we spent weeks on this benchmark @frontrunnerai + too much time deciding whether the bench joke was worth it...
we gave AI agents a year to run GTM:
Fable 5.1: $18M
Opus 5.5: $16M
GPT-6 Astra: $15M
results below
next up: giving agents a Stripe card + $10k budget and seeing how perform in our harness with real money..