Don’t assume you need one specific frontier model to orchestrate agents. So we filmed Codex doing it: printin its plan, spawnin workers, collecting their reports, and signin off with its own words "the chair does not care who sits in it." Whatever you already pay for can conduct.
The smartest model shouldn’t do all the work. In this run, #Fable5 — 1 on the Artificial Analysis Intelligence Index — wrote the plan, then hired each model for what it’s verifiably best at:
#Kimi K3 for UI (1 in blind human preference on the frontend arena)
2of3
#Codex for building (tied at the top of Terminal-Bench),
#MiniMax for audits (~5-10% of frontier cost),
#Grok for smoke tests.
Specialists get you ~95% of the quality at a fraction of the cost — so the premium tokens go where they compound: planning and verification.