I tested Haiku vs. Sonnet across 3 agent tasks – the cheap model won every time(github.com)
3 pointsby aimvik072 hours ago1 comment