for entertainment purposes only: if you take the ECI scores and fit them w the old METR time horizon data points on models they released estimates for (which they correlate well with) , you can calculate "time horizons" for current models. opus 5.5 sitting at 48hrs!

tbc this is like completely methodologically unjustified and has basically nothing to do with some genuine sense of how long it takes humans to complete tasks. but who cares it's a fun number and just *feels better* than raw ECI