On the FrontierCode coding benchmark, Fable 5.2 scored 4 points ahead of GPT-6 Astra and 3.5 points ahead of Opus 5.5