Do you need to use Claude differently starting with Opus 5?

Anthropic launched Claude Opus 5 on July 24 as part of the Claude 5 lineup, alongside Sonnet 5, and the article says the new generation is more specialized and cost-sensitive than before.

The piece argues that benchmarks matter less on their own now; real-world workflow tests and custom domain evaluations are better signals of performance.

Opus 5 is described as doing more built-in verification, shifting work that teams previously handled with prompts or external harnesses.

Anthropic’s updated guidance suggests trimming redundant instructions, adjusting effort levels by task, and choosing models based on task size rather than defaulting to one flagship model.