Anthropic's Claude Advisor Strategy pairs Opus with cheaper models for smarter agents

Anthropic has introduced a novel architectural approach called the 'Advisor Strategy' to its Claude Platform, fundamentally changing how developers can build cost-effective yet powerful AI agents. The strategy allows a developer to configure an agent where the expensive, top-tier Claude 3 Opus model acts as a strategic advisor, while a more affordable model like Claude 3.5 Sonnet or Haiku handles the execution of tasks. Crucially, this all happens within a single API request: when the executor hits a complex problem, it can pause and consult Opus, which returns a plan before the executor resumes. This creates a hybrid system that delivers intelligence approaching Opus levels without the full Opus price tag.

In internal evaluations, the performance gains are tangible. A Sonnet executor paired with an Opus advisor scored 2.7 percentage points higher on the challenging SWE-bench Multilingual coding benchmark compared to Sonnet working alone. More impressively, this smarter system was 11.9% cheaper to run per task than using Sonnet solo, as the costly Opus inference is only triggered for critical decision points. This represents a major shift in agent design, moving from a static, single-model architecture to a dynamic, multi-model team. By separating planning from execution, Anthropic provides a practical path for scaling sophisticated agentic workflows in production, where both performance and cost are paramount constraints. The feature is available in beta immediately on the Claude Platform.