Claude Platform launches Opus advisor tool

Yes, this looks like the commercial shape frontier AI is drifting toward: cheap default execution, expensive reasoning only when needed. It’s a cleaner product story than exposing the biggest model directly, but it also makes routing logic part of the product, not just the backend.

• –Anthropic is packaging Opus-level guidance as a server-side advisor, which lowers cost while preserving access to top-tier reasoning.
• –The executor/advisor split is easier for developers than managing separate orchestrator and worker models, because the handoff happens inside one request.
• –This pattern should improve agent workflows where most turns are routine and only a few need frontier judgment.
• –The tradeoff is opacity: developers may lose a clear sense of when the expensive model is actually being invoked, which matters for latency, cost, and debugging.
• –If Mythos-style models follow the same playbook, users may not “talk to” the frontier model directly so much as rent bursts of it through routing and escalation.