Claude & Claude Code in Detail

[ikaijua]

Claude & Claude Code in Detail

Claude is Anthropic's AI assistant and coding agent — chat, writing, long documents, and office automation; Claude Code is its terminal coding-agent form, sharing the same models and engine.

Positioning. Anthropic's AI assistant launched in 2023 — the most mainstream alternative to ChatGPT; Claude Code is the official AI coding agent.
Signature strengths. Claude: coding, 1M-token long context, Cowork agentic mode; Claude Code: autonomous coding in terminal / IDE.
Leaderboards. #1 on LMArena text leaderboard (Elo 1505); tied with GPT-6 Astra at the top of the Artificial Analysis Intelligence Index.
Pricing. Free tier available; Pro $20/mo, Team $25/user/mo, Max $100-200/mo; API billed per token.
Note. Mainland China is not a supported region — sign-up, payment, and long-term use carry account risk.

Floors below are grouped as «Claude product intro → Claude Code usage guide».

Table of Contents

Claude

1. What is Claude?
2. Where to Use Claude
3. Leaderboards & Independent Benchmarks
4. Core Capabilities
5. Cowork: Turn Chat into Getting Work Done
6. Typical Use Cases
7. Pricing & Plans
8. FAQ
9. Things to Know: Region & Account Policies
10. References

Claude Code

1. What is Claude Code?
2. Core Philosophy
3. Available Surfaces
4. Core Capabilities
5. Quick Start
6. Common Slash Commands
7. Configuration Files
8. Best Practices
9. Related Resources
10. License

Click a TOC entry to jump to that floor; if in-page clicking does not work, open the link in a new tab.

[ikaijua]

What is Claude?

In one sentence. Claude is Anthropic's AI assistant — it can chat, write, summarize long documents, help you code, and (with Cowork) operate your files and apps to finish multi-step tasks.

What you can do with it.

• Chat & writing: translate, summarize, rewrite, brainstorm
• Coding: write code, fix bugs, or have it explain code you don't understand
• Long documents: feed in a whole report or large codebase and ask about any detail
• Getting work done (Cowork): give it a goal and it works on your files and apps to deliver a finished result

What's special about it.

• Long context: supports 1M tokens — roughly the size of several books at once
• Safety & alignment: designed to be helpful, harmless, and honest, which is why enterprises trust it with sensitive data
• Claude Code: one of the most popular terminal AI coding agents, covered in its own group below

Where to use it. Regular users: web app at claude.ai or the mobile app. Developers: API or the Claude Code terminal tool.

How much it costs. A free tier is available for daily use; Pro is $20/month; heavy coding users can pick Max $100-200/month; API is billed per token (see Pricing & Plans).

Things to know first. Mainland China is not a supported region — sign-up, payment, and long-term use all carry account-risk, see "Things to Know: Region & Account Policies".

[ikaijua]

Where to Use Claude

Surface	Where	Notes
Web app	https://claude.ai	Free sign-up; chat, Artifacts, Cowork
Desktop app	https://claude.ai/download	macOS / Windows; main entry for Cowork
Mobile app	Search "Claude" in app stores	On-the-go chat
API	https://console.anthropic.com	Pay-per-token for developers
Claude Code	see the Claude Code group below	Terminal AI coding agent

Note: region checks are strict on web and mobile — a network location outside supported regions may not be able to access the service at all.

[ikaijua]

Leaderboards & Independent Benchmarks

Claude models sit at the top tier of third-party leaderboards as of Sep-Oct 2026:

• LMArena Text Arena: Claude Fable 5.1 ranks 编程工具可以添加aws提供的codewhisper #1 with Elo 1505 (2026-09-30 snapshot, 410 models)
• Artificial Analysis Intelligence Index: Fable 5.1 is tied with GPT-6 Astra for the lead; Opus 5.5 leads the max-effort tier at 57.6
• Agentic Index: Claude Fable 5.1 leads the Artificial Analysis Agentic Index

Note: these are independent third-party rankings based on real user votes or benchmark runs; each leaderboard weighs different things, and rankings change as models are updated.

[ikaijua]

Core Capabilities

• Coding: writing, reviewing, and debugging code in many languages; with Claude Code it can autonomously make cross-file changes
• Long-context reasoning: 1M-token context handles very long documents and large codebases
• Artifacts: generates previewable documents, charts, and code snippets right in the conversation (web app)
• Multimodal: takes image input — reading charts, interpreting screenshots, turning UI mockups into code
• Safety & alignment: designed to be helpful, harmless, and honest; passes multiple safety benchmarks
• API & ecosystem: official API and SDKs, integrated with Claude Code / Cowork / MCP

[ikaijua]

Cowork: Turn Chat into Getting Work Done

Cowork is Claude's agentic "get things done" mode (desktop, web and mobile in beta) — the same agentic architecture as Claude Code, but no terminal required.

How to use it. In the desktop app, describe a goal (e.g. "organize these 24 expense reports into the quarterly spreadsheet"). Claude plans the steps, works on your files and apps, and comes back with a finished deliverable and a summary.

What it can do.

• Organize files: sort, rename, and batch-process by rules
• Work with data: pull data, build Excel forecasting models, make reports
• Produce deliverables: formatted documents, research summaries, slide drafts
• Scheduled tasks: run on a schedule once set up (e.g. daily briefings)

How it differs from chat. Chat only replies with text; Cowork actually touches your computer — it reads and modifies only the folders you authorize, so give it a clearly scoped workspace.

[ikaijua]

Typical Use Cases

Write / read code.

Rewrite this snippet in Python, add error handling, and explain what each block does.

Long-document Q&A.

I've uploaded this 200-page research report. Summarize the key conclusions and list what changed versus last year.

Office automation (Cowork).

Check these 24 monthly expense reports, fill them into the quarterly sheet, group by department, and generate a summary.

Creative writing.

Rewrite my product intro in three styles — professional, playful, and minimal — 200 words each.

Tip: the more precisely you state the desired outcome and what "done" looks like, the better Claude's output.

[ikaijua]

Pricing & Plans

Plan	Price	Notes
Free	$0	Limited chat quota; good for a light trial
Pro	$20/mo	Daily-use mainstay; higher quota and latest models
Team	$25/user/mo	Pro + team collaboration, admin console (min 5 seats)
Max 5x	$100/mo	For moderate Claude Code users (5x Pro quota)
Max 20x	$200/mo	For heavy Claude Code users (20x Pro quota)
API	per token	For developers, priced per model

API pricing (per 1M tokens, input / output):

• Claude Fable 5.1 / Mythos 5.1: $10 / $50
• Claude Opus 5.5: $4 / $20
• Claude Sonnet 5.5: $2 / $10

Money-saving tip. Opus 5.5 matches the flagship Fable 5.1 on most tasks at roughly 40% lower cost — pick it first for everyday work; Sonnet 5.5 is faster and cheaper for simple tasks.

[ikaijua]

FAQ

Q: Is the free tier enough?
A: Fine for light chat and writing; for frequent use or the latest models, go Pro.

Q: Pro or Max — which one?
A: Chat and writing → Pro. Heavy Claude Code usage → Max 5x or 20x.

Q: What's the difference between Claude and ChatGPT?
A: They're comparable and trade wins. Claude has a stronger reputation for coding, long context, and safety; ChatGPT has a broader ecosystem and more plugins. Try both free tiers before deciding.

Q: Can I use it from mainland China?
A: Anthropic does not support mainland China — both access and payment carry risk-control issues; see "Things to Know: Region & Account Policies".

Q: Is my data safe?
A: Enterprise plans support a Data Processing Agreement (DPA); consumer subscriptions use conversations to improve models by default, which can be disabled in settings.

Q: How are Claude and Claude Code related?
A: Claude Code is the terminal coding-agent form of Claude — a separate product sharing the same underlying models and engine, covered in its own group below.

[ikaijua]

Things to Know: Region & Account Policies

Claude is powerful, but for users in mainland China there are real hurdles to know before paying:

• Mainland China is not officially supported: Anthropic's terms list it as an unsupported region; accounts "created from an unsupported location" can be disabled
• High ban risk: multiple large-scale account suspensions hit in 2026 (June affected a huge number of accounts; September swept up China customers of AWS/Microsoft); virtual cards, resold accounts, and IP/billing-address mismatches easily trigger risk controls, and the appeal success rate is very low
• Identity verification: some cases require a real-person ID check; even after verification the account can still be disabled for region reasons
• Enterprise compliance: since July 2026 some Chinese companies have been told to remove Claude-related tools; confirm compliance before using Claude with company data

Advice for mainland-China users. Try the free tier for a light trial; before committing to a paid plan or using it at work, weigh the account risk and alternatives (domestic models, compliant API resellers) rather than jumping straight to an expensive subscription.

[ikaijua]

References

• Claude official site: https://www.anthropic.com/claude
• Claude model docs: https://platform.claude.com/docs/en/models/overview
• Claude Cowork: https://claude.com/product/cowork
• Anthropic pricing: https://www.anthropic.com/pricing
• Region & account policy: https://support.claude.com (search "supported regions" / "account bans")

[ikaijua]

What is Claude Code?

Claude Code is Anthropic's official AI coding agent. Beyond the terminal CLI, it ships as a VS Code / Cursor / JetBrains IDE plugin, a standalone desktop app, a web interface at claude.ai/code, and — via Remote Control — the Claude iOS / Android mobile app. All surfaces share the same underlying engine, so CLAUDE.md files, MCP servers, Skills, Hooks, and memory stay consistent across devices.

[ikaijua]

Core Philosophy

Agent-first, not just completion. Claude Code understands your entire codebase. It plans across multiple files, edits code, runs commands, checks test results, and iterates on feedback until the task you described is done.

Project-level memory. Through the CLAUDE.md project instruction file and automatic memory, Claude Code remembers project structure, conventions, and common commands so you don't restate context every time.

Controllable autonomy. From strict plan mode to automatic mode to a fully permission-skipping YOLO mode, you choose how autonomous Claude should be based on task risk and trust.

Unified surfaces. Terminal, IDE, desktop app, web, or phone all connect to the same Claude Code engine; sessions and settings carry across devices.

[ikaijua]

Available Surfaces

Surface	Description
Terminal CLI	Full feature set, best for deep development
VS Code / Cursor	Plugin with inline diffs and chat panel
JetBrains	Plugin for IDEA, PyCharm, WebStorm, etc.
Desktop app	Visual diffs, side-by-side sessions, scheduled tasks
Web	claude.ai/code, no local setup required
Mobile app	Claude iOS / Android app connects to local sessions via Remote Control

[ikaijua]

Core Capabilities

Code understanding & editing

• Automatically explores codebase structure, dependencies, and call relationships
• Refactors across files, adds features, and fixes bugs
• Inline diffs and visual review

Terminal & toolchain integration

• Runs shell commands, tests, and build scripts
• Native Git operations and GitHub / GitLab integration
• Extensible to CI/CD, Slack, Chrome, and more

Extensibility

• MCP servers: connect to databases, browsers, external APIs
• Skills: reusable project-level workflows (invoke with /skill)
• Hooks: scripts triggered at lifecycle points such as file writes or command execution
• Subagents: split complex work among specialized agents that run in parallel

Cross-device collaboration

• Remote Control: continue a local Claude Code session from your phone or browser
• Cloud sessions: claude --cloud starts a task you can continue from mobile
• Files, credentials, and the local environment stay on the machine running Claude Code

[ikaijua]

Quick Start

Installation

Official install script (macOS / Linux):

curl -fsSL https://claude.ai/install.sh | bash

Homebrew:

brew install claude-code

npm (alternative):

npm install -g @anthropic-ai/claude-code

On Windows, use WinGet or the PowerShell install script. See the official docs for details.

First run:

claude

Log in with your Anthropic account, or set ANTHROPIC_API_KEY beforehand.

Common commands

claude                          # Start an interactive session
claude /path/to/project         # Start in a specific directory
claude "fix the login bug"      # Ask directly
claude --permission-mode plan   # Plan before making changes
claude --permission-mode auto   # Auto-approve routine actions
claude --cloud                  # Start a cloud session resumable on mobile

Permission modes

Mode	Description
default	Asks before sensitive actions (recommended for daily use)
plan	Explores and plans only, waits for confirmation before changes
auto	Auto-approves low-risk routine actions, still asks for sensitive ones
--dangerously-skip-permissions	Skips all confirmations (use only in trusted repos / CI)

Start with default or plan on unfamiliar projects, then move to auto once you trust the workflow.

[ikaijua]

Common Slash Commands

• /help - Show help
• /clear - Clear the current session
• /commit - Create a Git commit
• /review-pr - Review a Pull Request
• /remote-control - Enable cross-device Remote Control
• /mcp - View MCP server status
• /settings or /config - Open settings

[ikaijua]

Configuration Files

• ~/.claude/settings.json: user-level settings (permissions, model, hooks, environment variables)
• CLAUDE.md: project-level instructions and rules
• .claude/: project-local memory and configuration

[ikaijua]

Best Practices

1. Write clear task goals: describe expected behavior, relevant files, and tests to run.
2. Use CLAUDE.md: document conventions, safety guardrails, and common commands.
3. Start with plan/default: increase automation only after you know the project.
4. Review diffs and run tests: treat Claude's output like a colleague's PR.
5. Use Skills and Hooks: encapsulate repetitive workflows into reusable capabilities.

[ikaijua]

Related Resources

• Official Documentation
• GitHub Repository
• Claude Code Remote Control

[ikaijua]

License

Claude Code is developed and maintained by Anthropic. Use is subject to Anthropic's terms of service.