“Anthropic has deprecated the intelligence of its Claude model, with the Opus version now generating 70% fewer tokens for the exact same question.”