The doubling time for the METR time horizons has moved off the consistent 7 months from 2019-2024 to a new exponential doubling every 4 months at around the end of 2024.

(source: https://theaidigest.org/time-horizons)
Some causes, such as reasoning models, suggest this will revert to baseline. Others, such as AI accelerating research engineering strongly, suggest the trend will continue.
This is confounded by other possible phase transitions in research, but would still be an interesting part of the future to peer into.
@plex could you specify
whether you mean the time series for 50% time horizon or for 80% time horizon (or either/both)
what calculation you will do to determine whether this resolves YES or NO
whether you would resolve before December 31 under any circumstances?