Anthropic is addressing the high cost of Fable 5 by repositioning it as a manager that delegates tasks to the smaller Sonnet 5 model. The company recommends using Fable 5 primarily as a planner, with Sonnet 5 handling execution. This approach, known as the 'Advisor' pattern, allows Sonnet 5 to perform most tasks and only consult Fable 5 when needed. According to Anthropic, this combination achieves about 92 percent of Fable 5's solo performance at 63 percent of the cost on the SWE-bench Pro benchmark. Fable 5 is called roughly once per task in this setup. The company also introduced an 'Orchestrator' pattern where Fable 5 plans and distributes tasks to multiple Sonnet 5 workers, achieving 96 percent of Fable 5's performance at 46 percent of the cost on the BrowseComp benchmark. Both patterns operate through Claude Managed Agents, with each sub-agent using its own cache to avoid duplicate context costs. The official documentation provides more details on these strategies. Anthropic is likely sharing these tips due to growing price pressure from Chinese open-source models and the new GPT-5.6 Sol, which is much cheaper per token and reportedly more token-efficient.
Source: thedecoder
