· Updated
advisor_20260301 advisor tool on the Claude Platform, enabling a new "advisor strategy" for AI agents. Instead of a large model orchestrating smaller ones, a cost-effective executor like Claude Sonnet or Claude Haiku drives the task and invokes Claude Opus as an advisor only when it hits a difficult decision.This approach inverts traditional agent patterns to solve the trade-off between reasoning capability and inference (running a model to generate an output) costs. By limiting Claude Opus to short plans while the cheaper model handles execution, developers achieve performance gains on benchmarks like SWE-bench without the expense of a full frontier-model run.
Implement this by adding the advisor tool to a single /v1/messages API request, which handles the model handoff server-side. The feature is in beta and requires the advisor-tool-2026-03-01 header. Usage is billed at each model's rates; set max_uses to cap advisor calls per task.
We're bringing the advisor strategy to the Claude Platform. Pair Opus as an advisor with Sonnet or Haiku as an executor, and get near Opus-level intelligence in your agents at a fraction of the cost. https://t.co/fRkegyMs5t
More like this
ClaudeMay 12
AnthropicYesterday
OpenRouterMay 13