Anthropic Launches Advisor Strategy Tool: Combining Opus and Sonnet for Cost-Effective AI Agents
Key Takeaways
- ▸Advisor strategy tool enables cost-effective AI agents by pairing Opus (advisor) with Sonnet or Haiku (executor) in a single API call
- ▸Sonnet+Opus advisor improved performance by 2.7 points on SWE-bench while reducing costs by 11.9% compared to Sonnet alone
- ▸Haiku with Opus advisor doubled performance scores while costing 85% less per task than Sonnet, enabling cost-conscious high-volume deployments
Summary
Anthropic has introduced the advisor strategy tool on its Claude Platform, enabling developers to pair Claude Opus as an intelligent advisor with Claude Sonnet or Haiku as cost-effective executors. This approach allows agents to achieve near Opus-level intelligence while maintaining costs closer to Sonnet levels. The strategy inverts traditional orchestrator-worker patterns by having a smaller, faster model (executor) handle tasks end-to-end, escalating only to Opus when facing difficult decisions. In evaluations, Sonnet with Opus as an advisor demonstrated a 2.7 percentage point improvement on SWE-bench Multilingual while reducing cost per task by 11.9%.
The advisor tool integrates seamlessly into the Claude Messages API as a single-line change, with the model handoff occurring within a single API request without requiring extra round-trips or manual context management. Developers can set maximum usage limits and track advisor token spending separately from executor tokens. Haiku with an Opus advisor showed particularly strong results, scoring 41.2% on BrowseComp—more than double its solo performance of 19.7%—while costing 85% less per task than Sonnet alone, making it suitable for high-volume applications requiring balanced intelligence and cost.
- Advisor tokens are billed separately at advisor rates while executor tokens use lower executor rates, with built-in cost controls via max_uses parameters
Editorial Opinion
The advisor strategy represents a pragmatic approach to the intelligence-versus-cost tradeoff that has become increasingly important in the competitive generative AI market. By enabling developers to optimize their agentic systems through a simple, server-side tool rather than complex orchestration logic, Anthropic addresses a real pain point for production teams. This design choice—letting the smaller model make escalation decisions rather than having a larger model decompose work—is elegant and likely to become influential in how developers structure AI applications.
