The Paradox of Progress: Why Newer AI Models Are Getting Worse at Using Tools In a counterintuitive discovery that challenges our assumptions about AI progress, newer Anthropic Claude models (Opus 4.8 and Sonnet 5) are demonstrating worse performance at correctly calling tools compared to their older siblings. This regression, documented by researcher Armin Ronacher, reveals a fundamental tension in how modern language models are trained and optimized, with significant implications for the futur...

View Entire Post