Anthropic Unveils AI Consultant Tool with Specialized Model Roles

Anthropic has officially launched the Advisor Tool within the Claude Platform API, introducing developers to a novel hybrid intelligence model. In this setup, a more affordable and faster model handles the bulk of the workload, automatically consulting the most powerful model for strategic advice only when complex decisions arise. This design fundamentally rethinks the traditional Agent development approach of "large model plans, small model acts," delivering a truly optimal cost-to-performance solution.

Innovative Mechanism: Small Model Executes, Large Model Advises

This new feature positions Claude Sonnet or Haiku as the "executor," responsible for calling tools, processing tasks, and integrating results. The executor only passes the current context to Claude Opus for advisory input when it encounters a difficult problem it cannot solve independently.

Opus provides high-level solutions, course corrections, or strategic guidance before immediately returning control to the executor. Crucially, Opus does not directly call tools or output final results to users, acting instead as an invisible "strategist." This inverted design ensures the majority of token consumption occurs on the lower-cost model, reserving high-end computing power for critical junctures.

Performance and Cost: A Win-Win in Testing

Official test data reveals significant benefits:

The Sonnet + Opus Advisor combination showed a 2.7 percentage point improvement on the multi-language SWE-bench compared to Sonnet alone, while the overall cost per task decreased by 11.9%.

The Haiku + Opus Advisor pairing delivered even more striking results. Its score in the BrowseComp test jumped from 19.7% to 41.2%, effectively doubling performance. While it still trails Sonnet's solo performance by about 29%, its cost is only 15% of Sonnet's, making it ideal for high-concurrency, batch-processing scenarios with higher tolerance for error.

These results demonstrate that by intelligently routing key decisions, developers can move beyond the difficult choice between "Opus is too expensive for full-time use" and "Sonnet alone isn't powerful enough."

Simple to Use: Enable with a Single Line of Configuration

Developers can activate this feature simply by adding a tool definition of type advisor_20260301 to the tools array in the Messages API. Complex operations like model switching and context passing are managed automatically within a single API request, eliminating the need to manually orchestrate multiple calls or manage extra context.

The max_uses parameter allows flexible control over the maximum number of advisor consultations per request. For billing, token usage for the executor and the advisor are charged separately, ensuring transparency and control. Currently in Beta, using the feature requires adding `anthropic-beta: advisor-tool-2026-03-01` to the request header.

Industry Impact: Opening New Doors for Cost-Effective Agent Development

The launch of the Advisor Tool signifies an important evolution in AI Agent development paradigms. It enables developers to enjoy Sonnet's execution efficiency and near-Sonnet pricing for roughly 95% of tasks, while gaining Opus-level judgment at the 5% of critical decision points. This "executor + advisor" hybrid strategy not only slashes the cost of large-scale deployment but also dramatically enhances the viability of smaller models in complex workflows.

Whether building coding agents, browser automation tools, or high-volume data processing systems, the Advisor Tool offers an efficient, flexible, and economical option. Anthropic once again proves that breakthroughs in model capability stem not just from parameter scale, but from architectural and usage-pattern innovation.

As the Beta phase continues, this feature is poised to quickly become a standard tool for developers building performant, cost-conscious AI Agents. Its subsequent updates and further benchmark performances are certainly worth watching closely.