Claude Opus 5 shows how quickly frontier-level intelligence is moving toward lower prices—and why efficiency may become the industry’s most important benchmark.
For much of the generative AI boom, companies competed by promising the smartest model in the world. Anthropic’s latest release suggests the next stage of that race will be measured differently: how much useful intelligence customers receive for every dollar spent.
Released on July 24, Claude Opus 5 is positioned as an everyday model for developers, knowledge workers and businesses. Anthropic says it approaches the performance of its more advanced Claude Fable 5 model across many coding and professional tasks, but at roughly half the price.
Opus 5 costs $5 per million input tokens and $25 per million output tokens—the same base price as its predecessor, Opus 4.8. It is available across Anthropic’s paid products and has become the default model for Claude Max subscribers.
The price of a token, however, tells only part of the story. A cheaper model that repeatedly fails, uses unnecessary tools or requires extensive human correction can ultimately cost more. The metric that increasingly matters to businesses is cost per successfully completed task.
Anthropic says Opus 5 improves that equation by completing more work with fewer steps and by allowing users to select how much computational effort it should devote to a request. Lower settings prioritize speed and conserve tokens; higher settings let the model spend more time on complex problems. Developers can also change models during a task, reserving the most expensive intelligence for moments when it is actually needed.
Independent testing will be necessary to determine how consistently those claims hold across real business workloads. Benchmarks published by model developers do not always reflect performance inside a company’s specific software, data and approval processes.
Still, the direction is clear. As Axios notes, Opus 5 became Anthropic’s fourth Claude 5 release in less than two months.
The industry is moving from occasional blockbuster launches toward rapid cycles of improvement. The winning AI may not always be the model with the highest benchmark score. It may be the one that delivers nearly the same result faster, more reliably and at a price businesses can afford to use every day.
