Claude Sonnet 5.5 scores 70.6% on Terminal-Bench 4.0, up from Sonnet 5's 10.3%, while running 30% faster. Developers disabling thinking must switch to the
between_tools setting.💬 A security researcher notes that authorized bug bounty work is immediately blocked by the model's new cyber safeguards.
→ discussion
797 points · 533 comments · #daily · @HNDigestEn