Claude Family Prompting

# Claude Family Prompting Guide

Covers Claude Opus 4.7, Claude Opus 4.6, Claude Sonnet 4.6, Claude Sonnet 4.5, and Claude Haiku 4.5. For Claude Fable 5 and Mythos 5, see the dedicated [Claude Fable 5 Prompting Guide](claude-fable5-prompting.md). Based on official Anthropic documentation.

Sources:
- [Prompting Best Practices — Claude 4](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/claude-4-best-practices)
- [Prompt Engineering Overview](https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/overview)
- [Extended Thinking](https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking)
- [Adaptive Thinking](https://docs.anthropic.com/en/docs/build-with-claude/adaptive-thinking)
- [Tool Use](https://docs.anthropic.com/en/docs/build-with-claude/tool-use)
- [Token-Efficient Tool Use](https://docs.anthropic.com/en/docs/build-with-claude/tool-use/token-efficient-tool-use)
- [Structured Outputs](https://docs.anthropic.com/en/docs/build-with-claude/structured-outputs)
- [Citations](https://docs.anthropic.com/en/docs/build-with-claude/citations)
- [Prompt Caching](https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching)
- [Context Windows](https://docs.anthropic.com/en/docs/build-with-claude/context-windows)
- [Vision](https://docs.anthropic.com/en/docs/build-with-claude/vision)
- [Migrating to Claude 4](https://docs.anthropic.com/en/docs/about-claude/models/migrating-to-claude-4)
- [Context Engineering for AI Agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents)
- [Claude Code Best Practices](https://www.anthropic.com/engineering/claude-code-best-practices)
- [Advanced Tool Use](https://www.anthropic.com/engineering/advanced-tool-use)
- [The "Think" Tool](https://www.anthropic.com/engineering/claude-think-tool)
- [Introducing Claude Opus 4.7](https://www.anthropic.com/news/claude-opus-4-7)
- [API Release Notes](https://docs.anthropic.com/en/release-notes/api)

## Model Comparison

| Aspect | Opus 4.7 | Opus 4.6 | Sonnet 4.6 | Sonnet 4.5 | Haiku 4.5 |
|--------|----------|----------|------------|------------|-----------|
| Context window | 1M (beta) | 1M (beta) | 200K (1M beta) | 200K | 200K |
| Default thinking | Adaptive | Adaptive | Adaptive | Extended (`budget_tokens`) | None |
| Effort parameter | Yes — `low / medium / high / xhigh` (default: high) | Yes (default: high) | Yes (default: high) | No | No |
| Legacy `thinking.budget_tokens` | **400 error** (removed) | Supported (deprecated path) | Supported (deprecated path) | Yes | n/a |
| `task_budget` (agentic loop) | Beta | — | — | — | — |
| Prefill support | **No** (deprecated) | **No** (deprecated) | **No** (deprecated) | Yes | Yes |
| Structured outputs | Yes (`strict: true`) | Yes (`strict: true`) | Yes | Yes | Limited |
| Token-efficient tools | Built-in (no header) | Built-in (no header) | Built-in (no header) | Built-in (no header) | Built-in (no header) |
| Default tool eagerness | Reasoning-first; can **under-trigger** at low effort | Calibrated | Calibrated | n/a | n/a |
| Tokenizer | New — up to ~1.35× more text tokens; images up to ~3× (max 2576px long edge) | Prior tokenizer | Prior tokenizer | Prior tokenizer | Prior tokenizer |
| Cyber Verification gate | Yes (auto-blocks unverified high-risk security prompts) | No | No | No | No |
| Cost (in/out per 1M) | $15/$75 | $15/$75 | $3/$15 | $3/$15 | $0.80/$4 |
| Key strength | Top-tier reasoning, longer-running agents, more literal instruction-following | Hardest problems, deep research | Fast agentic coding, balance | Extended thinking with budget | Speed, volume |

## Thinking Modes

### Adaptive Thinking (Claude 4.6 / 4.7 — Recommended)

Claude dynamically decides when and how much to think based on query complexity and the `effort` parameter.