Best Machine Learning Certifications

For years, one exam sat at the top of every machine learning certification list: the AWS Certified Machine Learning – Specialty. You can no longer take it. AWS retired MLS-C01 on 31 March 2026, and Microsoft retired DP-100, the Azure Data Scientist Associate exam, on 1 June 2026. If the ML certification you had bookmarked is one of those two, your shortlist needs rebuilding from scratch — which is exactly what this article does.

The scope here is deliberately narrow: certifications that test machine learning engineering and modelling — training, tuning, deploying and operating models. Broad AI credentials that cover the whole field are ranked in our best AI certifications for 2026, and LLM-centred credentials get their own treatment in the generative AI certifications guide. What remains after those filters is a short, strong list.

What actually counts as an ML certification in 2026?

A working definition: a proctored exam whose objectives centre on the ML lifecycle — data preparation, model training and evaluation, deployment, and production operations (MLOps) — rather than on consuming pre-built AI services or on business literacy. By that standard, four credentials make the core list, with two ecosystem-specific additions worth considering. Course certificates such as TensorFlow-focused specialisations or Coursera programmes can teach you a great deal, but they are not invigilated credentials, so they sit outside this ranking.

The core four

AWS Certified Machine Learning Engineer – Associate

Who it fits: ML practitioners on AWS, and the default pick when no other constraint applies.

This is the closest successor to the retired Specialty and now AWS's flagship ML credential. The MLA-C01 exam runs 65 questions in 130 minutes for $150 USD (regional pricing varies), with a 720 scaled passing score and no prerequisites — AWS suggests about a year of hands-on work with SageMaker and related services. It uses AWS's newer question types (ordering, matching, case study) alongside multiple choice, which rewards genuine workflow understanding over recall.

Version alert: MLA-C02 registration opens on 1 September 2026, the last English sitting of MLA-C01 is 28 September 2026, and the C02 revision folds in generative AI, agentic AI and foundation-model workloads. The certification lasts three years, and earning it auto-recertifies the foundational AI Practitioner. If you are studying now, check AWS's exam page to confirm which version your date falls under.

Google Cloud Professional Machine Learning Engineer

Who it fits: experienced practitioners who want the most demanding, design-orientated ML exam on the market.

Google's Professional Machine Learning Engineer costs $200 USD plus tax and runs 50–60 multiple-choice and multiple-select questions over two hours. There are no exam codes (Google Cloud does not use them), no prerequisites, and no published passing score — results are pass/fail only. Google recommends three-plus years in industry with at least one year on Google Cloud, and although the exam does not directly assess coding, Google advises baseline Python and SQL so you can read code snippets.

The 2026 refresh matters: Google has updated the exam to reflect the transition from Vertex AI-era content toward the Gemini Enterprise Agent Platform, alongside product renames. Prep material written for the old Vertex-centric exam will steer you wrong. Timed Professional Machine Learning Engineer practice questions are worth using early here — not at the end of your prep — because the fastest way to discover this exam's scenario style is to fail some scenarios cheaply.

Databricks Machine Learning Associate and Professional

Who it fits: practitioners in Databricks shops, where the lakehouse platform is the ML workbench.

Databricks maintains a two-tier machine learning track — the Certified Machine Learning Associate and the Certified Machine Learning Professional — covering ML work on the Databricks platform at practitioner and advanced level respectively. We have not verified current formats and fees for these two exams against official pages, so take those details from Databricks' certification site directly. What we can say from the wider Databricks portfolio: its exams are proctored, scenario-based, and typically valid for two years with renewal by retaking the current exam — a heavier maintenance cycle than AWS's three years.

If your organisation runs Databricks, these certifications map to your daily tooling more tightly than any cloud-generic exam will. You can review the exam pages for the Databricks Certified Machine Learning Associate and Databricks Certified Machine Learning Professional to see the shape of each.

Microsoft's MLOps successor to DP-100

Who it fits: ML practitioners in Azure environments — once the new exam settles.

DP-100 and the Azure Data Scientist Associate certification are gone; Microsoft's replacement path is a new MLOps-focused associate certification via exam AI-300, orientated toward production machine learning operations and generative AI operations. At the time of research Microsoft had not published full official details for AI-300 — treat the exam code and positioning as announced, and confirm format, pricing and objectives on Microsoft Learn before committing study time. What is stable is Microsoft's policy layer: associate certifications are scored out of 1,000 with 700 to pass, expire after one year, and renew free through an open-book online assessment. Do not buy DP-100 prep material in 2026; it can no longer earn you anything.

AWS MLA vs Google PMLE: the comparison most people are really making

Searches for "AWS Machine Learning Specialty vs Google ML Engineer" now resolve to this match-up, since the Specialty is retired. The honest answer is that they test different altitudes of the same job.

Factor	AWS ML Engineer – Associate	Google Professional ML Engineer
Level	Associate	Professional
Difficulty	Applied, workflow-centred	Design-heavy, broader judgement calls
Prerequisites	None (~1 yr AWS ML suggested)	None (3+ yrs industry, 1+ yr GCP suggested)
Cost	$150 USD	$200 USD + tax
Format	65 Q, 130 min, scaled score 720 to pass	50–60 Q, 2 hrs, pass/fail only
Best for	Engineers shipping ML on AWS	Senior practitioners on Google Cloud
Career signal	Platform ML engineering competence	Architecture-level ML judgement
Renewal	3 years; higher AWS exams auto-renew it	Per Google's renewal FAQ; shorter professional-tier terms

Neither is universally better. Pick by platform first: certifying against a cloud your employer does not run converts to interview value slowly. If you genuinely work across both, the AWS exam is the gentler entry and the Google exam the stronger senior signal — many practitioners sequence them in that order. How that sequencing extends into a multi-year plan is the subject of our machine learning certification roadmap.