Best AI certifications in 2026 for engineers, by maker

The best AI certification for an engineer in 2026 is the one from the platform you build on, at the level of the work you do. On AWS that is the Generative AI Developer – Professional; on Azure, the Azure AI Apps and Agents Developer Associate (AI-103); on Databricks, the Generative AI Engineer Associate; on Google Cloud, the Professional Agentic Architect, whose full exam opens in mid-November; and on Claude, the Claude Certified Developer or Architect – Foundations, if your employer is a Claude partner. For data engineers, the platform's data engineering exam comes first, because the AI runs on that data.

Lists written before this summer are wrong in the places that matter. Microsoft retired AI-102, DP-100 and AZ-204 between June and July. AWS no longer offers the Machine Learning Specialty and is replacing its ML engineer exam. Google added an agent exam with a hands-on part. Anthropic and Databricks both launched exams that did not exist a year ago.

What changed in 2026

If you hold one of the retired Microsoft certifications, it stays valid on your transcript until it expires; you just cannot renew it after its retirement date. Our DP-203 to DP-700 piece covers the earlier Azure data change.

Building AI apps and agents

This is where most engineers are being asked to prove something, and every maker now has an exam for it.

AWS Generative AI Developer – Professional is the heaviest of the group: a professional-level exam, 3 hours and $300, about building generative AI on AWS services, Bedrock at the centre. If you already run Bedrock in production, it is the one that matches.

Microsoft AI-103 replaced AI-102 and moved the exam from Azure AI services to building agents and generative apps on Microsoft Foundry, including multi-agent orchestration, with Python assumed. AB-620 is for developers extending Microsoft's own agents for the enterprise and expects RAG, MCP and A2A knowledge; GH-600 tests working with coding agents in GitHub, Copilot, MCP servers and custom agents included. Pick AI-103 for apps you build on Azure, GH-600 if your job is shipping code with agents.

Google's Professional Agentic Architect is the one to watch. It is the only exam here with a practical part: pass the proctored multiple-choice exam, then complete hands-on labs. Google asks for 3 or more years building cloud solutions, including a year building agents on Google Cloud. Beta takers have until 31 December to finish the labs and hear 4 to 6 weeks after that; the full exam opens for registration in mid-November.

Databricks Generative AI Engineer Associate fits engineers building RAG and agents on lakehouse data; the newer Context Engineer Associate is about getting the right data in front of a model, retrieval and Genie included. Our Databricks certification guide compares all 8.

Claude certifications test building on the Claude API, Agent SDK, Claude Code and MCP directly, not a cloud's wrapper. They are well made and cheap, at $125 for the engineering exams, but only people at a Claude Partner Network organisation can register, with a company email. If that is you, the Claude certifications guide explains which of the 4 fits, and our Architect – Foundations practice exam is free.

Running ML in production

If your work is training, deploying and watching models rather than calling an LLM API:

• AWS Machine Learning Engineer – Associate. The current English exam is the MLA-C02 beta: 85 questions over 170 minutes for $75, a beta price. MLA-C01 is still offered in Japanese, Korean and Simplified Chinese until MLA-C02 is generally available.
• Microsoft Machine Learning Operations Engineer Associate (AI-300). The replacement for DP-100, and broader: MLOps plus GenAIOps on Azure, with GitHub Actions and the command line assumed. $165.
• Google Professional Machine Learning Engineer. 2 hours, 50 to 60 questions, $200, now updated for Gemini Enterprise Agent Platform, which replaced the Vertex AI name. Google says it does not directly test coding, though questions include Python and SQL snippets.
• Databricks Machine Learning Associate or Professional, if your models live on Databricks.

Engineering the data AI runs on

Most AI projects fail on the data, not the model, and every maker tests that separately:

The new one is DP-750: Microsoft's own exam for data engineering on Azure Databricks, pipelines, modelling, troubleshooting and Unity Catalog governance included. If you work on Azure Databricks, you now choose between Microsoft's exam and Databricks' own Data Engineer Associate. Databricks' is the wider-known badge across clouds; DP-750 sits better on a Microsoft partner's or employer's books.

The usual path for a data engineer moving into AI is data engineering first, then one AI exam on the same platform: Databricks Data Engineer Associate then Context Engineer, AWS Data Engineer then Generative AI Developer, DP-700 or DP-750 then AI-103.

Picking by platform and role

Not on any one platform yet? Choose the cloud your target employers use. The beginner exams, AWS AI Practitioner ($100, 90 minutes), Microsoft AI-901 and Google's Generative AI Leader ($99), are aimed at people new to AI or in business roles; an engineer with a year of building gets more from an associate or professional exam.

Are AI certifications worth it?

A certification proves you studied a platform's way of doing things; it does not prove you can ship. Its value is highest where an employer or partner programme counts them, such as cloud partner tiers and Claude partner tiers, and when you are moving into a platform you have not used at work. Pair it with something you built. Our free AI engineering course and data engineering course end with projects you can show, and the interview questions are closer to what a hiring panel asks than any exam.

Prices are in US dollars before tax and vary by country. Several exams above are in beta, which means a lower fee, results weeks later and content that may change before the full release.