Anthropic just benchmarked "Fable 5 orchestrates, cheap models execute": 96% of the performance at 46% of the cost. You can run this pattern in Claude Code today

Yesterday's ClaudeDevs thread published first-party numbers for two multi-model patterns ( docs ): Fable 5 as orchestrator, Sonnet 5 as workers: 96% of all-Fable performance at 46% of the cost (BrowseComp: 86.8% vs 90.8% accuracy, $18.53 vs $40.56 per problem) Sonnet 5 as executor, consulting a Fable 5 advisor: ~92% at ~63% (SWE-bench Pro) The orchestrator split won on both axes. Their post covers the API and Managed Agents side. If you're a Claude Code subscriber, the same split works natively today, and t