Claude Advisor Adds Opus Boost
Anthropic updated Claude Advisor to include Opus as a ‘silent strategist’ that can be invoked in a single API call, and Anthropic claims this improves task performance by about 11.9%. The change was announced on social channels as a way to simplify building agents that want a behind‑the‑scenes planning layer (x.com).
Anthropic has added an advisor tool to the Claude application programming interface that lets a cheaper model call Claude Opus 4.6 for behind-the-scenes strategy in one request. (platform.claude.com) Anthropic’s documentation says the setup pairs a faster “executor” model with a higher-intelligence “advisor” model that reads the full conversation, suggests a plan or course correction, and then hands the task back. The company’s April 10 blog post says developers can pair Opus with Claude Sonnet 4.6 or Claude Haiku 4.5. (platform.claude.com) (claude.com) Anthropic said the pattern can be invoked through the Messages application programming interface with a simple configuration instead of a separate orchestration layer. In the company’s example, the advisor typically returns 400 to 700 text tokens, with 1,400 to 1,800 total tokens including internal reasoning. (platform.claude.com) The underlying problem is cost: frontier models are better at planning, but they are slower and more expensive to run on every step. Anthropic lists Claude Opus 4.6 at $5 per million input tokens and $25 per million output tokens on its pricing page for the model. (anthropic.com) Anthropic’s pitch is to keep the cheaper model doing most of the work and bring in Opus only when the task gets hard. In its blog post, the company said this produced “near Opus-level intelligence” for agents at a lower cost. (claude.com) The company attached benchmark numbers to that claim. Anthropic said Sonnet 4.6 scored 72.1% on SWE-bench Multilingual by itself, and 74.8% when paired with Opus as an advisor, while cost per task came in 11.9% below running pure Opus. (claude.com) Anthropic announced the change on April 9, 2026, and third-party coverage the same day described it as a public release on the Claude Platform for application programming interface users. A social post circulating the launch framed it as a way to build agents with a planning layer that stays out of the user-facing exchange. (testingcatalog.com) (x.com) The release also fits Anthropic’s recent push to sell Opus as its premium model for coding, long-running agent work, and complex enterprise tasks. Anthropic introduced Claude Opus 4.6 on February 5, 2026, and described it as the strongest model it has shipped. (anthropic.com 1) (anthropic.com 2) For developers, the practical change is that “use Opus when needed” is now a built-in product feature instead of a custom pattern. Anthropic is betting that one-call access to a silent strategist will make higher-end planning easier to add without paying Opus rates for every token. (platform.claude.com) (claude.com)
