An Azure relational database service.
Forecasts for Serverless are inherently inaccurate, because it all depends on how you use the service. First you need to understand that when Serverless is paused it is rolled out and exists only disk, and in this state you pay only for storage. Once someone connects to it, it is rolled in and mounted on a server, and now you start to pay for compute.
Serverless is great for a database that you only use occasionally, like an hour a day and it is rolled out most of the day. It's a really bad option for a database that is always active, because the price for compute is higher for a provisioned general-purpose database. One risk with serverless is that you think you are not using it, but there is something that pings it every five minutes which prevents it from being paused. Keep in mind, it is the fact that it is mounted you pay for, not the actual CPU time you consume.
If you want something which is really budget-friendly, I recommend an S0 or S1 (i.e. a DTU-based model). They are really cheap. And, no, they are overly powerful, but if you are only learning and playing, they can still serve you well.
But going back to Serverless, it can be a great choice, if you understand the service. If you don't, it will be expensive.