In a rare Friday release, Opus 5 became today's headline. Although its performance on most official benchmarks has technically surpassed Fable, the official messaging still describes it as "very close". This mostly reflects the difficulty of evals (nowadays an AIE track drop) itself—it fails to capture that "big model intuition (big model smell)" that Anthropic clearly knows Fable still possesses, but which cannot be measured.
Fortunately, independent evaluations also confirm Opus's leading performance:
Moreover, beyond pricing, the efficiency gains narrative is also important... even though it barely matches GPT 5.6 Sol:
