What if you could make Claude Code more efficient by giving each AI model a different job?
Instead of using the most expensive model for everything, try a smarter multi-model setup:
Sonnet does the coding. Opus provides expert advice. Haiku explores the codebase.
Three models. Three roles. One coordinated workflow.
Here's how it works:
- Claude Sonnet: Your main coding agent. Handles implementation, edits, testing, and everyday development.
- Claude Opus: Your senior advisor. Reviews complex decisions, helps solve difficult problems, and provides guidance when Sonnet gets stuck.
- Claude Haiku: Your research assistant. Explores repositories, reads files, searches for dependencies, and reports findings.
The goal? Better resource allocation, potentially lower costs, and less unnecessary work for your most powerful models.
Step 1: Install the multi-model configuration
npx claude-code-templates@latest --setting model/sonnet-opus-advisor-haiku-subagents
During installation, choose where the settings should apply: User, Project, or Local.
Step 2: Make your Haiku subagents read-only
Here's an additional improvement I recommend.
When Haiku subagents investigate your codebase, they don't necessarily need permission to modify anything.
Let them research. Let Sonnet make the changes.
You can enforce this behavior using two Claude Code hooks:
Subagents Explore Only: Uses SubagentStart to instruct subagents to investigate and report findings without modifying files.
Subagent Read Only Guard: Uses PreToolUse to block common file-writing tools and state-changing commands when invoked by subagents.
Install both:
npx claude-code-templates@latest --setting hooks/subagents-explore-only --hook security/subagent-read-only-guard
Your main agent keeps its normal permissions, while research subagents work under additional restrictions. Note that the guard does not cover every possible write mechanism, including write-capable MCP tools.
DEV Community
Why does this matter?
Because more powerful AI models aren't always the answer.
Sometimes the biggest improvement comes from giving the right task to the right model.
Let Haiku do the research.
Let Sonnet do the work.
Let Opus handle the hard questions.
Save this configuration and try it in your next Claude Code session.
Claude Code Templates on GitHub:
https://github.com/davila7/claude-code-templates
#ClaudeCode #ClaudeAI #AIAgents #VibeCoding #AgenticCoding
YouTube / ABVX