# How Should You Prompt Claude Opus 5.5? Drop "Think Hard" and Ask for the Finished File (November 2026)
> Anthropic now says Claude Opus 5.5 is the default model on Pro, Max, Ultra, and Team plans, and that prompting habits from older models slow it down. Cut "think hard," start at Medium effort, put the whole task in one message, and ask for the finished document instead of an outline.
**Author:** [Alex Lowe](https://theaicareerlab.com/about) — Founder, The AI Career Lab
**Published:** 2026-11-14
**Canonical URL:** https://theaicareerlab.com/blog/how-to-prompt-claude-opus-5-5-2026
**Category:** industry-news
**Tags:** Claude, Claude Opus 5.5, Anthropic, prompting, AI tools, 2026
---> **TL;DR.** On November 12, 2026, Anthropic published a guide to using Claude Opus 5.5 for everyday knowledge work. It confirms Opus 5.5 is now the default model on Pro, Max, Ultra, and Team plans. Anthropic's main point is that habits from older models now get in the way. Stop writing "think hard," stop feeding Claude one instruction at a time, and stop retyping numbers from charts. Instead, leave effort on Medium, describe the whole job in your first message, attach the chart itself, and ask for the finished file.

---

If you learned to prompt Claude in 2024 or 2025, some of your best habits are now making it slower.

That's the message of a new Anthropic guide, ["Using Claude Opus 5.5 as your daily driver for knowledge work"](https://claude.com/resources/articles/using-claude-opus-5-5-as-your-daily-driver-for-knowledge-work), published November 12, 2026. It's written for people who use Claude for documents, analysis and research, not just for developers. [The Decoder](https://the-decoder.com/anthropic-says-outdated-prompting-habits-are-slowing-down-claude-opus-5-5/) covered it on November 14 under the headline "Anthropic says outdated prompting habits are slowing down Claude Opus 5.5."

Here's what changed, checked against Anthropic's own help center and documentation, and what to do about it.

## First: Opus 5.5 is probably already your default

Anthropic's guide says Opus 5.5 is "now the default model in Claude on Pro, Max, Ultra, and Team plans." If you're on one of those plans and haven't changed the model menu, you're already using it. The guide doesn't name a default for Free or Enterprise.

That matters because the advice below is specific to this model. If you're still on an older model, or you deliberately switched to Sonnet, the trade-offs are different. (For how the current models compare, see [Which Claude model should you use?](/blog/which-claude-model-should-you-use))

## Four habits to drop

### 1. "Think hard" and "think carefully"

These phrases date from early reasoning models, when you had to nudge the model to slow down. According to [Anthropic's help center](https://support.claude.com/en/articles/8664678-change-the-model-effort-and-thinking-settings), thinking is always on for Opus 5.5 and can't be turned off. The model decides how much to reason based on the task and the effort setting.

Anthropic says that in its tests, removing "think hard" and "think carefully" made replies start sooner with no clear quality drop. If you want deeper reasoning, raise the **effort** setting (below) instead of adding words to the prompt.

**Where to look:** saved prompts, Claude Project instructions, custom styles and any prompt templates your team shares.

### 2. One instruction at a time

Many people learned to break a task into small steps, sending one message per step and checking each answer. Anthropic says Opus 5.5 handles multi-part work well enough that you should describe the whole job up front. Say what the goal is, what the finished output should look like, and when Claude should stop and ask you something.

### 3. Retyping numbers from charts

With older models it was often safer to type a chart's values into the prompt by hand. Anthropic calls Opus 5.5 its "best Opus model for vision." It says that even at its lowest effort setting, Opus 5.5 read values from dense charts more accurately than Opus 5 did at its highest. Attach the chart, screenshot or diagram itself. In one of Anthropic's examples, the model caught a chart in a deck that didn't match its source data.

### 4. Vague style bans

Instructions like "avoid a generic look" don't work well. Anthropic says they tend to swap one default style for another. Name the specific things you don't want instead (Anthropic's examples are a cream background and pill-shaped buttons). Better still, paste your brand or writing guidelines, or point Claude at your best existing decks. Expect a few rounds before it matches your house style.

## What to do instead

**Leave effort on Medium.** In the Claude apps, click the model name next to the send button and choose **Effort**. Anthropic's help center lists Low, Medium, High, Extra high and Max. Medium is Opus 5.5's default, according to both the guide and [Anthropic's developer documentation](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence). Anthropic says Opus 5.5 at Medium "matched or beat" Opus 5 at High on its internal coding and knowledge-work tests. Its advice is to move up to High when Medium stalls, and to try a higher effort before switching to a bigger model. Higher effort uses more of your plan limits.

**Ask for the finished file.** Instead of "give me an outline for a board update," ask for the board update itself, as a document, spreadsheet or deck. Anthropic says Opus 5.5 often finishes it in one pass. An outline still helps when other people need to agree on the structure first.

**Connect your files.** Anthropic recommends connecting Claude to the tools where your work lives, so it can pull in what it needs instead of relying on what you paste. On Team and Enterprise plans, an owner has to enable a connector before members can use it.

**Use it as a reviewer.** The guide suggests asking Opus 5.5 to check long documents for mistakes and inconsistencies, and to research open questions and write up the findings as a memo or report.

## Should you still use Sonnet to save limits?

Many teams used the cheaper Sonnet by default and saved Opus for hard projects. Anthropic is now pushing back on that, at least for longer work. A more capable model can finish a task in fewer steps, which offsets much of its higher per-token price.

Anthropic's specific claim is that Sonnet 5.5 is still the right call for short, well-scoped tasks with a clear brief. But once you push Sonnet to Extra high or Max effort, its cost per task gets close to Opus 5.5's, and you're better off switching. Its developer docs show a similar pattern on a coding test: Opus 5.5 at Low effort solved about the same share of tasks as Sonnet 5.5 at its default, for about 7% more per solved task.

Two caveats. First, these figures come from Anthropic's own tests, and the guide's customer examples are quotes rather than independent studies. Second, on a subscription plan the cost shows up as how fast you use your limits, not as a bill. If you often hit your limits, try it on your own work before switching everything to Opus.

For the biggest jobs, Anthropic points to Fable 5.2. Its guide notes that Pro, standard Team and seat-based Enterprise seats don't include Fable 5.2 in their limits except during limited-time launch offers, so it runs on usage credits. (Details: [What is Claude Fable 5.2?](/blog/what-is-claude-fable-5-2-2026))

## A note for people who write prompts for the API

If you maintain prompts for an app or an internal tool, the same idea applies. Anthropic's developer docs flag patterns written for older models, such as "verify twice" and "be maximally thorough", as worth removing. In one of Anthropic's tests, removing "verify twice" cut cost per ticket by a third on Opus 5. On the API, Opus 5.5's default effort is `medium`, while most other models default to `high`.

## The bigger picture

Anthropic isn't alone in telling users to prompt *less*. OpenAI's July guide said much the same for ChatGPT: [lead with the result and skip the step-by-step script](/blog/openai-chatgpt-prompting-guide-july-2026). Models now handle more of the planning themselves, so the useful work has moved to describing the outcome clearly, giving the model the right files, and naming the few things that must not go wrong.

The practical step for this week: open your saved Claude prompts and Project instructions and delete every "think hard," "think carefully" and "go step by step." Then run one real task as a single, complete request and compare the result.

## Sources

- Anthropic, "Using Claude Opus 5.5 as your daily driver for knowledge work" (November 12, 2026): https://claude.com/resources/articles/using-claude-opus-5-5-as-your-daily-driver-for-knowledge-work
- Anthropic Help Center, "Change the model effort and thinking settings": https://support.claude.com/en/articles/8664678-change-the-model-effort-and-thinking-settings
- Anthropic developer docs, "Optimizing for cost and intelligence": https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence
- The Decoder, "Anthropic says outdated prompting habits are slowing down Claude Opus 5.5" (November 14, 2026): https://the-decoder.com/anthropic-says-outdated-prompting-habits-are-slowing-down-claude-opus-5-5/
## Frequently asked questions

### Is Claude Opus 5.5 now the default model in Claude?

On most paid plans, yes. In a November 12, 2026 guide, Anthropic says Opus 5.5 is "now the default model in Claude on Pro, Max, Ultra, and Team plans." The guide does not name a default for the Free plan or Enterprise. You can see and change the model in the menu next to the send button.

### Should I still write "think hard" or "think step by step" in Claude prompts?

Not for Opus 5.5. Anthropic says thinking is always on for this model and it decides how much to reason based on the task and the effort setting. In Anthropic's own tests, removing "think hard" and "think carefully" made replies start sooner with no clear drop in quality. If you want more reasoning, raise the effort setting instead.

### What effort level should I use with Claude Opus 5.5?

Start at Medium, which is Opus 5.5's default. Anthropic says Opus 5.5 at Medium matched or beat Opus 5 at High on its internal coding and knowledge-work tests. Move to High when Medium stalls on a task. The Claude apps offer Low, Medium, High, Extra high and Max; you change it by clicking the model name next to the send button and choosing Effort. Higher effort uses more of your plan limits.

### Should I ask Claude for an outline first, then the full document?

Anthropic now recommends the opposite for Opus 5.5: ask for the finished spreadsheet, document or deck directly, and describe the goal, the output and when Claude should stop and ask you a question, all in your first message. Opus 5.5 often finishes the file in one pass. Outlines still make sense when you need to agree on structure with someone else first.

### Should I use Sonnet 5.5 instead of Opus 5.5 to save money or limits?

For short, well-defined tasks, Sonnet 5.5 is faster and cheaper. But Anthropic says that once you push Sonnet to Extra high or Max effort, its cost per task gets close to Opus 5.5's, so at that point you may as well switch to Opus. These are Anthropic's own claims and have not been independently tested.

### Do these tips apply to ChatGPT and Gemini too?

Partly. OpenAI gave similar advice in its July 2026 prompting guide: lead with the result you want instead of scripting every step. The specific points about effort levels, always-on thinking and chart reading are about Claude Opus 5.5 and come from Anthropic's documentation.

---

*Canonical version: https://theaicareerlab.com/blog/how-to-prompt-claude-opus-5-5-2026*
*This document is the Markdown companion served for AI crawlers and answer engines. See the canonical URL for the rendered version with navigation, related content, and interactive elements.*