Sonnet 4.6: The Workhorse Model for Production Development

Where Sonnet 4.6 Sits in the Lineup

Sonnet 4.6 is the mid-tier reasoning model in the Claude 4 generation. It costs roughly half of Opus 4.7 and significantly less than Opus 4.6, while delivering better performance on most development tasks than Haiku 4.5. For teams using Claude Code in practice, Sonnet 4.6 is the default model for a majority of sessions — not because it is the best at everything, but because it is the best at most things at a reasonable price.

The context window is 200,000 tokens. That is enough to load an entire large codebase or multiple specification documents at once. Most developers never come close to saturating that window in a single task, which makes Sonnet 4.6 practical for long development sessions without the cost penalties of Opus.

When Sonnet 4.6 Is the Right Choice

Use Sonnet 4.6 as your default for:

• Feature implementation in known codebases
• Writing tests for existing code
• Bug fixes where you understand the problem domain
• Refactoring with clear direction
• Code review and explanation of unfamiliar code
• Documentation generation and updates
• Migration scripts with well-defined transformations

These are tasks where the work is bounded, the goals are clear, and the cost of being wrong is low enough that you review the output before it ships anywhere important. Sonnet 4.6 handles them efficiently without the reasoning overhead that makes Opus slower and more expensive on the same tasks.

When to Switch Away from Sonnet 4.6

Sonnet 4.6 has a specific failure mode: it produces confident, plausible-sounding wrong answers on tasks that require deeper reasoning. You spot these by noticing you are correcting the same thing repeatedly, or by seeing suggestions that sound right but do not actually solve the problem when you apply them.

The signal to switch to Opus mid-session:

• You have corrected Sonnet 4.6 three or more times on the same task
• The task involves tracing through multiple layers of abstraction
• The problem requires holding state across many files or a long conversation
• You are working on authentication, authorization, or security-sensitive code
• The task involves complex concurrency, distributed systems, or data consistency

Switch with /model opus-4.7 or /model opus-4.6 mid-session — you do not need to restart or lose context.

Configuring Sonnet 4.6 as Your Default

Set Sonnet 4.6 as the project default in claude_settings.json:

{
  "defaultModel": "sonnet-4-6",
  "maxTokens": 8192,
  "temperature": 0.7
}

Run a new session with Sonnet 4.6 explicitly:

claude --model sonnet-4-6

Override at session start, then switch as needed:

claude --model sonnet-4-6
# Later in session:
/model opus-4.7
# Switch back:
/model sonnet-4-6

Cost Comparison

The actual cost difference matters for budgeting. Sonnet 4.6 is approximately:

• 2x cheaper than Opus 4.7 for equivalent token volumes
• 1.5x cheaper than Opus 4.6
• 3-4x more expensive than Haiku 4.5 for equivalent tasks

For a typical development session of 50,000 input tokens and 10,000 output tokens, Sonnet 4.6 costs roughly $0.40–$0.60 USD. Opus 4.7 on the same session would be $0.80–$1.20 USD. The gap compounds over a full day of sessions.

The Honest Assessment

Sonnet 4.6 is not the best model in the lineup — Opus 4.7 beats it on reasoning depth. But it is the most practical model for most development work. The developers who use it well are not using it because they cannot afford Opus — they are using it because they know when Sonnet 4.6 is sufficient and when to escalate. That judgment is what separates effective Claude Code usage from expensive overuse of the flagship model.