精選 Extended Thinking Opus Sonnet Advanced

Claude Code 的 Extended Thinking：釋放深度推理能力

掌握 Claude Code 的 Extended Thinking。學習何時啟用思考模式、如何使用 effort 等級，以及如何運用深度推理來處理複雜的開發任務。

2026年1月10日 • 17 分鐘閱讀 • 作者：ClaudeWorld

Extended Thinking（也稱為「Thinking Mode」）讓 Claude 有更多時間在回應之前深入思考複雜問題。這就像給開發者時間在寫程式碼之前先思考架構一樣。

本指南涵蓋何時使用 Extended Thinking、如何設定它，以及最大化效果的模式。

什麼是 Extended Thinking？

Extended Thinking 讓 Claude 能夠：

  • 更長時間地深思熟慮再回應
  • 考慮多種方法來解決問題
  • 系統性地推理邊界情況
  • 徹底規劃多步驟實作
┌─────────────────────────────────────────────────────────────┐
│                    Normal vs Extended Thinking              │
│                                                             │
│  Normal:                                                    │
│  ┌────────┐      ┌────────┐                                │
│  │ Prompt │  ──► │Response│                                │
│  └────────┘      └────────┘                                │
│                                                             │
│  Extended Thinking:                                         │
│  ┌────────┐   ┌─────────────────┐   ┌────────┐            │
│  │ Prompt │ ► │ Internal Reason │ ► │Response│            │
│  └────────┘   │ (not visible)   │   └────────┘            │
│               └─────────────────┘                          │
│                     ▲                                       │
│                     │                                       │
│              Longer, deeper                                 │
│              deliberation                                   │
└─────────────────────────────────────────────────────────────┘

模型支援

模型Extended Thinking預設備註
Opus 4.5是開啟（v2.0.67+）最高品質推理
Sonnet 4.5是關閉為複雜任務啟用
Haiku 4.5是關閉首個支援思考的 Haiku

關鍵洞察： Opus 4.5 從 v2.0.67 起預設啟用思考功能，因為它能顯著提升品質。

啟用 Extended Thinking

透過 Config

在 /config 中，啟用思考模式：

Thinking Mode: ON

透過 Settings

在 .claude/settings.json 中：

{
  "model": "opus",
  "thinking": true
}

每個工作階段

在工作階段中切換：

"Enable thinking mode for this complex architecture discussion."

快速切換模型

在輸入提示時按 Option+P（macOS）或 Alt+P（Windows/Linux）快速切換模型。

Effort 等級（僅 Opus 4.5）

Opus 4.5 支援 effort 等級 來控制思考深度：

Effort思考時間使用情境
low最少快速任務、批次操作
medium平衡預設，大多數任務
high最多複雜分析、架構設計

設定 Effort

在提示中：

"Analyze this authentication system with high effort.
I need thorough security analysis."

在 Task tool 中：

Task({
  subagent_type: "general-purpose",
  model: "opus",
  prompt: `
    Effort: high

    Analyze the payment integration architecture.
    Consider all edge cases and failure modes.
  `
})

何時使用 Extended Thinking

架構決策

沒有思考：

"How should we structure our microservices?"

Claude: [Quick response, may miss considerations]

有思考：

"Enable high-effort thinking.

How should we structure our microservices?
Consider: scalability, team structure, data ownership,
deployment complexity, and failure isolation."

Claude: [Thorough analysis of trade-offs]

安全審查

"Using Opus with high effort, review the authentication
implementation for vulnerabilities.

Consider: OWASP Top 10, JWT best practices, session
management, rate limiting, and common attack vectors."

複雜重構

"Think through how to refactor the legacy payment system
to the new architecture. Consider:

- Backward compatibility
- Database migrations
- API versioning
- Rollback strategy
- Testing approach"

除錯複雜問題

"This bug only happens in production with high concurrency.
Use extended thinking to analyze potential causes.

Symptoms: [describe symptoms]
Environment: [describe environment]
Recent changes: [describe changes]"

Extended Thinking 模式

模式 1：架構審查

"I need to design a new feature. Use extended thinking.

Feature: Real-time notifications
Requirements:
- Support 100K concurrent users
- < 100ms delivery latency
- Support mobile and web
- Handle offline users

Analyze options and recommend architecture."

模式 2：權衡分析

"Think deeply about this trade-off.

Option A: Monolith with modular structure
Option B: Microservices from start

Context:
- 3-person team
- MVP stage
- Unknown scale requirements
- Limited DevOps experience

Analyze both options with pros, cons, and recommendation."

模式 3：安全審計

"Perform a deep security audit with high effort.

Focus areas:
- Authentication and authorization
- Input validation
- Data encryption
- API security
- Dependency vulnerabilities

For each issue found, provide:
- Severity (Critical/High/Medium/Low)
- Description
- Remediation steps
- Code examples"

模式 4：效能調查

"The API is slow under load. Think through potential causes.

Observations:
- Response time increases from 50ms to 2s at 500 RPS
- CPU stays at 30%, memory at 60%
- Database connections spike
- No errors in logs

Analyze potential bottlenecks and suggest investigation steps."

對 Context 的影響

重要： Extended Thinking 會影響 prompt caching 效率。

┌─────────────────────────────────────────────────────────────┐
│              Context Usage Comparison                        │
│                                                             │
│  Normal Mode:                                                │
│  ├── Prompt Cache: Highly effective                        │
│  ├── Context usage: Predictable                            │
│  └── Token efficiency: High                                 │
│                                                             │
│  Extended Thinking:                                          │
│  ├── Prompt Cache: Less effective                          │
│  ├── Context usage: Variable (thinking uses tokens)        │
│  └── Token efficiency: Lower (but higher quality)          │
│                                                             │
└─────────────────────────────────────────────────────────────┘

建議： 選擇性地為高價值決策使用 extended thinking，而非日常任務。

模型選擇指南

使用 Haiku 4.5（無思考）

Best for:
- File searches (Explore agent)
- Pattern matching (Grep)
- Simple code generation
- Bulk operations

Example:
"Find all uses of deprecated API in the codebase"

使用 Sonnet 4.5（思考可選）

Best for (without thinking):
- Standard implementations
- Code refactoring
- Test writing
- Documentation

Best for (with thinking):
- Complex implementations
- Multi-file refactoring
- Architecture-sensitive changes

Example:
"Implement the user authentication system following
our security patterns. Enable thinking for the token
refresh mechanism design."

使用 Opus 4.5（預設思考）

Best for:
- Architecture decisions
- Security audits
- Complex debugging
- Critical code paths

Example:
"Review the payment processing implementation.
Use high effort for the transaction handling logic."

Agents 中的 Thinking Mode

Explore Agent（Haiku）

Explore agents 使用 Haiku 而不啟用思考以提升速度：

Task({
  subagent_type: "Explore",
  model: "haiku",
  prompt: "Find authentication patterns (thoroughness: medium)"
})

不需要思考 — 純粹的搜尋和發現。

Implementation Agent（Sonnet + Thinking）

對於複雜實作，啟用思考：

Task({
  subagent_type: "general-purpose",
  model: "sonnet",
  prompt: `
    Enable thinking mode.

    Implement OAuth 2.0 with PKCE flow.
    Consider:
    - State management
    - Token storage
    - Refresh handling
    - Error recovery
  `
})

Review Agent（Opus + High Effort）

對於關鍵審查，使用 Opus 搭配 high effort：

Task({
  subagent_type: "general-purpose",
  model: "opus",
  prompt: `
    Effort: high

    Security review the payment integration.
    Check for:
    - PCI compliance issues
    - Data exposure risks
    - Authentication bypasses
    - Rate limiting gaps
  `
})

實際範例

範例 1：功能設計

"I need to add collaborative editing to our document app.

Enable extended thinking with high effort.

Constraints:
- Real-time sync (< 100ms)
- Handle offline mode
- Conflict resolution
- Undo/redo support

Analyze:
1. Technology options (CRDT, OT, etc.)
2. Architecture approach
3. Trade-offs of each option
4. Recommended path forward"

範例 2：遷移規劃

"We're migrating from MongoDB to PostgreSQL.

Use Opus with high effort to create a migration plan.

Current state:
- 50 collections
- 10M documents
- 5 services depending on data
- Zero downtime requirement

Deliverables:
1. Migration strategy
2. Schema design
3. Service update plan
4. Rollback strategy
5. Testing approach
6. Timeline estimate"

範例 3：除錯工作階段

"Production issue: Intermittent 500 errors on /api/checkout.

Think through this carefully.

Data points:
- Happens ~1% of requests
- No pattern in time of day
- No correlation with request size
- Started after last deploy
- Logs show: 'Connection reset by peer'

Analyze potential causes and investigation steps."

範例 4：程式碼審查

"Review this PR with extended thinking.

PR adds: Payment processing with Stripe

Focus areas:
- Error handling completeness
- Idempotency of operations
- State machine correctness
- Webhook handling security
- Test coverage adequacy

Provide detailed feedback with specific line references."

Thinking Mode 設定

專案層級

在 .claude/settings.json 中：

{
  "model": "sonnet",
  "thinking": false,
  "thinkingConfig": {
    "enableForSecurity": true,
    "enableForArchitecture": true,
    "defaultEffort": "medium"
  }
}

每個任務的觸發條件

設定思考自動啟用：

{
  "thinkingTriggers": {
    "keywords": ["security", "architecture", "design"],
    "paths": ["auth/", "payment/", "admin/"],
    "operations": ["create-file", "refactor"]
  }
}

成本效益分析

何時 Extended Thinking 值得

情境時間成本品質提升ROI
架構決策+30秒避免數月的返工非常高
安全審查+45秒捕捉漏洞非常高
複雜重構+20秒更好的設計高
Bug 調查+15秒更快找到根因高
日常編碼+10秒邊際改善低

何時跳過

  • 檔案搜尋
  • 模式匹配
  • 簡單的程式碼變更
  • 文件更新
  • 常規測試

最佳實踐

1. 明確說明要使用思考

"Enable extended thinking for this security analysis."

而不是：

"Analyze this code."

2. 提供豐富的上下文

Extended thinking 在有上下文時效果更好：

"Using high-effort thinking, design the caching layer.

Context:
- Current: No caching, 200ms average response
- Goal: <50ms for 80% of requests
- Constraints: Redis available, 5GB memory budget
- Traffic: 10K RPM, 80% reads

Consider cache invalidation, TTL strategies, and
cache stampede prevention."

3. 要求結構化輸出

"Think through this architecture decision.

Output format:
1. Options considered (3-5)
2. Evaluation criteria
3. Analysis matrix
4. Recommendation with rationale
5. Risk assessment"

4. 與 Agents 結合

使用思考來進行協調，使用速度來執行：

"Phase 1 (Opus, high effort): Design the migration strategy
Phase 2 (Sonnet, no thinking): Implement each migration step
Phase 3 (Haiku, no thinking): Run tests in parallel"

監控思考使用量

檢查 Context 使用量

狀態列（v2.0.65+）顯示 context 使用量：

Context: 45% used | Thinking: enabled | Model: opus

根據工作階段調整

對於長時間工作階段，考慮在日常任務中停用思考以保留 context：

"Disable thinking for these routine file updates."

開始使用

今天：

  1. 為一個架構決策啟用思考
  2. 比較有無思考的輸出品質
  3. 記錄時間差異

本週：

  1. 為你的專案設定模型選擇
  2. 識別受益於思考的任務
  3. 建立能運用深度推理的提示

本月：

  1. 為團隊建立思考模式指南
  2. 設定自動啟用觸發條件
  3. 衡量對決策品質的影響

Extended Thinking 將 Claude 從快速回應者轉變為深思熟慮的夥伴。策略性地用於重要決策，你會看到架構決策品質顯著提升。

來源：Claude Code Documentation、Claude Code CHANGELOG