1.1 — The Claude family: reasoning and long context
Anthropic ships three models in the Claude 4.6 family. Each targets a specific segment of the development workflow, and you'll find them directly in Claude Code via the --model flag or the /model command.
The three tiers
| Model | Target | Context | Typical use case |
|---|---|---|---|
| Opus 4.6 | Deep reasoning | 1M tokens | Architecture, multi-file refactoring, complex decisions |
| Sonnet 4.6 | Perf/cost ratio | 1M tokens | Daily development, code review, test generation |
| Haiku 4.5 | Speed and minimal cost | 200K tokens | Classification, reformatting, lightweight subagent tasks |
Picking the model in Claude Code
# Launch a session with Opus
claude --model opus
# Switch mid-session
/model sonnet
# Set a default model in /config
# Or specify model: haiku in a subagent's config
Key takeaways
• Opus 4.6 is the flagship model for complex reasoning and large-scale code tasks
• Sonnet 4.6 covers 80% of daily tasks at a cost significantly lower than Opus
• The 1M tokens window lets you analyze an entire project — architecture, tests, configuration — in a single pass
• Extended thinking gives Claude the ability to "think" before answering, step by step
• The safety-first approach (Constitutional AI) reduces hallucinations: Claude prefers to flag uncertainty rather than generate plausible but wrong code