AI 原生开放教材

外观

Claude Code Everything You Need to Know工程手册·英文原文 ↗仓库 ↗

FAQ — Claude Code & Claude Plans ​

← Back to README

A consolidated FAQ covering models, pricing, tokens, plans, Fast Mode, worktrees, and Pro-plan optimization. For Skills-specific questions, see docs/skills.md → Skills FAQ.

Models ​

Q: What Claude models are available in 2026? ​

As of July 2026: Sonnet 5 (default), Opus 4.8, Fable 5 (Mythos class), and Haiku 4.5, with Opus 4.7/4.6 and Sonnet 4.6 available as legacy models. Full specs, model IDs, and pricing live in one place — Models — so this FAQ doesn't duplicate them.

Key features:

  • Adaptive thinking — current models scale reasoning depth automatically; on Fable 5 it can't be disabled.
  • 1M context window — standard on Sonnet 5, Opus 4.8, and Fable 5 (no beta flag, no long-context surcharge).
  • Fast Mode — up to 2.5× faster output at 2× pricing, on Opus 4.8. Toggle with /fast.

Anthropic ships new models often. Check the official model overview for the current list.

Q: Can I use the 1M context window? ​

Yes. 1M-token context is standard on Sonnet 5, Opus 4.8, and Fable 5 — in Claude Code and via the API — and the full window bills at standard per-token pricing.

Best for: analyzing entire large codebases at once, processing extensive documentation, complex multi-file refactoring.

Reasoning effort ​

Q: How many reasoning effort levels are there, and what's my default? ​

The API knows five (low / medium / high / xhigh / max, default high); Claude Code adds a sixth, ultracode (xhigh reasoning + automatic multi-agent workflow orchestration). max and ultracode are session-only by design.

Current defaults (July 2026): Opus 4.8 → high on all surfaces; Sonnet 5 → high on API and Claude Code. Run /effort to see and change yours. Full breakdown: Effort levels.

Q: Does ultrathink actually raise effort? ​

It adds an in-context "think harder" instruction but does not change the API effort level. If you're already on xhigh or max, the keyword is essentially redundant. For a deliberately deeper turn, set /effort max, prompt, then /effort auto to reset.

Q: Why can't I persist max or ultracode in settings.json? ​

They're session-only by design. "effortLevel" in settings.json accepts low / medium / high / xhigh; set max or ultracode per session with /effort.

Skills & commands ​

Q: What's the difference between custom slash commands and skills? ​

They're officially one system now — .claude/commands/deploy.md and .claude/skills/deploy/SKILL.md both create /deploy. See docs/skills.md → Skills FAQ for security best practices, team sharing, and troubleshooting.

Tokens & limits ​

Q: How many messages do I get on the Pro plan? ​

Anthropic no longer publishes exact message counts. Third-party estimates put Pro at roughly ~45 messages per 5-hour rolling window — and Claude Code's five-hour rate limits were doubled on May 6, 2026 for Pro/Max/Team, with peak-hours limit reductions removed (announcement). Limits reset every 5 hours; Sonnet 5, Opus 4.8, and Haiku 4.5 (plus legacy 4.x models) are included — Fable 5 draws usage credits on subscription plans since June 23, 2026.

Q: How are tokens counted? ​

Everything in your interaction counts: your prompts, Claude's responses, system instructions, tool definitions, and any file content Claude reads.

InputToken cost
English text~1 token per 0.75 words (1,000 tokens ≈ 750 words)
Punctuation / code symbolsEach counts as ~1 token
Images(width_px × height_px) / 750 tokens
File readsEach line counts; CLAUDE.md adds context every session

💡 Use the Claude tokenizer for an exact count, or /usage inside Claude Code to track live usage. Note: Fable 5 and Opus 4.7+ use a tokenizer that produces ~30% more tokens for the same text than older models.

Q: How much code can I work on per 5-hour window? ​

Enough for a focused feature session on a small-to-mid project. Tokens are shared across reading existing code, Claude's analysis, your back-and-forth, and the file edits themselves — so treat any lines-per-window number as an order-of-magnitude guess, not a budget.

Pricing & plans ​

Q: What are the Claude subscription plans? ​

As of July 2026 (verify on claude.com/pricing before committing):

PlanPriceModelsUsage limitClaude Code
Free$0LimitedBaseNot included
Pro$20/mo ($17/mo annual)All~5-hour rolling windowFull
Max 5xfrom $100/moAll5× ProFull
Max 20x$200/moAll20× ProFull
  • Unified subscription — one subscription covers both web (claude.ai) and CLI (Claude Code).
  • All models means the current lineup (Sonnet 5, Opus 4.8, Haiku 4.5) plus legacy 4.x models; Fable 5 draws usage credits on subscription plans since June 23, 2026.

Q: API pricing (pay-as-you-go)? ​

See the Models page for the full table — headline rates: Sonnet 5 $2/$10 (intro through Aug 31, 2026, then $3/$15), Opus 4.8 $5/$25, Fable 5 $10/$50, Haiku 4.5 $1/$5.

Q: Pro vs. Max 5x vs. Max 20x? ​

Same models, more capacity — Max 5x is ~5× Pro's usage, Max 20x is ~20×.

Upgrade if: you frequently hit Pro limits, run multiple parallel sessions or agent teams, or need guaranteed availability during peak hours.

Fast Mode ​

Q: What is Fast Mode and when should I use it? ​

Fast Mode delivers up to 2.5× faster output at 2× pricing, running on Opus 4.8 (the default fast-mode model since Claude Code v2.1.154).

  • Toggle with /fast (CLI-only, v2.1.36+). The ↯ indicator confirms it's on.
  • On subscription plans, fast mode draws from usage credits rather than plan limits.
Standard Opus 4.8Fast Mode
Input (per MTok)$5$10
Output (per MTok)$25$50

⚠️ Fast mode on Opus 4.7 ($30/$150) is deprecated and removed July 24, 2026; Opus 4.6 silently runs at standard speed since June 29, 2026.

Use it for: rapid iteration, live debugging, time-sensitive deployments, demo prep. Skip it for: long background tasks and exploratory work — speed doesn't help work you're not watching.

Sessions, terminals & worktrees ​

Q: What happens when I run multiple Claude Code sessions in different terminals? ​

Usage limits are shared. All Claude Code sessions on your account draw from the same 5-hour budget — running 3 parallel sessions exhausts it ~3× faster.

Conversation context is independent. Each terminal has its own history; Claude in tab 1 doesn't know what happened in tab 2. Each session analyzes the codebase independently.

Q: How do Git worktrees affect Claude Code sessions? ​

Worktrees create a hybrid:

  • Separate working directories — each worktree shows different file contents. Edits in feature-A don't appear in feature-B.
  • Shared .git — git history, branches, and commit logs are visible across all worktrees. Claude can see commits from other features (but not their working changes).
bash
# Terminal 1: feature/auth
git commit -m "Add auth middleware"

# Terminal 2: feature/payment
cat middleware.js        # Won't show auth changes
git log --oneline --all  # WILL show the auth commit

Q: How do I get fully separate Claude Code contexts? ​

Use separate clones instead of worktrees:

bash
~/project-auth/     # Complete separate clone
~/project-payment/  # Complete separate clone

Each clone has its own .git, no shared state, fully isolated views.

Optimization ​

Q: How do I maximize my Pro usage? ​

Session management:

  • /compact to reduce context in long sessions.
  • Close unused sessions to avoid accidental token consumption.
  • Start fresh contexts when switching task types.

Token efficiency:

  • Batch related work — let one session deliver a feature, don't re-explain context across sessions.
  • Use /model strategically: Haiku for simple tasks, Sonnet 5 for everyday, Opus 4.8 when reasoning depth matters.
  • Be specific in prompts to avoid back-and-forth.

Project structure:

  • Keep CLAUDE.md concise — it loads every session.
  • Point Claude at specific files rather than letting it scan the whole repo.

← Back to README · Skills FAQ · Changelog