Anthropic Claude Opus 5

Anthropic just made the most important number in AI smaller.

On July 24, it released Claude Opus 5 — a model that matches, and on some benchmarks beats, its flagship Fable 5, at roughly half the effective cost. Pricing held flat at $5 per million input tokens and $25 per million output, but the real shift is a new low/medium/high "effort" toggle that lets you dial capability against spend on a per-task basis.

That toggle matters more than the benchmark scores. For most of the last two years, the enterprise AI question was "which model is smartest?" Opus 5 reframes it as "how much intelligence does this specific task actually need?" — and hands that decision to the buyer instead of the vendor.

This is what a maturing market looks like. When frontier performance becomes a knob you can turn down to save money, the competitive edge stops being raw capability and starts being cost discipline: knowing which workflows deserve maximum reasoning and which are fine at low effort. The leaders won't be the teams with the best model. They'll be the ones who stop overpaying for it.

Anthropic also calls Opus 5 its safest release yet, with less deceptive behavior — a quiet but real signal as more companies hand agents autonomous, multi-step work.

If your team could toggle AI cost against capability per task, who in your org actually owns that decision today?