About 5 months out. This is what polling error looks like historically. We are getting into the period where polls are kinda useful. But you'll notice the major election models are still heavy on fundamentals at this point as there is a lot of room for the polls to improve.
In recent decades, there is evidence this shift in polling error is smaller than previous decades, but still large enough to matter in such a close election.
Also my regular reminder that polling error is not correlated between election cycles. Even for the same pollster! So don't go assuming polling error and direction of the error will be similar to 2022 or 2020. We won't know until the election.