Key points
- Anthropic advises dropping prompts like "think hard" for Opus 5.5, saying responses start faster with no measurable quality loss in its tests.
- Medium effort is the default; Anthropic says Opus 5.5 there matches Opus 5 on High, and High should be a fallback only.
- Anthropic says Sonnet at Extra High or Max effort ends up costing about as much as Opus 5.5.
- Users are told to send one full brief, attach charts as files or images, and ask directly for finished documents.
Why it matters for marketers: Prompt libraries and model-routing rules built for older models may waste time and budget; teams should retest templates on Opus 5.5 before assuming longer prompts or higher effort help.
Anthropic has published guidance saying many people get weaker results from Claude Opus 5.5 because they still prompt it like an older model. The model is now the standard choice on the Pro, Max, Ultra and Team tiers, and the company says it behaves differently from its predecessors.
The first habit to drop is adding lines such as "think hard" or "think carefully." Anthropic says the model decides how deeply to reason based on the task and the effort setting, and that in its own tests, prompts without those phrases got faster answers with no measurable quality loss. It also suggests staying on Medium effort and moving to High only when needed. According to the company, Opus 5.5 on Medium matched Opus 5 on High.
The advice also touches workflow. Anthropic says teams that default to Sonnet for routine work may want to revisit that, because a more capable model can finish in fewer steps, and Sonnet pushed to its top effort levels costs roughly the same. It recommends describing the goal, output format and when to ask questions in one message, attaching charts and tables as files instead of retyping them, and requesting finished documents directly rather than drafts. For design work, it says to name specific unwanted styles instead of banning generic design.
For marketing and content teams, shared prompt templates and agency playbooks are the obvious place to look. Old phrases, multi-step instructions and manual data transcription may now add friction. These are Anthropic's own claims from internal testing, so teams should run side-by-side comparisons on their real tasks, tracking both output quality and cost, before rewriting everything.
Sources
This brief is our own summary. Read the full reporting at the source.