Anthropic's Opus 5 and the Effort Dial: Frontier Intelligence Starts Competing on Price Per Decision

Claude Opus 5, released in late July as the new default on Anthropic's Max tier, is notable less for its benchmark position — approaching the company's own flagship — than for its pricing posture: roughly half the flagship's cost, with an effort dial that lets users explicitly trade capability against spend on a per-task basis. The effort dial deserves attention as a product-design signal. Until recently, model choice was binary and coarse: pick the cheap model or the smart one, per application. A tunable effort setting moves that decision to the granularity of individual tasks — routine work runs at low effort and low cost, hard problems get the full reasoning budget — without switching models, prompts, or providers. For teams running AI orchestration at scale, this converts model economics from a procurement decision into a runtime parameter. The competitive context is a general repricing of intelligence. Capability that cost a premium six months ago now ships at mid-tier prices across every major provider, and the differentiation is migrating toward controllability: how precisely can an operator match spend to task value? That question matters most for agentic systems, where a single workflow may contain dozens of model calls of wildly different difficulty. Flat-rate seats with strong included models — the setup many AI-first teams now run on — push the calculus further: when marginal inference is effectively free, the operating discipline shifts from minimizing calls to allocating attention, deciding which tasks deserve the model's maximum effort. The winners of this phase will be teams that treat model effort as a budgeted resource, not a fixed setting.