Benchmark data published July 2 by AI testing platform BridgeMind shows Claude Fable 5's TypeScript debugging scores collapsed 70% after the model's July 1 relaunch — not because the model got worse, ...