@fchollet As per the claim abt other domains:
The taste that Opus 5.5 shows towards things like video editing, 3d modeling, raw js artwork, among others, seems like a discrete jump.
Perhaps RLVR but for sure applied in ways we normally thought outside its scope.
@scaling01 imo 6.1 is insane value and we would've given a hand to have that just 2 months ago.
but agree that 5.5 is insane taste/intelligence at very reasonable price and wins the day
@TokenGremlin 6.1-xhigh uses *half* as many tokens as O5.5-high.
so the gap the far smaller than you make it out to be.
token efficiency matters a great bunch