Claude Opus 5 Prices Frontier Performance at Half the Cost of Fable 5

Anthropic launched Claude Opus 5 on July 24, 2026, priced at $5 per million input tokens and $25 per million output tokens. That's roughly half what Claude Fable 5 costs, while Anthropic is positioning Opus 5 as targeting comparable intelligence rather than a step down in capability to match the lower price.

The benchmark numbers back that positioning up to a point: Opus 5 topped both Frontier-Bench and GDPval-AA, two of the more closely watched capability benchmarks this year.

The more practically useful addition is a new user-toggleable "effort" setting, low, medium, or high, that lets a developer trade off cost against capability on a per-request basis rather than picking a fixed model tier upfront and living with it. That's a different shape of pricing decision than the usual "which model tier do I default to" question. It pushes the cost-versus-capability tradeoff down to the request level, where a developer can spend more compute on a hard query and less on a routine one within the same model, instead of provisioning for the hardest case across every request.

For teams already running Fable 5 in production, the math is simple enough to check directly: same intelligence tier, half the per-token cost, plus a lever to spend less on the requests that don't need it.