Claude (Anthropic)

Overview

Claude is Anthropic's family of AI models. Anthropic was founded in 2021 by former OpenAI researchers (Dario and Daniela Amodei) with a focus on AI safety. Claude is known for strong coding ability, long context handling, and careful instruction-following.

---

Model Lineup

Anthropic offers three tiers, each optimized for different use cases:

┌───────────────────────────────────────────────────────────────┐
│  Opus 4.6          Sonnet 4.6          Haiku 4.5              │
│  ────────          ──────────          ─────────              │
│  Deepest           Best coding          Fastest               │
│  reasoning         model, balanced      and cheapest          │
│                                                               │
│  Complex analysis  Daily development    High-volume tasks     │
│  Research          Agentic workflows    Quick classification  │
│  Architecture      Production apps      Lightweight agents    │
└───────────────────────────────────────────────────────────────┘

Current Models (as of early 2026)

---

Pricing

Prompt caching reduces input costs by up to 90% for repeated system prompts and context. Cached input pricing: Sonnet $0.30/1M, Haiku $0.08/1M.

---

Key Features

200K-1M Context Window

Claude can process very long inputs. 200K tokens is standard across all models; Opus and Sonnet support up to 1M tokens.

200K tokens ≈ a 500-page book
1M tokens   ≈ an entire codebase (thousands of files)

Use cases: analyze entire repos, process long legal documents, compare multiple papers.

Extended Thinking

Claude can reason internally before answering. It gets a hidden "scratchpad" to work through complex problems step by step.

Without thinking: Immediate response, may miss nuances
With thinking:    Model reasons for 5-30 seconds, then gives a more accurate answer

Toggle in Claude.ai: Option+T (macOS) / Alt+T (Windows/Linux).

In the API:

python
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    temperature=1,
    thinking={"type": "enabled", "budget_tokens": 10000},
    messages=[{"role": "user", "content": "Your complex question"}]
)

Tool Use (Function Calling)

Claude can call external tools/functions you define:

python
tools = [{
    "name": "get_weather",
    "description": "Get current weather for a city",
    "input_schema": {
        "type": "object",
        "properties": {
            "city": {"type": "string"},
            "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]}
        },
        "required": ["city"]
    }
}]

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    tools=tools,
    messages=[{"role": "user", "content": "What's the weather in Paris?"}]
)
# Claude responds with: tool_use block calling get_weather(city="Paris")

Vision (Image Understanding)

Claude can analyze images: screenshots, diagrams, charts, photos, handwritten notes.

python
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    messages=[{
        "role": "user",
        "content": [
            {"type": "image", "source": {"type": "base64", "media_type": "image/png", "data": "..."}},
            {"type": "text", "text": "What's wrong with this UI? List accessibility issues."}
        ]
    }]
)

PDF Reading

Claude can read and analyze PDF documents directly, including complex layouts, tables, and figures.

---

Claude.ai (Web Interface)

The consumer-facing product at claude.ai:

Projects

Projects are one of Claude's standout features:

Project: "Backend API Development"
├── Knowledge: architecture.md, api-spec.yaml, db-schema.sql
├── Custom instructions: "Use TypeScript. Follow our REST conventions."
└── Conversations: (all share the project context)

Every conversation in the project has access to the uploaded knowledge and instructions without re-uploading.

Artifacts

When Claude creates substantial content (code, SVGs, HTML, markdown documents), it renders them in an interactive side panel:

• Code — Syntax highlighted, copy-to-clipboard
• HTML/CSS/JS — Live preview in the artifact panel
• SVG/Diagrams — Visual rendering
• Documents — Formatted markdown

---

Claude Code (CLI)

Claude Code is Anthropic's command-line tool for agentic coding. It reads your files, writes code, and runs commands.

bash
# Install
npm install -g @anthropic-ai/claude-code

# Run in your project directory
claude

# Or pass a task directly
claude "Add input validation to the user registration endpoint"

What It Can Do

• Read and understand your entire codebase
• Write, edit, and create files
• Run terminal commands (build, test, lint)
• Search across files with grep/glob
• Manage git operations
• Multi-file refactoring

Key Concepts

Plan mode:   Claude explains what it will do before doing it
             Toggle with Option+P / Shift+Tab

Thinking:    Extended reasoning before acting
             Toggle with Option+T

Auto-accept: Let Claude run commands without confirmation
             (Use carefully)

When to Use Claude Code vs Claude.ai

---

Claude API

Messages API (Basic Call)

python
import anthropic

client = anthropic.Anthropic()  # Uses ANTHROPIC_API_KEY env var

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    system="You are a helpful coding assistant.",
    messages=[
        {"role": "user", "content": "Write a Python function to validate email addresses."}
    ]
)

print(response.content[0].text)

Streaming

python
with client.messages.stream(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain microservices."}]
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

Structured Output via Tool Use

Force a specific JSON schema by defining a tool and requiring its use:

python
response = client.messages.create(
    model="claude-sonnet-4-20250514",
    tools=[{
        "name": "extract_info",
        "description": "Extract structured information",
        "input_schema": {
            "type": "object",
            "properties": {
                "name": {"type": "string"},
                "age": {"type": "integer"},
                "skills": {"type": "array", "items": {"type": "string"}}
            },
            "required": ["name", "age", "skills"]
        }
    }],
    tool_choice={"type": "tool", "name": "extract_info"},
    messages=[{"role": "user", "content": "John is 28 and knows Python, Go, and Rust."}]
)
# Returns: {"name": "John", "age": 28, "skills": ["Python", "Go", "Rust"]}

---

Claude vs GPT-4o vs Gemini

When to Choose Claude

• You need long context (200K-1M tokens)
• Coding tasks (especially agentic multi-file work)
• Instruction following and careful constraint adherence
• Safety-sensitive applications
• Complex reasoning with extended thinking

When to Choose Alternatives

• GPT-4o: Image generation, web browsing, ecosystem (plugins, GPTs)
• Gemini: Google integration, multimodal (video), very long context at lower cost
• Open source: Privacy, offline use, cost at scale (see 13 - Open Source Models)

---

Quick Tips

1. Use extended thinking for complex problems. The accuracy improvement is significant.
2. Use projects in Claude.ai to avoid re-uploading context every conversation.
3. Prefer Sonnet for 90% of tasks. Use Opus only for the hardest reasoning problems.
4. Use Haiku for high-volume classification, extraction, or routing.
5. Cache system prompts via the API to reduce costs by up to 90%.
6. Use XML tags in prompts. Claude handles structured input with XML particularly well.

---

Resources

• 🔗 Anthropic Documentation
• 🔗 Claude API Reference
• 🔗 Claude Code
• 🔗 Anthropic Cookbook (examples)
• 🔗 Claude Pricing
• 🔗 Anthropic Research

---

Previous: 09 - System Prompts & Instructions | Next: 11 - ChatGPT & OpenAI