Triage a long document to what matters
Read a long doc and surface the three things that change your decision, not a proportional summary.
A practical index of 101 Claude Opus 4.7 tasks, grouped into ten parts. Each entry is a one-line brief plus a rough input-token estimate, so you can pick a task, build the prompt, and price it at your real volume. The structure is inspired by 101 Astonishing and Useful Things You Can Do With Claude Opus 4.7 by Bill Wear (2026); the wording, estimates, and cost math are ours.
Read a long doc and surface the three things that change your decision, not a proportional summary.
Get an explanation calibrated to what you already know instead of a generic intro.
Reason through a decision together rather than handing you a verdict.
Generate the strongest case against your current view before you commit.
Step through an unfamiliar spreadsheet cell by cell until it makes sense.
Clean a rambling voice note into a structured outline you can act on.
Catch the specific mistakes you repeat, not generic grammar fixes.
Shape a half-formed idea into a one-page brief with a clear ask.
Practice a difficult conversation before it happens, with pushback.
Surface the real question underneath the one you typed.
Draft the email you have been putting off, in your voice, with one clear ask.
Shift a piece into a different tone or register without losing the point.
Translate and back-translate to catch meaning drift before you send.
Draft a format you have never written before, with structure provided.
Find the thesis hiding in a draft that is not sure it has one.
Generate a sharper subject line than the one you have.
Learn your voice from samples and ghostwrite in it.
Write to a specific reader with their concerns in mind.
Produce a headline with the three sentences beneath it.
Simplify a paragraph that is trying to do too much.
Respond to criticism without matching its heat.
Get a first lesson in a field you know nothing about, paced for beginners.
Work problems at your actual skill level, not a generic one.
Have Claude play a skeptical examiner and test you.
Move jargon to plain language and back to check understanding.
Re-derive a known result from scratch to actually learn it.
Map the landscape of a field you are entering.
Convert notes into spaced-repetition flashcards.
Learn a language by actually using it in conversation.
Build a study guide from a long text.
Get explanations at five different levels of depth.
Synthesize what several articles agree and disagree about.
Pull the argumentative structure out of a long document.
Build a research plan for a question you cannot yet answer.
Generate a competitive analysis framework.
Pressure-test a strategy before committing to it.
Write the postmortem before the project ends, while it is honest.
Draft talking points for a meeting you need to run.
Convert feedback into a prioritized action list.
Outline the document you keep failing to write.
Get a second opinion on your own reasoning.
Debug code you did not write, with ranked hypotheses.
Write a one-off script to solve a one-off problem.
Review code before you commit, with blockers separated from nits.
Refactor without changing observable behavior.
Add tests to code that has none.
Translate code between programming languages.
Orient yourself in an unfamiliar codebase fast.
Pair-program through a design decision.
Trace a bug from symptoms to cause.
Document code you skipped at the time.
Run an ultrareview before a tough merge.
Write a migration plan for a technical change.
Draft a scene you are stuck on.
Generate plot alternatives at a branching point.
Develop a character beyond archetype.
Stress-test a fictional world’s internal consistency.
Art-direct a visual you will make yourself.
Structure a song lyric.
Experiment with poetic form.
Format a screenplay scene with correct conventions.
Design branching for interactive fiction.
Critique your own draft ruthlessly.
Plan the coming week when you are overwhelmed.
Tame an inbox that became its own project.
Decide between similar options like offers, apartments, or cars.
Frame a financial decision honestly, with the trade-offs visible.
Plan meals with real constraints like budget and time.
Build a travel itinerary for actual travelers.
Plan the household project that has been haunting you.
Prepare for a hard conversation with a family member.
Restructure deadlines when you are overcommitted.
Reflect at year’s end without the usual cliches.
Clean a CSV you inherited.
Explore a dataset you have never seen.
Generate SQL from a plain English description.
Find outliers and anomalies in your data.
Do a quick analysis without writing full code.
Explain a statistical result in plain English.
Generate realistic test data.
Get help with a spreadsheet formula.
Design a pivot table (or the equivalent).
Choose the right chart for the data.
Read small text in a photo of a document.
Extract data from a chart you only have as an image.
Identify UI/UX issues from screenshots.
Analyze diagrams, schematics, and technical drawings.
OCR handwritten notes that matter.
Compare two versions of a visual design.
Describe complex visuals for accessibility or context.
Interpret maps, floor plans, and spatial layouts.
Turn a whiteboard photo into structured notes.
Use the xhigh effort level strategically for hard problems.
Run long tasks with task budgets.
Use Claude in Chrome as a browser agent.
Run Claude Code agentic workflows.
Do multi-turn tool use for real workflows.
Connect MCP servers to extend what Claude can do.
Use prompt caching for repeated context.
Use Claude in Excel for spreadsheet agents.
Use the ultrareview command for code.
Each part has a different typical prompt size. The table prices an average prompt for that part across three models, at 1,000 calls. Output tokens are extra and usually dominate on the reasoning-heavy tasks.
| Part | Tasks | Avg input tokens | Claude Sonnet 5.5 | GPT-6.1 Sol | Gemini 3.8 Flash |
|---|---|---|---|---|---|
| Everyday Thinking | 10 | 530 | $1.06 | $1.06 | $0.40 |
| Writing and Communication | 11 | 445 | $0.89 | $0.89 | $0.33 |
| Learning Anything | 10 | 560 | $1.12 | $1.12 | $0.42 |
| Work and Research | 10 | 630 | $1.26 | $1.26 | $0.47 |
| Code and Technical Work | 12 | 950 | $1.90 | $1.90 | $0.71 |
| Creative Work | 10 | 580 | $1.16 | $1.16 | $0.43 |
| Life Admin | 10 | 470 | $0.94 | $0.94 | $0.35 |
| Data and Spreadsheets | 10 | 610 | $1.22 | $1.22 | $0.46 |
| Images, Vision, Multimodal | 9 | 378 | $0.76 | $0.76 | $0.28 |
| Power Moves | 9 | 478 | $0.96 | $0.96 | $0.36 |
Prices are list input rates from the TokenCalc AI catalog and exclude output tokens, caching, batching, retries, and negotiated discounts. Output tokens are usually the expensive half of a Claude conversation; cap them with a length rule in the prompt.
No. The ten-part structure and the task ideas are inspired by 101 Astonishing and Useful Things You Can Do With Claude Opus 4.7 by Bill Wear (2026). Every description, token estimate, and cost figure on this page is original work for TokenCalc AI. The book is a map of what people actually do with Claude; the prompts and the math are ours.
They are rough input-token ranges for a typical prompt for that task, not measurements. Real prompts vary with how much context you paste. Use the Tokenizer Arena for an exact BPE count and the Prompt Builder to assemble the brief before you send it.
Claude Opus 4.7 is the most capable option for the reasoning-heavy and agentic tasks (parts V and X). For everyday writing and admin tasks, Claude Sonnet 5.5 or a cheaper model often does the job at a fraction of the cost. The table below prices the same task across three models so you can see the spread.
Almost all of them. The book notes which entries rely on Opus 4.7 specifics such as the larger image resolution, the xhigh effort level, and improved agentic coding. Those are flagged in the book; the rest transfer to any recent Claude model.
Open the Prompt Builder, pick the closest template, and fill the eight fields. Then trim it in the Trimmer and price the result in the Calculator at your real request volume.
The ten-part structure and the task ideas are inspired by 101 Astonishing and Useful Things You Can Do With Claude Opus 4.7 by Bill Wear (2026). Every description, token estimate, and cost figure here is original. The book is a useful map of what people actually do with Claude; the prompts and the cost math are ours. Claude, Claude Opus, Claude Code, and related names are trademarks of Anthropic PBC.
Token estimates are rough input-token ranges, not measurements. Prices are list input rates from the TokenCalc AI catalog. Output tokens, caching, batching, retries, and negotiated discounts are excluded.