is worth more than the courses wrapped around it.
But be clear about what you are getting. The badge certifies attendance. The credential that would certify competence costs $99 to $175, is proctored by Pearson, and you cannot sit it unless your employer is an Anthropic partner. For an experienced engineer the honest reading is four hours well spent — the two Claude Code courses, agent skills, subagents — and about 60 hours you can safely skip.
If you are already an AI-native developer, this is the whole list: Introduction to subagents (45 min), Introduction to agent skills (1 hr), Claude Code in Action (1 hr), and Introduction to Model Context Protocol (1 hr) only if you are about to build a server. That is three to four hours. Skip Claude Code 101, skip all nine AI Fluency variants, skip the 9-hour API course, and skip MCP: Advanced Topics until it is re-recorded.
Take fewer courses. Take the right ones.
- Your Agents Eat Your Subscription — what headless Claude and Codex actually draw down, and why a company seat is smaller than a personal plan.
- The Slash Is Load-Bearing — subagents and slash commands are not the same mechanism, and the Academy course is 45 minutes on one of them.
- Tech Debt Disguised as Best Practices — why vendor training tends to teach the happy path.