Best MLOps certifications 2026: which test real MLOps

The best MLOps certification is the one for the platform your models run on, and only one vendor exam is actually built for MLOps: Microsoft AI-300, which earns Microsoft Certified: Machine Learning Operations Engineer Associate. Databricks Certified Machine Learning Professional goes deepest on operations within one platform, while Google and AWS test MLOps as about half of a broader machine learning exam.

This guide ranks five exams you can book today by how much of each blueprint is operations work: pipelines, registries, deployment, monitoring and drift. You'll also find the format and price of each, three retired exams that many lists still recommend, the free official prep, and an order that works with a point at which to stop.

What counts as an MLOps certification

MLOps is the work of turning a trained model into a maintainable service: repeatable pipelines, registries, controlled deployment, monitoring, drift detection and retraining. A certification does not need the word "MLOps" in its name to cover that work, but a model-development exam with one serving objective is not automatically an MLOps certification.

The other line to draw is between a certificate and a certification. A course-completion certificate shows that you finished the coursework. The credentials below are proctored vendor exams: free training can prepare you for them, but it does not replace the paid exam.

The five exams, by how much MLOps they test

Microsoft AI-300: Machine Learning Operations Engineer Associate

Microsoft Certified: Machine Learning Operations Engineer Associate, earned by passing AI-300, is live and generally available. It replaced DP-100, and it is the only option in this comparison whose credential is specifically for machine learning operations.

It is for intermediate MLOps engineers, data scientists and DevOps engineers who train, deploy, monitor and optimise machine learning and generative AI on Azure. Microsoft expects useful background in data science, Python, entry-level DevOps and infrastructure as code, but requires no earlier certification.

As of October 2026, AI-300 is an English-only, 120-minute proctored exam; Microsoft does not publish its question count. Microsoft prices exams by country or region, with Associate exams typically US$165 as of October 2026, so do not rely on a third-party local-currency price.

Classic MLOps infrastructure is 15-20% and model lifecycle and operations is 25-30%, making classic MLOps 40-50% of the blueprint. GenAIOps infrastructure adds 20-25%, so operations work including GenAIOps accounts for 60-75%; evaluation, tracing and cost are part of the remaining observability and optimisation material. Microsoft associate certifications last one year and renew through a free, short online assessment in the six-month renewal window.

The registry objective is about preserving lineage across environments, not merely naming a model: try a practice question on cross-workspace registry promotion. For production monitoring, use a practice question on data drift monitoring, where labels are not available in time for an accuracy check. CI/CD is also in scope, including a practice question on GitHub Actions OIDC for keyless Azure access.

Databricks Certified Machine Learning Professional

Databricks Certified Machine Learning Professional is live; Databricks publishes no separate exam code or retirement date. It is the senior of the two Databricks machine learning certifications, not an entry-level credential with a similar name.

It is for machine learning engineers who design, implement and manage enterprise-scale ML solutions using advanced Databricks capabilities. There is no formal prerequisite, although the vendor recommends related training and about one year of hands-on Databricks experience.

As of October 2026, the English-language exam has 59 scored multiple-choice questions, up to 10 unscored questions and 120 minutes, with no test aids. The price is US$200 plus applicable local taxes as of October 2026, and it can be taken online under proctoring or at a test centre.

ML Ops is 44% and Model Deployment is 12%, so 56% of the blueprint is direct operations and deployment work. The remaining 44% is Model Development, which includes advanced MLflow and Feature Store work but should not be counted again as operations. The certification is valid for two years; to renew, you pass the current live exam again.

Drift is tested as an operational decision rather than a definition. Work through a practice question on baseline drift statistical tests to separate a reference window from a consecutive-window comparison and choose a test for the column type.

Google Cloud Professional Machine Learning Engineer

Google Cloud Professional Machine Learning Engineer is live. Its exam guide was updated for the move from Vertex AI to Gemini Enterprise Agent Platform, so prepare with the current product names rather than assuming older Vertex AI material uses the names that appear on the exam.

It is for engineers who build, evaluate, productionise and optimise AI solutions on Google Cloud. There are no formal prerequisites, but Google recommends substantial industry and Google Cloud experience, so it is a professional target rather than a first exposure to machine learning.

As of October 2026, it has 50-60 multiple-choice and multiple-select questions in two hours, offered in English and Japanese. It costs US$200 plus applicable tax as of October 2026 and is available through remote or test-centre proctoring.

Serving and scaling models is about 20%, automating and orchestrating ML pipelines about 18%, and monitoring AI solutions about 13%. Those three direct MLOps sections total about 51% of the blueprint; data, experimentation and model-building work make up the rest. The certification is valid for two years and renewal requires the standard exam, rather than a shorter renewal test.

Monitoring includes comparing training and serving behaviour, as in a practice question on training-serving skew detection. Pipeline automation also means changing the build path when components change; see a practice question on CI/CD pipeline component changes.

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

AWS Certified Machine Learning Engineer – Associate MLA-C02 is a beta, not generally available; its beta registration code is ME1-C02. It succeeds MLA-C01, and beta delivery is English only. AWS's September 2026 announcement puts general availability on January 14, 2027, although the certification page itself still lists the GA dates as TBD.

It is for backend developers, DevOps engineers, data engineers, MLOps engineers, data scientists, ML engineers and LLMOps engineers working with SageMaker AI and Bedrock. There is no formal prerequisite, although AWS recommends experience in a related role and with the relevant services.

As of October 2026, the MLA-C02 beta has 85 questions and 170 minutes. Its beta price is US$75 as of October 2026; do not confuse that discount with a generally available exam price.

Deployment and orchestration is 24%, while operating, monitoring and securing ML and AI solutions is another 24%. That makes its direct MLOps share 48%; data preparation can support pipelines, but it is not included in that calculation. AWS certifications are valid for three years and are renewed by passing the current exam or an eligible higher-level role-based exam.

There is no MLA-C02 question bank on this site yet. For the version change and what the beta replaces, use the AWS MLA-C01 vs MLA-C02 guide rather than booking an archived exam by mistake.

Databricks Certified Machine Learning Associate

Databricks Certified Machine Learning Associate is live, with its current blueprint effective March 1, 2025. Databricks does not publish an exam code, and this Associate credential is not the same assessment as the Professional exam above.

It is for ML engineers and data scientists learning foundational Databricks work: AutoML, Unity Catalog feature tables, MLflow tracking and registry, training and deployment. It has no required prerequisite, though six months of hands-on experience and related training are recommended.

As of October 2026, it has 48 scored multiple-choice and multiple-selection questions in 90 minutes, with up to 10 unscored items. It is offered in English, Japanese, Brazilian Portuguese and Korean, and costs US$200 plus applicable local taxes as of October 2026.

The Databricks Machine Learning domain is 38% and is the exam's MLOps layer: it covers strategy, feature tables, MLflow runs, the Unity Catalog registry, aliases and promotion. Model Deployment adds 12%, for a 50% MLOps-and-deployment share, but the outline has no monitoring, drift, retraining or CI/CD domain. The credential is valid for two years and is renewed by passing the current live exam again.

Promotion is deliberately decoupled from application code; try a practice question on Unity Catalog model alias promotion. Before promoting anything, find the winning experiment correctly with a practice question on MLflow search for the best run.

How they differ

Platform: AI-300: Azure ML and Microsoft Foundry / Databricks Professional and Associate: the Databricks lakehouse / Google: Google Cloud and Gemini Enterprise Agent Platform / AWS: SageMaker AI and Bedrock.
MLOps share: AI-300: 40-50% classic MLOps, plus 20-25% GenAIOps / Databricks Professional: 56% operations and deployment / Google: about 51% serving, pipelines and monitoring / AWS MLA-C02 beta: 48% deployment, operation and monitoring / Databricks Associate: 50% registry, feature, workflow and deployment coverage without a monitoring or CI/CD domain.
GenAIOps: AI-300: a major, measured part of the blueprint / Google and AWS: generative AI included in broader ML and AI scope / both Databricks ML exams: choose them for the platform's ML lifecycle rather than a dedicated GenAIOps credential.
Level: Databricks Associate: foundations on one platform / AI-300 and AWS MLA-C02 beta: associate-level role credentials / Google PMLE and Databricks Professional: experienced-engineer targets.
Price: AI-300: country-based, Associate exams typically US$165 / Databricks Professional, Databricks Associate and Google PMLE: US$200 plus applicable taxes / AWS MLA-C02: US$75 beta price.
Renewal: AI-300: one year with a free online renewal assessment / Databricks and Google: two years and a current standard exam / AWS: three years and the current or an eligible higher-level role-based exam.

Which MLOps certification should you take

You work in an Azure shop: take AI-300. It is the direct Azure MLOps certification and covers both the standard model lifecycle and GenAIOps. If you need the wider Microsoft sequence, see the Microsoft AI certification path.
You work on a Databricks lakehouse: take Databricks Machine Learning Associate if registry, MLflow and serving are new to you; take Professional if you already operate production Databricks pipelines and monitoring. The latter is the deeper single-platform operations exam.
You work on Google Cloud: take Professional Machine Learning Engineer. There is no separate Google MLOps certification; this is the closest Google credential, and about half of its blueprint is serving, orchestration and monitoring.
You work on AWS: MLA-C02 is the AWS answer only if you understand that it is still a beta. Use it for the current SageMaker AI and Bedrock direction; do not book an MLA-C01 English slot that no longer exists.
You are new to ML: start with hands-on learning and choose Databricks Associate only if Databricks is your intended platform. It has no formal prerequisite, but an exam is useful after you can train, register and serve a model rather than before.
You are already a data scientist: choose the platform that will own your model in production. Databricks Professional suits established Databricks work, while AI-300 or Google PMLE makes more sense when Azure or Google Cloud is the deployment environment.
You are a DevOps engineer moving to ML: AI-300 maps most directly to infrastructure, identity, CI/CD, endpoints and monitoring on Azure. On AWS, MLA-C02 beta has a similar deployment-and-operations emphasis, but it remains a beta choice.

An order that works, and when to stop

First, choose the cloud or lakehouse you will actually use and build one small path from training data to a deployed model, with a registry and a monitoring decision. Then take the exam that matches that platform: AI-300 for Azure, Databricks Associate before Professional where you need the platform fundamentals, Google PMLE for an established Google Cloud role, or MLA-C02 only if the AWS beta status suits your timing.

Take a second certification only when it proves a real change in scope. For example, Databricks Associate to Professional makes sense when you move from model registry basics to automation, retraining and Lakehouse Monitoring. Stop after the credential that matches your job plus a working project; collecting one exam from every cloud is not a substitute for operating a model in the stack an employer uses. For a wider platform comparison, read the best AI certifications.

Exams you will still see recommended, but can't take

AWS Certified Machine Learning – Specialty (MLS-C01): retired on 2026-03-31. Old lists and even stale vendor material can still surface it, but it is not a booking option.
Microsoft DP-100, Azure Data Scientist Associate: retired on 2026-06-01. AI-300 is its current MLOps-oriented replacement.
AWS MLA-C01 in English: retired on 2026-09-28. Other-language delivery continues only until MLA-C02 reaches general availability, so do not treat the English version as a current exam.

What they cost and how to prepare for free

As of October 2026, Microsoft Associate exams typically cost US$165 but are priced by the country or region of proctoring. Databricks Professional, Databricks Associate and Google PMLE are US$200 plus applicable taxes, while MLA-C02's US$75 is specifically a beta price. If you are booking from India or Singapore, the final amount in local currency, with tax, shows at checkout; none of these vendors publishes a fixed rupee or Singapore-dollar price on its exam page.

Use the official free material before paying for any third-party practice product. Microsoft provides two AI-300 Learn paths, a practice assessment and an exam sandbox. Databricks provides exam guides, an AI Prep Guide, free self-paced Academy courses and Free Edition hands-on work; the Associate has sample questions but no official practice exam. Google provides an exam guide, sample questions and documentation, while its Skills learning path is only partly free. AWS offers a free exam guide, a two-hour Exam Prep course and Cloud Quest; its full-length practice exams need a Skill Builder subscription.

Once you have worked through the official material, practise with explained questions. This site has free AI-300 practice tests, Databricks ML Professional practice tests, Databricks ML Associate practice tests and Google PMLE practice tests, each with an explanation for every option. To revise one topic at a time, browse the full lists: AI-300 questions, Databricks ML Professional questions, Databricks ML Associate questions and PMLE questions. They are written to teach the reasoning behind a blueprint topic, never copied from a real exam.

Frequently asked questions

Is there an AWS MLOps certification?

AWS's relevant credential is AWS Certified Machine Learning Engineer – Associate. MLA-C02 is in beta as of October 2026, with booking code ME1-C02, 85 questions, 170 minutes and a US$75 beta price; AWS has announced general availability for January 14, 2027. MLS-C01, the older Machine Learning – Specialty, retired on 2026-03-31.

Is there an Azure MLOps certification?

Yes. Microsoft AI-300 earns Microsoft Certified: Machine Learning Operations Engineer Associate and is the only exam here built specifically for MLOps. It replaced DP-100, which retired on 2026-06-01, and combines classic MLOps with a large GenAIOps share.

Is there a free MLOps certification?

None of the four vendors offers a free proctored MLOps certification; every exam here has a fee. What is free is the preparation: Microsoft Learn paths and the AI-300 practice assessment, Databricks Academy courses, Google's exam guide and sample questions, and AWS's exam guide. A free course certificate shows you finished a course, which is not the same as passing an exam.

Which MLOps certification should I get first?

Pick the one for the platform where you will deploy models: AI-300 for Azure, Databricks Associate for Databricks fundamentals, Google PMLE for an experienced Google Cloud role, or the AWS MLA-C02 beta if its status is acceptable. There is no general first exam that is better than learning the registry, deployment and monitoring tools in your own stack.

Is an MLOps certification worth it?

It is useful when it documents skills in the platform your current or target role uses. It is not enough on its own: a deployed model with versioning, release control and a monitoring choice shows the practical work that an operations-focused blueprint describes.