Anthropic Makes Opus 5.5 the Default on Every Paid Plan and Tells Users to Stop Writing 'Think Hard'

Anthropic has quietly completed a product change with real billing consequences: Claude Opus 5.5 is now the default model on Pro, Max, Ultra and Team plans. Alongside the switch, the company published a knowledge-work guide, reported by The Decoder on November 14, that tells users to unlearn prompting habits inherited from older model generations, starting with the most reflexive one, adding "think hard" or "think carefully" to a request.

The reasoning is architectural. Opus 5.5 always reasons on its own and scales its depth to the task and the selected effort level, so prompt text trying to force more thinking is, by Anthropic's account, dead weight. In the company's tests, responses without those phrases started faster with no measurable drop in quality. The phrases date back roughly a year and a half, to the early days of reasoning models, when users could sometimes coax extra performance out of a model or steer the router behind the chat box.

The second habit to break is defaulting to maximum effort. Medium is the default for Opus 5.5, and Anthropic says that in internal tests Opus 5.5 at Medium matched the performance of its predecessor Opus 5 at High, while higher settings cost more without always producing better results. The setting lives in the model menu next to the send button, and the company advises stepping up to High only when Medium falls short. Those comparisons are Anthropic's own, published without scores or test details, so teams should treat them as vendor-reported rather than independently verified.

The guide also pushes back on the Sonnet-first cost strategy many teams adopted. Anthropic argues that a more capable model finishes a task in fewer steps, partly offsetting its higher per-token price, and that anyone regularly running Sonnet at Extra High or Max effort ends up paying about as much as with Opus 5.5. The guide's price list puts Opus 5.5 at $4 and $20 per million input and output tokens, with Fable 5.2 at $8 and $40, Sonnet 5.5 at $2 and $10, and Haiku 5.5 at $0.10 and $0.50.

The rest of the advice reframes Claude as something to delegate to rather than guide turn by turn. Describe the final result in a single message, including the output format and when the model should check back with questions. Attach diagrams and tables as images or files instead of transcribing them; Anthropic claims Opus 5.5 at its lowest effort level reads values from dense charts more accurately than Opus 5 did at its highest. Ask for the finished document or spreadsheet directly rather than a draft first. For design work, name the unwanted elements, a cream-colored background or pill-shaped buttons, instead of vaguely banning "generic design." Share your house style. And connect Claude to your files and systems so it can pull context itself.

The internal numbers attached to that pitch are specific: of 18 generated reports, 16 passed automated checks on every number and quotation; a fictional M&A analysis took 63 minutes on Opus 5.5 against 93 minutes on Opus 5, at about half the cost; and an early tester at Factory reported results comparable to Opus 5 at High with 20 to 25 percent fewer output tokens. All of it is company-reported, with no independent benchmark published alongside.

The larger signal is how the interaction model has shifted. Effort control has moved from the prompt into a settings menu, the router-steering era is officially over, and the marginal value of prompt engineering is migrating toward task design: what you ask for, what context you attach, and which systems the model can reach. Users who keep writing for the old model are not just being inefficient, they are paying for effort the model no longer needs.