Best AI certifications 2026: AWS, Google, Microsoft, Claude

The best AI certification for most people is the entry-level exam from the platform they already use at work: AWS Certified AI Practitioner, Google Cloud Generative AI Leader or Microsoft's AI-901. If you build AI applications, go one level up to the engineer exam on that same platform, such as Microsoft AI-103 or Databricks Generative AI Engineer Associate.

This guide sorts the AI certifications from AWS, Google Cloud, Microsoft, Anthropic (Claude) and Databricks into three levels, with the same facts for each: who it is for, format, price and status. It also explains the difference between a course certificate and a certification exam, lists what changed in 2026, and suggests which one to take for your situation. Every exam fact was checked on the vendor's own pages on October 20 and 21, 2026, which matters because four AI exams retired or changed version this year.

How AI certifications are organized

Across the five vendors, AI certifications fall into three levels:

Entry level: what AI and generative AI are, which service fits which job, and responsible use. Little or no coding. These suit managers, analysts, sellers and anyone new to AI.
Builder level: building and running AI applications on one platform, such as agents, retrieval-augmented generation (RAG), evaluation and deployment. These assume hands-on work.
Professional level: designing production AI systems end to end, usually with two or more years of experience recommended.

None of the five vendors makes an entry-level exam a requirement for a higher one. The levels describe difficulty, not a fixed order.

Entry-level AI certifications

AWS Certified AI Practitioner (AIF-C01)

AWS writes this exam for people with up to six months of exposure to AI and machine learning on AWS who use AI solutions but don't necessarily build them. It has 65 questions in 90 minutes, a passing score of 700 out of 1,000, and costs 100 USD, which AWS also lists as 8,553 INR (as of October 2026). It is valid for three years. Applications of foundation models (28%) and generative AI fundamentals (24%) are its largest domains, and the AI and ML fundamentals domain (20%) asks choices like this practice question on traditional ML vs a foundation model. Practice tests are on the AI Practitioner page, and our AWS certification roadmap shows where it leads.

AWS also launched a business-level exam, AI Business Strategist (AIB-C01), which is in beta as of October 2026.

Google Cloud Generative AI Leader

Google aims this exam at anyone in any role, technical or not, and it tests strategy rather than implementation. It has 50 to 60 multiple-choice questions in 90 minutes and costs 99 USD (as of October 2026); Google does not publish a passing score. It is valid for three years and offered in English, Japanese, Spanish and Portuguese. Google Cloud's own AI offerings are its biggest section (35%), so it rewards knowing the product names, which our map of Google Cloud's generative AI offerings covers. Try a practice question on grounding answers in internal policies, or start the Generative AI Leader practice tests.

Microsoft Azure AI Fundamentals (AI-901)

AI-901 replaced AI-900, which retired on June 30, 2026. Microsoft pitches it at people at the beginning of a career in AI solution development. Unlike the old concepts-only exam, 55-60% of it is implementing AI solutions in Microsoft Foundry, and Microsoft expects you to know basic Python syntax. The passing score is 700, and fundamentals certifications do not expire. Microsoft prices exams by country and does not print the AI-901 price on its exam page.

Claude Certified Associate – Foundations (CCAO-F)

Anthropic's entry exam is for consultants, sellers and delivery leads who guide customers toward Claude use cases. It has 60 items in 120 minutes, a passing score of 720, and costs 99 USD (as of October 2026). The catch is eligibility: every Claude certification is open only to people working at a company in the Claude Partner Network. Our Claude certifications comparison covers all four exams and the partner rules.

Builder-level AI certifications

Microsoft AI-103: Azure AI Apps and Agents Developer Associate

AI-103 replaced AI-102, which retired on June 30, 2026. It is for engineers who build and deploy agents and AI solutions on Microsoft Foundry, in Python. It runs 120 minutes with a passing score of 700, and its biggest domain is generative AI and agentic solutions (30-35%). Microsoft says associate exams typically cost 165 USD, varying by country. Associate certifications expire every year, and renewal is a free online assessment. Expect security questions like this practice question on indirect prompt injection; our AI-103 exam guide covers the rest, and the AI-103 practice tests are free.

For the operations side, AI-300 (Machine Learning Operations Engineer Associate) covers MLOps and GenAIOps on Azure: 120 minutes, a passing score of 700, English only. There are AI-300 practice questions here too.

AWS Certified Machine Learning Engineer – Associate (MLA-C02, beta)

MLA-C02 covers both classic machine learning on SageMaker AI and foundation models on Amazon Bedrock, including RAG, agents and Guardrails. As of October 2026 it is in beta: 85 questions in 170 minutes, English only, for 75 USD, and AWS lists its general availability date as "TBD". The English MLA-C01 exam ended on September 28, 2026. Our MLA-C01 vs MLA-C02 guide explains the change.

Databricks Certified Generative AI Engineer Associate

This exam is for engineers who build and deploy generative AI applications on Databricks, with six months of hands-on experience recommended. It has 45 scored multiple-choice questions in 90 minutes and costs 200 USD (as of October 2026); Databricks does not publish a passing score. It is valid for two years, and you renew by passing the current exam again. Application development is the largest section (30%), with retrieval details like this practice question on chunk size and the embedding limit. See the Databricks GenAI Engineer practice test.

Databricks also lists a new Context Engineer Associate exam, about designing and governing the context that AI agents use.

Claude Certified Developer and Architect – Foundations

The Developer – Foundations exam (CCDV-F) is for engineers who build with the Claude API, Claude Code and the Model Context Protocol; it has 53 items and costs 125 USD. Architect – Foundations (CCAR-F) is for people who design Claude solutions end to end; it has 60 items and costs 125 USD. Both run 120 minutes with a passing score of 720, and both are valid for 12 months (as of October 2026). Architect questions look like this practice question on choosing a workflow or an agent, and there are Claude Architect practice tests here.

Google Cloud Professional Machine Learning Engineer

This is Google's exam for ML engineers who build, productionize and optimize both traditional and generative AI on Google Cloud. It sits here rather than with the professional exams because it is the builder exam on Google's side. Google recommends three or more years of industry experience, including a year on Google Cloud. It has 50 to 60 questions in two hours and costs 200 USD (as of October 2026), in English and Japanese. Google has updated it for the rename of Vertex AI to Gemini Enterprise Agent Platform, so older study material uses the old names. Practice with the Professional ML Engineer questions.

Professional-level AI certifications

AWS Certified Generative AI Developer – Professional (AIP-C01)

AWS's professional generative AI exam is for developers with two or more years building production applications and a year of hands-on generative AI work. It has 75 questions in 180 minutes, a passing score of 750, and costs 300 USD, or 25,659 INR (as of October 2026). It is valid for three years. The beta ended on March 31, 2026, and the standard exam now covers Amazon Bedrock AgentCore. Foundation model integration, data management and compliance is its largest domain (31%).

Google Cloud Professional Agentic Architect (beta)

This brand-new Google exam is for experienced developers and architects who build autonomous agentic workflows. It comes in two parts: about 80 multiple-choice questions in three hours, then hands-on labs for those who pass. As of October 2026 it is in beta, with beta registration closed and general availability expected in November 2026. The beta price was 120 USD, 40% off a 200 USD retail price. It is valid for only one year, and custom agents are its biggest domain (33%). The Professional Agentic Architect guide covers it in detail.

Claude Certified Architect – Professional (CCAR-P)

The only Professional level in Anthropic's program so far: 63 items in 120 minutes, a passing score of 720, 175 USD (as of October 2026), and the same partner-only eligibility as the other Claude exams.

Databricks Certified Machine Learning Professional

For people who design and manage enterprise-scale ML on Databricks, with a year of hands-on experience recommended. It has 59 scored questions in 120 minutes, costs 200 USD, is English only, and is valid for two years (as of October 2026). Model development and ML operations are 44% each. Practice with the ML Professional questions.

Course certificates vs certification exams

Many "free AI certification" results are course certificates, not certifications. A course certificate proves you finished a course; a certification proves you passed a proctored exam.

Google AI Essentials: a course of under five hours on prompting and everyday AI use. In the US and Canada it costs 49 USD a month on Coursera after a seven-day trial. No exam.
Google AI Professional Certificate: eight courses on Coursera at the same 49 USD a month. Graded assessments inside the courses, but no proctored exam.
Introduction to Generative AI (Google Skills): a 45-minute introductory course that earns a badge.
Career Essentials in Generative AI by Microsoft and LinkedIn: five courses, about four hours, with a certificate of completion on LinkedIn Learning.
Claude Academy: free Anthropic courses; Claude 101 ends with a certificate of completion.
AWS Skill Builder: a free tier of over 900 courses, including the free foundational AI training in AWS's AI Practitioner exam prep plan.

These are good, cheap ways to learn. If an employer asks for a certification, though, they usually mean a proctored exam like the ones above.

How the AI certifications differ

AWS vs Google at entry level: AI Practitioner tests AI and ML fundamentals plus AWS services; Generative AI Leader focuses on strategy and Google Cloud's offerings for non-technical roles.
Microsoft vs everyone else on renewal: Microsoft associate certifications expire yearly but renew free online; AWS certifications last three years; Databricks two; Claude 12 months with a free renewal; Google's Agentic Architect one year.
Cloud platforms vs Claude: AWS, Google, Microsoft and Databricks exams are open to anyone; Claude exams need a Claude Partner Network employer.
Published pass marks: AWS (700-750), Microsoft (700) and Anthropic (720) publish them; Google and Databricks do not.
Beta status: MLA-C02, AWS AI Business Strategist and Google's Professional Agentic Architect are in beta as of October 2026; everything else in this guide is generally available.

Which AI certification should you take

You are new to AI and not a developer: Google Generative AI Leader or AWS AI Practitioner, depending on which cloud your employer uses.
Your company runs on Azure: AI-901 to start, then AI-103 if you build applications.
You build generative AI apps on AWS: AI Practitioner, then Generative AI Developer – Professional once you have the experience it recommends.
You work on a lakehouse or data platform: Databricks Generative AI Engineer Associate.
You are an ML engineer: Google Professional ML Engineer, Databricks ML Professional, or MLA-C02 once it is out of beta.
Your company is a Claude partner: the Claude Foundations exam that matches your role.
You only want to learn, not certify: a free course such as Claude Academy or Skill Builder, then decide.

What changed in AI certifications in 2026

If a list you are reading recommends any of these, it is out of date:

AWS Machine Learning – Specialty (MLS-C01): retired on March 31, 2026.
Microsoft AI-900 and AI-102: both retired on June 30, 2026, replaced by AI-901 and AI-103.
AWS MLA-C01: the English exam ended on September 28, 2026; MLA-C02 is in beta.
Google's AI platform names: Vertex AI is now Gemini Enterprise Agent Platform in Google's exam guides.

AI certification cost and preparation

As of October 2026, entry-level AI exams cost about 100 USD: 100 USD for AWS AI Practitioner, 99 USD for Google Generative AI Leader and Claude Associate. Leaving out the MLA-C02 beta (75 USD), builder exams run from 125 USD (Claude) to 200 USD (Databricks, Google Professional ML Engineer), and professional exams from 175 USD (Claude Architect – Professional) to 300 USD (AWS). AWS publishes rupee prices; Google, Databricks and Anthropic list USD; Microsoft prices by country at booking.

Official preparation exists for all of them. AWS Skill Builder has a free tier, and its 20-question Official Practice Question Sets are free; Google publishes exam guides, sample questions and Skills Boost learning paths; Microsoft offers a free practice assessment for AI-103; Anthropic publishes an exam guide with sample questions for each exam and prep courses in its Partner Academy; Databricks offers free Academy training but no official practice exams.

The practice tests here are free, need no account, and explain every option. Each cert has a full question list, so you can study by topic: AI Practitioner questions, Generative AI Leader questions, AI-103 questions, Claude Architect questions and Databricks GenAI Engineer questions. If RAG is new to you, the RAG vs fine-tuning guide covers a choice every one of these exams tests.

Frequently asked questions

Which AI certification is best for beginners?

AWS Certified AI Practitioner and Google Cloud Generative AI Leader. Both are around 100 USD, take 90 minutes, and assume no AI background. Pick the one that matches the cloud your employer or target employer uses.

Is an AI certification worth it?

It is worth it when it matches the platform you work on or want to work on, because it shows you know that vendor's tools and their limits. On its own it won't replace hands-on projects, so pair a builder-level exam with something you have built.

Is there a free AI certification?

The proctored certifications in this guide all cost money. Free options are course certificates, such as Claude Academy's Claude 101 and AWS Skill Builder's free tier, which prove you finished a course rather than passed an exam.

Google vs AWS vs Microsoft: which AI certification is best?

None is best in general. At entry level, Generative AI Leader is the most business-focused, AI Practitioner balances concepts and services, and AI-901 expects some Python. For builders, AI-103 is Microsoft's agent-building exam. Choose by the platform you use.

Do I need to know how to code?

Not for the entry-level exams from AWS, Google or Anthropic. Microsoft's AI-901 expects basic Python syntax, and every builder and professional exam assumes you write code.

How long do AI certifications last?

It varies: AWS three years, Google Generative AI Leader three years, Databricks two, Microsoft associate one year with free renewal, Claude 12 months with free renewal, and Microsoft fundamentals do not expire.