I had the same use case and the hint of @GrahamDumpleton to run oc with --loglevel 9 was very helpful. This is what oc scale does:

1. It makes a get request to the resource, receiving some JSON object
2. Then it makes a put request to the resource, with a modified JSON object (the number of replicas changed) as payload

If you're doing this you don't have to worry about setting the apiVersion, you just reuse what you get in the first place.

Here is a small python script, that follows this approach:

"""
Login into your project first `oc login` and `oc project <your-project>` before running this script.

Usage:
    pip install requests
    python scale_pods.py --deployment-name <your-deployment> --nof-replicas <number>

"""
import argparse
import requests
from subprocess import check_output
import warnings
warnings.filterwarnings("ignore")  # ignore insecure request warnings


def byte_to_str(bs):
    return bs.decode("utf-8").strip()


def get_endpoint():
    byte_str = check_output("echo $(oc config current-context | cut -d/ -f2 | tr - .)", shell=True)
    return byte_to_str(byte_str)


def get_namespace():
    byte_str = check_output("echo $(oc config current-context | cut -d/ -f1)", shell=True)
    return byte_to_str(byte_str)


def get_token():
    byte_str = check_output("echo $(oc whoami -t)", shell=True)
    return byte_to_str(byte_str)


def scale_pods(deployment_name, nof_replicas):
    url = "https://{endpoint}/apis/apps.openshift.io/v1/namespaces/{namespace}/deploymentconfigs/{deplyoment_name}/scale".format(
        endpoint=get_endpoint(),
        namespace=get_namespace(),
        deplyoment_name=deployment_name
    )
    headers = {
        "Authorization": "Bearer %s" % get_token()
    }
    get_response = requests.get(url, headers=headers, verify=False)

    data = get_response.json()
    data["spec"]["replicas"] = nof_replicas
    print(data)

    response_put = requests.put(url, headers=headers, json=data, verify=False)
    print(response_put.status_code)


def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--deployment-name", type=str, required=True, help="deployment name")
    parser.add_argument("--nof-replicas", type=int, required=True, help="nof replicas")
    args = parser.parse_args()
    scale_pods(args.deployment_name, args.nof_replicas)


if __name__ == "__main__":
    main()