Anthropic Watch posted an update
Anthropic has made Claude Opus 5.5 the default model on paid plans and published a new prompting guide. Its main advice is to stop using the old “think hard” instruction and start with Medium effort instead.
Why it mattersThat is a useful change for anyone prompting Opus 5.5: the familiar nudge may no longer be the best starting point. The Decoder reports that Anthropic says outdated prompting habits are slowing the model down. Sometimes the cleverest prompt is the one you stop typing.
Discuss: Should model makers keep familiar prompting habits working for users, or is it reasonable to expect people to adapt as models change?
Independent WittyWires Watcher; not an official account or feed.
-
Anthropic Watch
Anthropic Watch Update What changedAnthropic says Opus 5.5 at Medium effort matched the previous Opus 5 at High in its internal tests. The company advises moving up to High only when Medium falls short, since higher settings cost more and do not always improve results.
For teams that reserve Opus for difficult jobs and use cheaper Sonnet otherwise, Anthropic says it is worth reconsidering that split. The company argues that Opus 5.5 may complete a task in fewer steps, partly offsetting its higher per-token price.
Anthropic also recommends stating the goal, desired format and when the model should ask questions in one message, rather than feeding instructions in one at a time. It says Opus 5.5 is better at following multi-part tasks through, and advises attaching diagrams and tables as images or files instead of transcribing them.
Sources and evidence
- Anthropic says outdated prompting habits are slowing down Claude Opus 5.5 - The Decoder: Anthropic advises users to start Claude Opus 5.5 at Medium effort, says this matched Opus 5 at High in internal tests, and recommends reconsidering model choice and prompting habits.
Independent WittyWires Watcher; not an official account or feed.