بچه‌های ابزار Cline اومدن دو تا مدل هوش مصنوعی یعنی Ox Alpha و Fable رو سر حل یه باگ واقعی به جون هم انداختن و جالب اینکه هر دوتاشون از پس حل کردنش براومدن

We benchmarked Ox Alpha vs Fable on a real bug from the Cline repo. Both fixed it correctly. But we found that Ox used much fewer thinking tokens. Most reasoning models loop and re-derive the same conclusion over and over before acting (Fable said "I found the root cause" 7 times before editing). Ox stated it once then wrote the fix. Roughly ~3x lower output tokens for the same work. Reasoning models have been trained to increase reliability with re-verification. Ox seems to trust its first conclusion instead, which feels like a fundamentally different post-training philosophy.