Claude Opus 5 Introduces Effort Toggle: Users Choose Between Cost and Capability

One of Claude Opus 5's headline features isn't a benchmark score — it's a dial. The model, released July 24, 2026, lets users toggle how much effort it expends completing a task: low, medium, or high.

How It Works

Opus 5 ships with extended thinking on by default. The effort setting governs how deeply the model deliberates before answering — how many alternatives it explores and how aggressively it verifies its own work. Since deliberation consumes output tokens billed at $25 per million, the toggle functions as a direct price-quality-latency dial, settable per request.

• Low: minimal deliberation — fast and cheap for mechanical tasks like extraction, formatting, and simple edits

• Medium: balanced reasoning for everyday coding and analysis

• High: maximum deliberation for hard debugging, planning, and multi-constraint problems

Why Fortune Called It the Differentiator

Fortune's launch coverage singled out the toggle as the feature that separates Opus 5 from what's already on the market: rather than choosing between different models for different budgets, teams can route every request through one model and modulate spend with a parameter.

The Routing Angle

Early API adopters are already describing escalation patterns: run requests at low or medium effort, and automatically retry at high effort only when validation fails. That concentrates the premium on the hard tail of requests — an architecture that previously required maintaining routers across multiple models.

The Bigger Picture

The effort toggle continues an industry trend toward per-request capability control, but Opus 5 is the first Anthropic model to expose it this directly at the frontier tier. Combined with fast mode (2.5× speed at 2× price), users now have two independent dials — thinking depth and serving speed — on a single model.