Core capabilities
Eight things Claude can do in a plain chat that most people either don't know about or have never tried. No setup required. Available on every plan.
These are the features baked into every conversation on claude.ai, the desktop app, and the mobile app. You do not need to configure anything, install anything, or be on a special plan. They are just there, and most of them are underused.
1. Vision and image understanding
Claude can read images. Attach a photo, a screenshot, a chart, a diagram, a whiteboard, a receipt, or a handwritten note, and Claude will see it and respond to what is in it. Supports JPEG, PNG, GIF, and WebP. Up to 20 images per message on the web interface.
Describing a chart vs attaching it
You are translating a visual into words, then asking Claude to turn it back into analysis. Information is lost in the middle, especially that Q3 note you could not remember. The commentary will be weaker than the chart.
Reads the chart directly, including the Q3 adjustment footnote you had forgotten. Produces commentary that references specific numbers from the image, names the adjustment, and flags the recovery trend. You did not have to transcribe anything.
Try it now: take a photo of the next whiteboard, handwritten note, or chart you encounter. Attach it to Claude and ask a question about it.
2. File uploads
Claude can read documents you attach: PDFs, Word files, Excel spreadsheets, CSVs, plain text, HTML, JSON, EPUB, and more. Up to 20 files per conversation, each up to 30 MB. Instead of pasting text into the chat, attach the file directly and ask Claude to work with it.
Which format for which job
Click between the tabs to see when each format works best.
PDF. Best for contracts, reports, board packs, anything with fixed formatting. Claude reads the text content but not embedded images. For very long PDFs (100+ pages), consider splitting into chapters.
Excel and CSV. Best for data analysis, reformatting, and extraction. CSV is the most reliable format for tabular data. Excel files work but complex formatting, macros, and multiple sheets can sometimes be missed. When in doubt, export to CSV first.
Word. Best for drafts, proposals, and anything you want Claude to rewrite or restructure. Claude reads the text and basic structure (headings, lists). Complex formatting, track changes, and embedded objects may not come through perfectly.
Image. Attach screenshots, photos, charts, diagrams, handwritten notes. Claude reads them via vision. JPEG, PNG, GIF, WebP. Up to 20 images per message. For text-heavy images, attaching the original document format is usually more reliable.
3. File creation
Claude does not just read files. It creates them. Ask Claude to produce an Excel spreadsheet, a PowerPoint presentation, a Word document, or a PDF, and it will generate a real file you can download or save to Google Drive. This is not a screenshot or a description of a file. It is the actual file, with formulas, formatting, and structure.
A real .xlsx file with real formulas. Click download or save to Drive. Open it in Excel and the formulas work.
Try it now: ask Claude to "create a spreadsheet tracking [whatever you track] with columns for [whatever matters], and include a totals row with formulas".
4. Web search
Claude's training data has a cut-off date. Without web search, it does not know what happened yesterday. With web search enabled, it can look things up in real time: current news, recent company filings, live documentation, competitor activity, anything publicly available on the web.
Stale knowledge vs live information
Either admits it does not know ("I do not have access to information after my training cutoff") or, worse, confidently states something from six months ago as if it were current. Both are unhelpful.
Fetches the actual announcement, cites the source, gives you the date and the rate, and can follow up with analysis if you ask. Current information, not stale training data.
How to enable it: look for the web search toggle in your chat settings. On paid plans it is available by default. Once enabled, Claude will search when it judges the question needs current information.
5. Extended thinking
Extended thinking is Claude pausing to reason through a hard problem step by step before answering, rather than jumping straight to a response. On newer models it happens automatically when the task benefits from it. You may see a "thinking" indicator while it works.
The same question, but Claude takes a moment to reason before committing to an answer.
When it matters: multi-step reasoning, complex analysis, anything where the first instinct might be wrong. Claude will use extended thinking automatically, but you can encourage it by saying "think through this step by step" or "walk me through your reasoning first".
6. Research mode
Research mode is extended thinking plus web search, combined into a single deep-synthesis workflow. Instead of one quick answer, Claude reads 5+ sources, synthesises across them, and produces a structured report with citations. Best for competitor analysis, market research, technical investigations, and anything that would normally take you an afternoon of tab-hopping.
Key findings
- Northwind raised a Series C in Q1 2026, led by Meridian Ventures, at a reported valuation of around £120m. [Source: TechCrunch, 14 Feb 2026]
- Their Rotterdam hub opened in March 2023, not January as their website claims. Three press releases confirm March. [Source: Lloyd's List, Port of Rotterdam authority]
- Employee count on LinkedIn: 487, up from 312 a year ago. Growth concentrated in engineering and operations. [Source: LinkedIn company page, accessed today]
Contradictions across sources
Two sources cite 34% YoY revenue growth; a third cites 28%. The discrepancy may be organic vs total revenue. Worth clarifying in your analysis.
How to use it: look for the Research toggle in the model picker. It takes longer than a normal response (30 seconds to a few minutes) but the output is dramatically more thorough.
7. Voice mode
You can talk to Claude instead of typing. Five voice options, 20 languages, continuous listening or push-to-talk. Available on the desktop app and mobile. Useful for brainstorming, dictating first drafts, or working through a problem when your hands are busy.
Voice mode is not a transcription service. Claude hears your speech, understands the intent, and responds with a spoken answer plus a text version in the chat. You can switch between voice and text mid-conversation.
Try it now: on the mobile app or desktop, tap the microphone icon and say "help me think through [something you are working on right now]". It is genuinely useful for the kind of thinking-out-loud that is hard to type.
8. Code execution in the sandbox
Claude can write code and run it, not just describe what code would do. The output comes back into the chat: computed numbers, processed data, generated charts. This runs in a sandboxed environment (no internet access, no persistent storage), so it is safe but limited to computation rather than system access.
Asking Claude to calculate vs asking it to compute
Claude does the maths in its head. Language models are famously shaky on arithmetic. The answer might be right. It might be off by a decimal. You will not know without checking, which defeats the point.
Writes the script, runs it in the sandbox, and returns the result: 18.0% CAGR. The answer is computed deterministically, not guessed. You can see the script if you want to verify the formula.
Rule of thumb: if the answer needs to be precisely right (maths, date calculations, data transforms), ask Claude to write and run a script rather than computing in its head. The script is deterministic. The head-maths is probabilistic.
All eight at a glance
| Capability | What it does | Enable how |
|---|---|---|
| Vision | Reads images, charts, screenshots, handwriting | Attach an image |
| File uploads | Reads PDFs, Word, Excel, CSV, and more | Attach a file |
| File creation | Produces real .xlsx, .pptx, .docx, .pdf files | Ask for one |
| Web search | Fetches current information from the web | Enable in settings |
| Extended thinking | Reasons step by step on hard problems | Automatic, or say "think step by step" |
| Research mode | Deep multi-source synthesis with citations | Toggle in model picker |
| Voice | Speak instead of type, 20 languages | Tap the microphone |
| Code execution | Writes and runs scripts for precise computation | Ask it to "write and run" a script |
These are the foundation. Every other tool in the handbook (Projects, Artifacts, Connectors, Skills, Cowork, Chrome, Extensions) builds on top of these eight. Start with the ones that match your work, and come back to the rest when you need them.
Next: Projects is where most people should go after core capabilities. If you find yourself pasting the same context into every chat, Projects is the fix.