Properly catch boto3 Errors

Use Boto3 exceptions: https://www.programcreek.com/python/example/97944/boto3.exceptions

client = boto3.client('logs')
try:
   client.create_log_group(logGroupName=LOG_GROUP)
except client.exceptions.ResourceAlreadyExistsException:
   pass

You've summarized the situation well. The old boto had a simple hardcoded approach to supporting AWS APIs. boto3, in what appears to be an attempt to reduce the overhead of keeping Python client synced with evolving features on the various apis, has been more squishy around exceptions, so the ClientError approach you outlined above used to be the canonical way.

In 2017 they introduced the second mechanism you highlight: 'modeled' exceptions available on the client.

I am not familiar with SNS but in my experience with other AWS products, the ClientError naming matches up with the HTTP apis, which tend to be well documented. So I would start here: https://docs.aws.amazon.com/sns/latest/api/Welcome.html

It looks like the new-style modeled exceptions are generated from service definition files that live in botocore module. I can't find any documentation about it, but go browse around the AWS service models in https://github.com/boto/botocore/tree/master/botocore/data.

Also, it's good to know that if you are not (in contrast to OP's code) dealing directly with the low-level client, but instead are using a high-level AWS ServiceResource object, a low-level client is still easily available at my_service_resource.meta.client so you can handle exceptions like this:

try:
    my_service_resource.do_stuff()
except my_service_resource.meta.client.exceptions.NotFoundException as e:
    # handle exception