Opus 5 beats Fable 5 on benchmarks — but Anthropic is relaxing the guardrails
anthropic launched opus 5, a model that outperforms fable 5 on benchmarks while offering cheaper pricing and fewer data retention restrictions, challenging the notion that capability requires heavy control.
anthropic launched opus 5, a new model that outperforms the heavyweight fable 5 on benchmarks and improves at self-verification and computer vision tasks. the model launches two months after version 4.8, completing the company's 5-series family with the exception of haiku. opus 5 features lighter safety restrictions than fable 5, with an 85% lower classifier engagement rate and no 30-day data retention policy, making it more attractive for privacy-conscious developers. additionally, a new automatic fallbacks beta feature routes blocked requests to smaller models to ensure functional api responses.