Is Q$-$learning an on$-$policy method or an off$-$policy method?

Is Q$-$learning an on$-$policy method or an off$-$policy method?

On$-$policy method. Q$-$learning learns about the policy that is being used to generate the actions.

On$-$policy method. Q$-$learning learns about the greedy policy and then uses the new greedy policy for next action selection.

Off$-$policy method. Q$-$learning learns about the greedy policy and then uses the new greedy policy for next action selection.

Off$-$policy method. Q$-$learning learns about the greedy policy, which gradually becomes the optimal policy. And this policy is independent of the one that is actually being followed by the agent when it chooses actions.