๐ November 14, 2026 โข Anthropic
Anthropic published a knowledge-work guide that positions Claude Opus 5.5 as the default model in Claude on Pro, Max, Ultra and Team plans โ and asks users to drop prompting habits inherited from older model generations. The article appeared on November 12 per web-archive records and was highlighted by The Decoder on November 14. It recommends treating Opus 5.5 as the daily driver for research, analysis and finished documents, decks and spreadsheets.
The headline change is about instructions: per Anthropic, thinking is always on in Opus 5.5 and the model decides how deeply to reason, so lines like "think hard" or "think carefully" can be deleted โ in the company's testing, replies started sooner with no clear drop in quality. Second: start at the Medium effort level, the new default โ per Anthropic's coding and knowledge-work evaluations, Opus 5.5 at Medium matched or beat Opus 5 at High, while higher settings think longer, cost more and do not always produce better work.
The guide lists seven habits to update: connect Claude to work files and systems, describe the finished result in the first message, delete think-hard lines, stay at Medium effort by default, attach charts instead of retyping them, name specific unwanted design styles instead of asking to avoid a generic look, and share your house style. For teams tracking spend, the same page lists pricing of 4 and 20 dollars per million input and output tokens for Opus 5.5, against 8 and 40 for Fable 5.2, 2 and 10 for Sonnet 5.5 and 0.10 and 0.50 for Haiku 5.5.
Internal tests cited in the article: 16 of 18 generated reports passed an automated check on every figure and quote, a fictional merger analysis took 63 minutes against 93 for Opus 5 at half the cost, and an early tester at Factory reports matching Opus 5 on High with 20 to 25 percent fewer output tokens. The platform docs add a hard constraint: adaptive thinking on Opus 5.5 cannot be turned off โ requests that try get an error, and effort is the control.
IMPACT
For anyone running Claude in production workflows the advice is concrete and cheap to test: audit saved prompts and remove reasoning-coaxing lines, pin effort at Medium for daily work and escalate per task, and stop transcribing charts into text. The deeper shift is architectural โ reasoning depth is now a model-side control, so prompt engineering moves from coaxing the model to think toward specifying outcomes and wiring in context. Vendor guidance about its own model is vendor-reported by definition, but the Medium-versus-High finding is easy to replicate on your own tasks before rewiring anything.
Sources: claude.com โ the-decoder.com โ platform.claude.com
Sources:
- https://claude.com/resources/articles/using-claude-opus-5-5-as-your-daily-driver-for-knowledge-work
- https://the-decoder.com/anthropic-says-outdated-prompting-habits-are-slowing-down-claude-opus-5-5/
- https://platform.claude.com/docs/en/build-with-claude/effort
Continue the conversation: https://t.me/AIAutomationArchitecture ยท Bluesky
Top comments (0)