Advisor model
A cheaper executor model does the work and consults a stronger advisor model at key moments, such as planning or a final review.
An advisor model setup pairs two models. A faster, cheaper executor runs the task end to end, calling tools and iterating. At key moments, such as planning an approach, getting past an obstacle or reviewing work before finishing, it consults a stronger advisor model, which reads the conversation and returns guidance. The advisor doesn’t call tools or talk to the user. It is the lightest form of a multi-agent system: one agent works, a second model only advises.
Anthropic offers this as the advisor tool on the Messages API (in beta) and as a session setting in Claude Managed Agents. Consultations are billed at the advisor model’s rates, and each one sends the advisor the conversation so far, so late consultations in a long session use more input tokens. In Anthropic’s April 9, 2026 post, Sonnet with an Opus advisor scored 2.7 percentage points higher on SWE-bench Multilingual than Sonnet alone while costing 11.9% less per agentic task. The docs add that the gain shrinks as the executor’s own capability approaches the advisor’s, and recommend testing on your own workload.
This inverts orchestrator-workers: there the big model plans and delegates to small ones, here the small model drives and escalates. It also differs from model routing, which picks one model per request up front. Judge it on cost per task and pass rate in your own evals.
Sources
- Claude Docs: Advisor tool platform.claude.com
- Claude Docs: Multiagent orchestration platform.claude.com
- Anthropic: The advisor strategy: Give Sonnet an intelligence boost with Opus claude.com
Related terms
- Multi-agent system (multi-agent orchestration): Several AI agents with separate roles and contexts working on one task, usually coordinated by a lead agent or a fixed workflow.
- Model routing and fallback: Sending each request to the model that fits it, and to another one when the first fails.
- Orchestrator-workers pattern (supervisor): One model plans the task, hands parts of it to worker models or subagents, and combines their results.
- Plan-and-execute: A planner model writes a step-by-step plan, executor models carry out the steps, and the planner revises the plan when a step fails.
- Cost per task (agent unit economics): What one finished agent task really costs: tokens, retries, tool calls and the human time to review it, compared with doing it another way.
- Evals: A repeatable test set that scores an AI feature, so changes can be measured instead of guessed.
- Tokens: The pieces of text models read and write, and the unit you pay for.
