Cognitive Class occupies an unusual position. It is run by IBM's Skills Network, it is genuinely free all the way through including the certificate at the end, and it has been running long enough to have accumulated a catalogue that most paid platforms would envy. Three million learners have passed through it. And yet almost nobody talks about it, which tells you something about how much of online education is marketing spend rather than substance.
The single best thing here is the lab environment. Every serious course drops you into a browser based notebook with the data already loaded and the dependencies already resolved. If you have ever tried to teach someone data science, you know that the first two weeks are usually lost to Python installations, path variables, conda versus pip arguments and the particular flavour of despair that comes from a package that will not build. Cognitive Class deletes all of that.
You click a button and you are writing pandas against a real dataset in about ninety seconds. For a beginner that is worth more than any amount of polished video, because it moves the point of first contact with actual code from week three to minute one. The catalogue is broad. Data analysis with Python, machine learning with Python, deep learning fundamentals, SQL and relational databases, data visualisation, plus a long tail of cloud, container, blockchain and generative AI material of varying vintage.
The learning paths are the right way in, because the raw course list is a jumble and the paths impose an order that actually makes sense. The guided projects, of which there are several hundred, are the underrated part. They are short, they are specific and they give you something to point at, which is more than most free courses manage. Now the problems, and they are real.
Maintenance is patchy. This is the recurring complaint and it is fair. Some courses are kept current and some have visibly not been touched in years, still teaching against library versions and interfaces that have since changed. Nothing is more demoralising for a beginner than following instructions exactly and getting an error, because they have no way to tell whether they made a mistake or the course did.
You will hit this. When you do, the fix is almost always a deprecation, and knowing that in advance will save you an evening. The second problem is the IBM gravity well. This is IBM's platform and it teaches IBM's stack.
Watson Studio, IBM Cloud, IBM's own service endpoints. Some of this is unavoidable and some of it is fine, because the underlying concepts transfer. But a beginner cannot tell which parts are industry standard and which parts are vendor specific, and Cognitive Class does not go out of its way to draw that line. You can finish a course feeling competent in something that almost no employer outside the IBM ecosystem uses.
Keep a mental note: pandas, scikit-learn, PyTorch and SQL are the transferable parts. Everything with a Watson prefix is a product, not a skill. Third, the badges. IBM issues them through Credly and they look official, which is precisely the problem.
They are awarded for completing material and passing a light quiz, not for demonstrating capability against a standard. Early on they carried some novelty value on LinkedIn. That window has closed. Recruiters have now seen enough profiles wallpapered with free badges to discount them entirely, and a stack of them can actively read as someone who collects certificates instead of building things.
Take the courses for the learning and put one or two of the most relevant badges on your profile if you like, but do not mistake the collection for a portfolio. There is also an organisational muddle worth naming. IBM runs Cognitive Class, IBM SkillsBuild and a large set of paid IBM Professional Certificates on Coursera, and the boundaries between them are not explained anywhere obvious. Broadly: Cognitive Class is the free technical catalogue with the good labs, SkillsBuild is the free careers and employability programme aimed at students and jobseekers, and the Coursera certificates are the paid, structured, credential bearing route.
Content flows between them. If you have done the free version you will find real overlap with the paid Coursera material, which is a good argument for starting here before you pay IBM anything. My rating reflects a straightforward calculation. The price is zero, the labs are excellent, the breadth is real and the friction of getting started is close to nothing.
Against that, the maintenance is uneven, the vendor lean is significant and the credential is worth very little. If this cost money I would rate it lower, because the currency issues would be inexcusable. As a free resource that lets a complete beginner write working machine learning code on the same evening they decide to try, it is one of the better things IBM does. Use it as a hands-on sandbox and a way to find out whether you actually like this work.
Then go and build something of your own, which is the only credential that has ever really counted.