Mountain ViewFable 5 scored 84.5% on the new benchmark.

Google added cost data too — Fable 5 and GPT 5.5 both burn over $130 in tokens per run.

Developers can now submit their own tasks for the official test.

Sources: Ars Technica